Can Language Models Play the Wikipedia Game?
Language models increasingly support or automate information retrieval from document collections too large to index. This task is difficult when it requires specialist linguistic or scientific knowledge. We used the Wikipedia Game to test language models on an analogous information retrieval task. In the game, players must reach a target article using only article links. We compared sentence embedding models and LLMs across five Wikipedia languages and five academic-literature datasets. We construct very large local graphs of Wikipedia pages and of paper citations using OpenAlex and provide code to build and navigate them. This allows the study of language models' navigation capabilities without relying on third-party APIs. We compared the models with human and TF-IDF benchmarks. The Wikipedia Game format enables robust comparisons across techniques, languages, and domains. No model significantly outperformed humans, although the best models (GPT-o3 mini, GPT-4o, QwQ:32B, and DeepSeek-R1:70B) achieved comparable success. Performance and rankings varied across languages and topics, with implications for practitioners utilising these techniques.
Updated: 2026-10-09 17:59:56
标题: 语言模型能玩维基百科游戏吗？
摘要: 语言模型越来越支持或自动化从太大以至无法索引的文档集合中检索信息。当需要专业的语言学或科学知识时，这项任务就变得困难。我们利用维基百科游戏来测试语言模型在类似的信息检索任务上的表现。在游戏中，玩家必须仅使用文章链接来到达目标文章。我们比较了五种维基百科语言和五种学术文献数据集上的句子嵌入模型和LLMs。我们使用OpenAlex构建了非常大的维基百科页面和论文引用的本地图，并提供代码来构建和导航它们。这使得研究语言模型的导航能力无需依赖第三方API。我们将模型与人类和TF-IDF基准进行了比较。维基百科游戏的格式使得可以在技术、语言和领域之间进行强大的比较。没有模型明显胜过人类，尽管最好的模型（GPT-o3 mini、GPT-4o、QwQ:32B和DeepSeek-R1:70B）取得了可比较的成功。性能和排名在不同语言和主题之间变化，这对于利用这些技术的从业者具有重要意义。
更新时间: 2026-10-09 17:59:56
领域: cs.AI,cs.IR
Measuring Harness-Induced Belief Divergence in Multi-Step LLM Agents
Software agents are becoming increasingly important in modern coding workflows, where they solve complex tasks through multi-step interaction with code repositories. A central component of these systems is the execution harness, which governs how the underlying LLM observes, acts on, and verifies its environment. We show that the harness can in turn shape the agent's evolving beliefs about the environment, even when the task, environment, and base LLM remain fixed. To reveal this effect, we introduce a belief-rollout diagnostic that traces how an agent's anticipated world state and future decisions evolve under alternative harnesses. We further separate immediate divergence introduced by the interface from divergence that develops over longer horizons. Across controlled coding tasks include HIBench-Code, SWE-bench Verified and Terminal-Bench, we find substantial cross-harness belief divergence that is often invisible from terminal success alone and can propagate into different downstream actions. Building on this observation, we introduce \textbf{Belief-Invariant World-Modeling (BIWM)}, which exposes evidence hidden or compressed by harness execution and aligns the resulting belief trajectories across different harnesses. Our findings reveal that harness design can systematically shape how software agents understand and act on their environments, providing practical guidance for building and evaluating more reliable coding agents.
Updated: 2026-10-09 17:59:01
标题: 测量多步LLM代理中由马具引起的信念分歧
摘要: 软件代理在现代编码工作流程中变得越来越重要，它们通过与代码存储库的多步交互解决复杂任务。这些系统的核心组件是执行框架，它规定了基础LLM如何观察、对环境进行操作和验证。我们展示了框架可以反过来塑造代理关于环境的演变信念，即使任务、环境和基础LLM保持不变。为了揭示这种影响，我们引入了一种信念演化诊断，追踪代理的预期世界状态和未来决策在不同框架下的演变。我们进一步区分了界面引入的即时分歧与随着时间推移而发展的分歧。在包括HIBench-Code、SWE-bench Verified和Terminal-Bench在内的受控编码任务中，我们发现跨框架信念分歧是相当显著的，通常仅通过终端成功是看不到的，并且可能传播到不同的下游行动。基于这一观察，我们引入了Belief-Invariant World-Modeling（BIWM），它暴露了框架执行隐藏或压缩的证据，并使不同框架下的结果信念轨迹对齐。我们的研究结果表明，框架设计可以系统地塑造软件代理对环境的理解和行动方式，为构建和评估更可靠的编码代理提供实用指导。
更新时间: 2026-10-09 17:59:01
领域: cs.AI
Evolving Hybrid Quantum-Classical Architectures for Image Classification
Hybrid quantum classical neural networks integrate parameterized quantum circuits (PQCs) with established deep learning architectures, but their performance depends strongly on the choice of quantum circuit architecture, a choice that remains largely manual. Most existing approaches rely on hand-designed or fixed circuit ansätze, requiring circuit structure, gate composition, and qubit connectivity to be specified in advance with no guarantee that they suit the task. This limitation is especially acute in image classification, where quantum circuits must transform features extracted by classical networks while remaining compact enough for practical training, requirements that generic, task-agnostic ansätze are unlikely to satisfy simultaneously. We extend EXAQC, an evolutionary framework for automated quantum circuit discovery, to image classification. EXAQC evolves PQCs as intermediate processing modules while retaining classical feature-extraction and prediction layers. On MNIST, Fashion-MNIST, and CIFAR-10, EXAQC achieves 98.42%, 90.62%, and 85.47% accuracy, respectively, while using comparable gate counts to other quantum architecture-search methods. Against classical networks, evolved hybrid models maintain comparable accuracy with substantially fewer trainable parameters, reaching 85.68% on CIFAR-10 with over 25$\times$ fewer parameters than a 10-layer CNN. Encoding choice also matters: rotation-based encodings (RX, RY, U3) outperform amplitude encoding by 22-25 points on CIFAR-10. These results demonstrate that automated circuit discovery yields compact quantum modules that can replace larger classical components in vision architectures while retaining competitive accuracy.
Updated: 2026-10-09 17:57:17
标题: 不断演变的混合量子-经典架构用于图像分类
摘要: 混合量子经典神经网络将参数化量子电路（PQCs）与已建立的深度学习架构结合，但它们的性能强烈依赖于选择量子电路架构，这一选择仍然大部分是手动的。大多数现有方法依赖于手工设计或固定的电路方案，需要事先指定电路结构、门组成和量子比特连接性，但无法保证其适用于任务。在图像分类中，这种限制尤为严重，量子电路必须将经典网络提取的特征转换，同时保持足够紧凑以进行实际训练，通用的、任务不可知的方案很难同时满足这些要求。我们将自动化量子电路发现的进化框架EXAQC扩展到图像分类。EXAQC将PQCs演化为中间处理模块，同时保留经典特征提取和预测层。在MNIST、Fashion-MNIST和CIFAR-10上，EXAQC分别实现了98.42％、90.62％和85.47％的准确率，同时使用与其他量子架构搜索方法相当的门数。与经典网络相比，演化的混合模型保持着相当的准确性，但可训练参数却大大减少，达到了CIFAR-10上85.68％的准确率，比10层CNN少了25倍的参数。编码选择也很重要：基于旋转的编码（RX、RY、U3）在CIFAR-10上比振幅编码表现出更好的效果，提高了22-25个点。这些结果表明，自动化电路发现产生了紧凑的量子模块，可以替代视觉架构中更大的经典组件，同时保持竞争性的准确性。
更新时间: 2026-10-09 17:57:17
领域: cs.NE,cs.AI,cs.CV,cs.LG,quant-ph
Hide and Seek: HIDE Concepts to SEEK explanations for black box models
Interpreting machine learning model predictions is a central challenge, particularly for black-box vision models. Concept-based attribution methods interpret models using human-understandable concepts. Existing concept-based methods can only interpret classification and regression models. They require access to model weights or large concept-annotated datasets, and most methods cannot provide image-level attribution. We propose Hide and Seek (HnS), a new method for concept-level attribution that does not require access to model weights or labeled concept datasets. HnS can interpret a wide range of vision models. Given a concept and image set, HnS retrieves concept-aligned images using a vision-language model, localizes the concept within each image, masks the localized region, and measures attribution from the change in the black-box model's output between the original and masked images. This enables faithful attribution by directly measuring the model's behavior and supporting local attributions. We validated HnS on CelebA and COCO-Stuff. Compared to existing methods, HnS showed better agreement with ground-truth attributions. We demonstrated the applicability of HnS on diverse tasks, including image captioning and satellite image segmentation, where no existing concept attribution method can be used. We provided our anonymous code in the supplementary material.
Updated: 2026-10-09 17:56:59
标题: 躲猫猫：HIDE概念，为黑盒模型寻找解释
摘要: 机器学习模型预测的解释是一个中心挑战，特别是对于黑盒视觉模型。基于概念的归因方法使用人类可理解的概念来解释模型。现有的基于概念的方法只能解释分类和回归模型。它们需要访问模型权重或大型概念注释数据集，大多数方法无法提供图像级别的归因。我们提出了“Hide and Seek”（HnS），这是一种新的概念级别归因方法，不需要访问模型权重或标记的概念数据集。HnS可以解释各种视觉模型。给定一个概念和图像集，HnS使用视觉-语言模型检索与概念对齐的图像，定位每个图像中的概念，遮罩定位的区域，并通过原始和遮罩图像之间黑盒模型输出的变化来衡量归因。这通过直接测量模型的行为并支持局部归因实现了忠实的归因。我们在CelebA和COCO-Stuff上验证了HnS。与现有方法相比，HnS与地面真实归因显示出更好的一致性。我们展示了HnS在各种任务上的适用性，包括图像字幕和卫星图像分割，在这些任务中没有现有的概念归因方法可用。我们在附录材料中提供了我们的匿名代码。
更新时间: 2026-10-09 17:56:59
领域: cs.CV,cs.AI,cs.LG
DEAF: A Benchmark for Diagnostic Evaluation of Acoustic Faithfulness in Audio Language Models
Recent Audio Multimodal Large Language Models (Audio MLLMs) demonstrate impressive performance on speech benchmarks, yet it remains unclear whether these models genuinely process acoustic signals or rely on text-based semantic inference. To systematically study this question, we introduce DEAF (Diagnostic Evaluation of Acoustic Faithfulness), a benchmark of over 2,700 conflict stimuli spanning three acoustic dimensions: emotional prosody, background sounds, and speaker identity. Then, we design a controlled multi-level evaluation framework that progressively increases textual influence, ranging from semantic conflicts in the content to misleading prompts and their combination, allowing us to disentangle content-driven bias from prompt-induced sycophancy. We further introduce diagnostic metrics to quantify model reliance on textual cues over acoustic signals. Our evaluation of seven Audio MLLMs reveals a consistent pattern of text dominance: models are sensitive to acoustic variations, yet predictions are predominantly driven by textual inputs, revealing a gap between high performance on standard speech benchmarks and genuine acoustic understanding.
Updated: 2026-10-09 17:56:35
标题: DEAF：音频语言模型中声音保真度的诊断评估基准
摘要: 最近的音频多模大型语言模型（Audio MLLMs）在语音基准上展现出令人印象深刻的性能，但目前尚不清楚这些模型是否真正处理声学信号，还是依赖于基于文本的语义推理。为了系统地研究这个问题，我们引入了DEAF（声学忠实性诊断评估），这是一个涵盖情感语调、背景声音和说话者身份三个声学维度的超过2700个冲突刺激的基准。然后，我们设计了一个受控的多层次评估框架，逐渐增加文本影响力，从内容中的语义冲突到误导性提示及其组合，使我们能够分离基于内容驱动的偏见和提示诱导的谄媚。我们进一步引入诊断指标来量化模型对文本线索而非声学信号的依赖程度。我们对七个音频MLLMs的评估显示出一致的文本主导模式：模型对声学变化敏感，但预测主要由文本输入驱动，揭示了在标准语音基准上高性能和真正的声学理解之间存在差距。
更新时间: 2026-10-09 17:56:35
领域: cs.AI,cs.SD,eess.AS
Constructive Bounds for the Topological Expressivity of Neural Networks
The expressivity of a neural network $Φ$ describes the class of functions that a network can express. The term topological expressivity refers to a family of measures particularly suited to understand binary classification and is often formulated in terms of the Betti numbers of a decision boundary $β(Φ^{-1}(0))$. Despite its popularity, this measure is not as useful for regression and other problems that do not involve a decision boundary. Instead, we propose to study topological expressivity in terms of the Betti numbers of the output of a neural network $β(Φ(X))$. We formally prove that a ReLU network with architecture $(d,\sum\limits_{k=1}^{d } β_k(3k+2),\sum\limits_{k=1}^{d}β_k(k+1))$ can express spaces homotopy equivalent to wedge sums of spheres with arbitrary Betti numbers. This gives upper bounds for any arbitrary connected ($β_0=1$) sequence of Betti numbers, but by constructing spaces homotopy equivalent to products of $N$ spheres of dimension $d_1,...,d_N$ we achieve much better bounds $(\sum\limits_{n=1} d_n, 3\sum\limits_{n=1}\limits^N d_n + 2N, \sum\limits_{n=1}\limits^N d_n + N) $ for Betti numbers given by $\prod_n (1+x^{d_n})$. Finally we conjecture a much stronger tight bound which gives the architectural bound $(d,d+1,d+1)$ for a space homeomorphic to the sphere.
Updated: 2026-10-09 17:55:33
标题: 神经网络拓扑表达能力的建设性界限
摘要: 神经网络$Φ$的表现力描述了网络可以表达的函数类。拓扑表现力一词指的是一组特别适合理解二元分类的度量标准，通常用决策边界$β(Φ^{-1}(0))$的贝蒂数来表述。尽管这一度量标准很受欢迎，但并不适用于回归和其他不涉及决策边界的问题。相反，我们建议根据神经网络输出的贝蒂数$β(Φ(X))$来研究拓扑表现力。我们正式证明，具有结构$(d,\sum\limits_{k=1}^{d} β_k(3k+2),\sum\limits_{k=1}^{d}β_k(k+1))$的ReLU网络可以表达与任意贝蒂数等价的楔积空间。这为任意连接的（$β_0=1$）贝蒂数序列提供了上界，但通过构建与$N$个维度为$d_1,...,d_N$的球的乘积等价的空间，我们可以获得更好的界限（$\sum\limits_{n=1} d_n, 3\sum\limits_{n=1}^N d_n + 2N, \sum\limits_{n=1}^N d_n + N$），其贝蒂数由$\prod_n (1+x^{d_n})$给出。最后，我们提出了一个更强的严格界限猜想，为与球同胚的空间提供了结构界限（$d,d+1,d+1$）。
更新时间: 2026-10-09 17:55:33
领域: math.AT,cs.LG
AI Models Can Provably Hide Arbitrary Capabilities
Capability evaluations aim to surface dangerous model behaviors before deployment, but their reliability depends on the assumption that hidden capabilities can be elicited. We show this assumption does not hold under adversarial conditions by constructing backdoor attacks that embed encrypted neural networks within host model weights, decrypting and executing them only upon receiving a secret trigger. Using digital lockers, these hidden circuits remain provably hard to elicit and interpret under cryptographic assumptions, even given full access to model weights, extending prior work from hiding fixed strings to arbitrary computations. We implement our proof of concept constructions in PyTorch and evaluate elicitation through fine-tuning, weight noising, gradient-based trigger search, and latent adversarial training on a small chemical reaction prediction task - a proxy for CBRN-relevant capabilities. Our results demonstrate resistance to these methods and, together with our cryptographic guarantees, reveal a limit of elicitation-based capability audits. We release our models to support the improvement of defenses.
Updated: 2026-10-09 17:55:06
标题: AI模型可以显式隐藏任意能力
摘要: 能力评估旨在在部署之前暴露危险的模型行为，但其可靠性取决于隐藏能力可以被引出的假设。我们通过构建后门攻击来展示在对抗条件下这一假设并不成立，后门攻击将加密的神经网络嵌入主机模型权重中，只有在接收到秘密触发器时才解密并执行。使用数字锁定器，这些隐藏的电路在密码学假设下可以证明很难被引出和解释，即使完全访问模型权重，也可以延伸之前的工作从隐藏固定字符串到任意计算。我们在PyTorch中实现了我们的概念证明构造，并通过微调、权重噪声、基于梯度的触发器搜索以及潜在对抗训练在一个小的化学反应预测任务上进行评估-这是用于CBRN相关能力的代理。我们的结果表明对这些方法的抵抗力，并且结合我们的密码学保证，揭示了基于引出的能力审计的局限性。我们发布我们的模型以支持防御措施的改进。
更新时间: 2026-10-09 17:55:06
领域: cs.LG,cs.CR
KL Before and After Mixing: A Round-Complexity Separation in Parallel Sampling
A parallel sampler can choose at random which coordinates to generate together. For a target $p$ and the output law $q_U$ of a schedule $U$ drawn afresh for each output, we compare the schedule-averaged divergence $\mathbb{E}_U D_{\mathrm{KL}}(p\|q_U)$ with the output-law divergence $D_{\mathrm{KL}}(p\|\mathbb{E}_U q_U)$, in nats. Rounds commit independent draws from exact conditional marginals frozen at the round's start, each output's schedule is drawn before sampling from a finite mixture of schedules, and every scheduled round is charged. For the uniform distribution on binary strings of even length $n$ with $n/2$ ones, the least expected number of rounds at a fixed tolerance $\varepsilon>0$ is $Θ_\varepsilon((\log n)^2)$ under the first criterion but $(e^{-\varepsilon}+o(1))\log_2\ln n$ under the second, and the double-logarithmic order is attainable with the same round count on every run. For this target, a count-state recursion gives the exact least output divergence at every round cap over all causal randomized policies. The separation extends to weighted fixed-count targets with bounded weight ratios and quota fractions bounded away from zero and one, without permutation symmetry, and to the full-support Beta$(1,1)$--Bernoulli target, where the sharp expected cost under the output-law criterion is $(e^{-\varepsilon}/2+o(1))\log_2\log_2n$, attained with rare $n$-round runs. Prior work implies the order of the schedule-averaged cost on the balanced target and its $O_\varepsilon((\log n)^2)$ upper bounds on both extensions.
Updated: 2026-10-09 17:54:16
标题: 混合前后的KL散度：并行采样中的复杂度分离
摘要: 一个并行采样器可以随机选择要一起生成的坐标。对于目标$p$和每个输出都重新绘制的调度$U$的输出法则$q_U$，我们比较了调度平均散度$\mathbb{E}_U D_{\mathrm{KL}}(p\|q_U)$与输出法则散度$D_{\mathrm{KL}}(p\|\mathbb{E}_U q_U)$，以nats为单位。每一轮从准确的条件边缘中进行独立抽样，每个输出的调度在从一系列调度中抽样之前进行抽样，并且每个调度轮都会被计费。对于长度为偶数$n$且$n/2$个为1的二进制字符串的均匀分布，在固定容差$\varepsilon>0$下，预期最少轮数是$Θ_\varepsilon((\log n)^2)$，但在第二个标准下是$(e^{-\varepsilon}+o(1))\log_2\ln n$，双对数阶可以通过相同的轮数在每次运行中实现。对于这个目标，计数状态递归给出了所有因果随机策略中每轮上的最小输出散度。该分离扩展到带有有界权重比和远离零和一的配额分数的加权固定计数目标，没有置换对称性，并且对完全支持的Beta$(1,1)$--伯努利目标，输出法则标准下的尖锐预期成本是$(e^{-\varepsilon}/2+o(1))\log_2\log_2n$，在极少数的$n$轮运行中实现。先前的工作暗示了在平衡目标上的调度平均成本的顺序以及在两个扩展上的$O_\varepsilon((\log n)^2)$上界。
更新时间: 2026-10-09 17:54:16
领域: cs.LG,cs.IT
Goal-Oriented Bundle Management and Flow Control for Deep-Space Communications via Reinforcement Learning
Deep-space communications operate under intermittent connectivity, long propagation delays, and relay storage constraints. In such networks, the Bundle Protocol enables store-carry-forward communication, requiring relay nodes to manage limited storage and scarce contact opportunities. Relay-side scheduling and dropping decisions are inherently goal-oriented: transmission resources should be allocated to bundles most valuable to downstream mission objectives. Conventional schedulers, such as FIFO, do not explicitly optimize task utility and may provide unfair service across heterogeneous planetary sources. This paper presents a closed-loop modeling and learning framework for relay-side bundle and flow management in a contact-driven deep-space network, where lunar and Martian updates are forwarded through a shared-buffer relay to Earth. We quantify destination-side information freshness using excess age $Δ_i^e(t)=[Δ_i(t)-d_i(t)]^+$, where $Δ_i(t)$ denotes the Age of Information of source $i$ and $d_i(t)$ is a source-dependent delay baseline. Based on this metric, we define a goal-oriented loss that supports source-dependent nonlinear penalties, allowing prioritization according to mission relevance. We formulate relay control as a constrained semi-Markov decision process (CSMDP) and solve it using masked Lagrangian Proximal Policy Optimization (PPO). Feasibility-aware action masking restricts transmissions to bundles that can complete within the remaining contact window. DTNSim simulations indicate improved age-based utility and inter-flow fairness compared with baseline schedulers. The proposed framework jointly optimizes relay-side bundle scheduling, dropping, and flow control under contact-driven dynamics with explicit fairness considerations across heterogeneous sources.
Updated: 2026-10-09 17:53:46
标题: 面向目标的束管理和流控制在深空通信中的应用：基于强化学习的方法
摘要: 深空通信在间歇性连通性、长传播延迟和中继存储约束下运行。在这种网络中，Bundle协议实现了存储-携带-转发通信，要求中继节点管理有限的存储空间和稀缺的联系机会。中继端的调度和丢弃决策是目标导向的：传输资源应分配给对下游任务目标最有价值的数据包。传统的调度器，如FIFO，不显式优化任务效用，可能在异质行星来源之间提供不公平的服务。本文提出了一个闭环建模和学习框架，用于在基于联系的深空网络中管理中继侧数据包和流量，其中月球和火星更新通过共享缓冲中继转发到地球。我们使用过时程度$Δ_i^e(t)=[Δ_i(t)-d_i(t)]^+$来量化目的地端信息新鲜度，其中$Δ_i(t)$表示源$i$的信息时代，$d_i(t)$是源相关的延迟基线。基于这个指标，我们定义了一个目标导向的损失，支持源相关的非线性惩罚，允许根据任务相关性进行优先排序。我们将中继控制形式化为受约束的半马尔可夫决策过程（CSMDP），并使用掩码拉格朗日近端策略优化（PPO）进行解决。考虑可行性的行动掩码将传输限制在可以在剩余联系窗口内完成的数据包上。DTNSim模拟显示，与基准调度器相比，基于时代的效用和跨流公平性得到改善。所提出的框架共同优化中继端数据包调度、丢弃和流量控制，考虑到跨异构来源的明确公平性考虑。
更新时间: 2026-10-09 17:53:46
领域: cs.NI,cs.LG,eess.SY
DyssomniaNN: Multi-Scale Spiking Neural Modeling of Human Sleep Disorders and Perceived Restedness
Clinical sleep medicine repeatedly documents profound divergences between objective sleep duration and subjective daytime impairment, notably exemplified by sleep state misperception in paradoxical insomnia and boundary instability in narcolepsy type 1. Here, we present DyssomniaNN: a multi-scale biophysical computational framework coupling Borbély's two-process regulation of sleep homeostasis and circadian drive with Saper's hypothalamic reciprocal flip-flop circuit and a recurrent corticothalamic leaky integrate-and-fire spiking neural network. We formalize five clinical cohorts across an identical 72-hour photoperiod: healthy control, circadian rhythm sleep-wake disorder (delayed-phase), chronic insomnia, paradoxical insomnia, and narcolepsy type 1. The synthetic local field potential converges closely with empirical human electroencephalography spectra from the PhysioNet Sleep-EDF database across all canonical frequency bands (spectral mean absolute error of 0.33--0.35). Furthermore, we introduce the Well-Rested Index to quantify restorative efficacy. Simulations reveal that sleep duration fails to predict restorative recovery: paradoxical insomnia achieves normal total sleep time (21.15 h) yet yields low restoration (37.6%) from unsuppressed cortical beta-band intrusion, whereas hypocretin deficiency in narcolepsy produces severe nocturnal fragmentation (781 min of wake after sleep onset) and 35 sleep-onset rapid eye movement periods. DyssomniaNN provides an in silico platform for mechanistic sleep research and virtual neuropharmacology.
Updated: 2026-10-09 17:52:23
标题: DyssomniaNN：人类睡眠障碍和感知到的休息感的多尺度脉冲神经建模
摘要: 临床睡眠医学反复记录客观睡眠持续时间和主观白天功能障碍之间的深刻分歧，尤其在悖论性失眠中的睡眠状态误知和1型嗜睡症中边界不稳定方面得到明显体现。在这里，我们提出了DyssomniaNN：一个多尺度生物物理计算框架，将Borbély的睡眠稳态和昼夜节律的两过程调节与Saper的下丘脑互逆翻板回路和一个重复皮层丘脑泄漏积分-放电神经网络相结合。我们在一个相同的72小时光周期内形式化了五个临床群体：健康对照组、循环节律睡眠-觉醒障碍（延迟相）、慢性失眠、悖论性失眠和1型嗜睡症。合成的局部场电位与PhysioNet Sleep-EDF数据库中的实验人类脑电波谱密切一致，涵盖所有经典频率波段（频谱平均绝对误差为0.33-0.35）。此外，我们引入了休息指数来量化恢复效果。模拟表明，睡眠持续时间无法预测恢复性恢复：悖论性失眠实现了正常的总睡眠时间（21.15小时），但通过抑制皮质β波干扰，恢复效果低（37.6%），而嗜睡症中的佝偻素缺乏导致严重的夜间碎片化（睡眠开始后781分钟的清醒）和35次睡眠开始快速眼动期。 DyssomniaNN为机械性睡眠研究和虚拟神经药理学提供了一个体外平台。
更新时间: 2026-10-09 17:52:23
领域: cs.NE,cs.ET,cs.LG,q-bio.NC
Designing Protein Structures from Natural Language
A central challenge in de novo protein design is generating proteins with desired structural and functional properties. Existing design models typically require a dedicated input representation for each property, limiting the design objectives they support. We develop Prompteina, the first all-atom structure generation model that supports new and broader design objectives specified in natural language, analogous to text-to-image generation. To train Prompteina, we construct Prompteina-Data, a paired text--protein structure dataset of 1.8M structures and 50M+ captions describing global and residue-specific structural and functional properties. We adapt Proteina-Complexa, a pretrained structure generation model, through text-conditioned pretraining and finetuning that aligns residues with their corresponding text. Prompteina outperforms baselines on 15 of 18 structural properties and generates proteins that more closely match descriptions of their intended functions. In binder design, it leads baselines on at least 13 of 20 targets and improves interface confidence by 14\% over the strongest baseline. With iterative in silico feedback, Prompteina's gain over non-iterative generation is 57\% larger than the next-best baseline on the hardest targets. Finally, we experimentally validate Prompteina designs against GM2A, a target with no reported de novo binders. We identify nine binders, with the strongest at an apparent $K_{\mathrm{D}}$ of 107 nM. These results establish natural language as an effective and powerful conditioning interface for protein structure generation.
Updated: 2026-10-09 17:51:55
标题: 从自然语言设计蛋白质结构
摘要: 在从头设计蛋白质的过程中，一个核心挑战是生成具有所需结构和功能特性的蛋白质。现有的设计模型通常需要为每个特性提供专门的输入表示，从而限制了它们支持的设计目标。我们开发了Prompteina，这是第一个支持自然语言指定新的和更广泛设计目标的全原子结构生成模型，类似于文本到图像的生成。为了训练Prompteina，我们构建了Prompteina-Data，一个包含180万个结构和5000万多个描述全局和残基特定结构和功能特性的标题的文本-蛋白质结构数据集。我们通过文本条件预训练和微调，使预训练的结构生成模型Proteina-Complexa适应了残基与其对应文本的对齐。Prompteina在18种结构特性中的15种上优于基线，并生成的蛋白质更接近其预期功能的描述。在结合物设计中，它在至少20个目标中领先基线，在接口置信度上比最强基线提高了14\%。通过迭代的体外反馈，Prompteina相对于非迭代生成的增益比最好的基线在最困难的目标上大57\%。最后，我们通过实验验证了Prompteina设计对GM2A的效果，这是一个没有报道的从头设计结合物。我们鉴定了九个结合物，其中最强的表观$K_{\mathrm{D}}$为107 nM。这些结果确立了自然语言作为蛋白质结构生成的有效和强大的条件接口。
更新时间: 2026-10-09 17:51:55
领域: cs.LG,q-bio.BM
Revisiting Tabular Deep Regression: Objectives and the Limits of Global Early Stopping
While deep learning models for tabular data have recently received significant research attention, the choice of objective functions for regression tasks remains an understudied degree of freedom. Several lines of research offer alternatives to the standard Mean Squared Error (MSE), but differences in datasets, backbones, and evaluation protocols make their practical benefits difficult to compare. In this work, we address this gap by performing a systematic re-evaluation of regression objectives across a broad collection of established tabular datasets and multiple neural backbones. Our comparison reveals that many proposed objectives provide limited or inconsistent gains over a properly tuned MSE baseline, while quantile loss yields consistent aggregate improvements over MSE across the tested backbones. We then identify a limitation that remains even with this strong objective: regions with different levels of target noise can reach their best predictive performance at different training epochs. Consequently, a single global early-stopping checkpoint can impose a compromise between uncertainty regions, even when the objective can represent heteroskedastic target distributions. Motivated by this limitation, we introduce a heteroskedasticity-aware extension of quantile loss that uses input-dependent label smoothing as a potential remedy and improves over quantile loss on several backbones. Overall, our results provide a modern empirical picture of regression objectives for tabular deep learning and highlight how differences in optimal stopping times across uncertainty regions can limit the effectiveness of global early stopping. The code is available at https://github.com/yandex-research/tabular-regression.
Updated: 2026-10-09 17:49:32
标题: 重新审视表格式深度回归：目标和全局提前停止的限制
摘要: 尽管近年来深度学习模型在表格数据方面受到了重视，但回归任务的目标函数选择仍然是一个较少研究的自由度。一些研究方向提供了替代标准均方误差（MSE）的选择，但数据集、骨干网络和评估协议的差异使它们的实际优势难以比较。在本研究中，我们通过在广泛的已建立的表格数据集和多个神经网络骨干上进行系统的回归目标函数重新评估来填补这一空白。我们的比较显示，许多提出的目标函数相对于经过适当调整的MSE基准提供了有限或不一致的增益，而分位数损失在测试的骨干上始终比MSE提供了一致的总体改进。然后我们确定了一个即使使用这个强大的目标函数仍然存在的限制：具有不同目标噪声水平的区域可以在不同的训练时期达到最佳的预测性能。因此，即使目标可以表示异方差目标分布，单个全局早停止检查点也可以在不确定性区域之间施加妥协。受到这一限制的启发，我们引入了适用于分位数损失的异方差感知扩展，它使用依赖于输入的标签平滑作为潜在的补救措施，并在几个骨干上优于分位数损失。总的来说，我们的结果为表格深度学习的回归目标提供了一个现代经验图片，并强调了不确定性区域之间最佳停止时间的差异如何限制全局早停止的有效性。代码可在https://github.com/yandex-research/tabular-regression 上找到。
更新时间: 2026-10-09 17:49:32
领域: cs.LG,cs.AI
Hitting-Time Bounds for Asynchronous Q-Learning
We study the last iterate of standard asynchronous Q-learning from a single trajectory of a finite Markov decision process. The trajectory follows a fixed behavior policy whose state--action chain is irreducible and may be periodic. For discount factor $γ$, write $H=(1-γ)^{-1}$, and let $μ_{\min}$ be the minimum stationary state--action probability and $\mathfrak h_\star$ the largest expected hitting time of the reverse state--action chain. We prove that zero-initialized Q-learning with a constant step size achieves sup-norm error at most $\varepsilon$ with high probability using $\widetilde{O}(H^4/(μ_{\min}\varepsilon^2)+\mathfrak h_\star)$ transitions, for $0<\varepsilon\le1$. The statistical term retains the quartic effective-horizon dependence of standard Q-learning, while initialization cost is additive in $\mathfrak h_\star$. For aperiodic behavior chains, a comparison with mixing time yields a transient term $\widetilde{O}(t_{\mathrm{mix}}/μ_{\min})$. An empirical state--action count rule selects a constant step size for a subsequent training run and attains the same rate without prior knowledge of model parameters, target accuracy, or confidence level. The output is the last training iterate, with both stages included in the sample count. The proof uses two deterministic Bellman envelopes to separate statistical error from fixed-policy initialization error and adapts reverse-time hitting-cost compensation from TD learning to state--action propagation. An exploration lower bound holds for arbitrary predictable step sizes. Together with the existing statistical lower bound and its extension to pilot-selected constant steps, it yields matching mean-square rates up to logarithmic factors on the stated large-sample parameter classes.
Updated: 2026-10-09 17:44:10
标题: 异步Q学习的击中时间界限
摘要: 我们研究了从有限马尔可夫决策过程的单一轨迹中标准异步Q学习的最后迭代。该轨迹遵循一个固定的行为策略，其状态-动作链是不可约的，可能是周期性的。对于折扣因子$γ$，记$H=(1-γ)^{-1}$，令$μ_{\min}$为最小的稳态状态-动作概率，$\mathfrak h_\star$为逆状态-动作链的最大期望命中时间。我们证明，使用恒定步长的零初始化Q学习可以在高概率下通过$\widetilde{O}(H^4/(μ_{\min}\varepsilon^2)+\mathfrak h_\star)$次转换实现至多$\varepsilon$的超范数误差。统计项保留了标准Q学习的四次有效地平方依赖性，而初始化成本在$\mathfrak h_\star$中是可加的。对于非周期性的行为链，与混合时间的比较会产生一个瞬态项$\widetilde{O}(t_{\mathrm{mix}}/μ_{\min})$。一个经验的状态-动作计数规则为后续的训练运行选择一个恒定的步长，并在没有先验模型参数、目标准确度或置信水平的情况下实现相同的速率。输出是最后的训练迭代，包括了样本计数中的两个阶段。证明使用两个确定性贝尔曼包络来分离统计误差和固定策略初始化误差，并将逆时间命中成本补偿从TD学习调整到状态-动作传播。对于任意可预测步长，存在一个探索下界。与现有的统计下界以及其对飞行员选择的恒定步长的扩展一起，在所述的大样本参数类上得到匹配的均方速率，直到对数因子。
更新时间: 2026-10-09 17:44:10
领域: stat.ML,cs.LG
ResidencyRL: Reinforcement Learning in Simulated Clinical Environments
In medical education, physicians convert academic knowledge into clinical expertise through residency: years of training across thousands of encounters, with diverse sources of feedback and progressively greater autonomy. Much of clinical reasoning relies on the patient encounter, a dialogue in which a clinician elicits history, refines diagnostic hypotheses, and decides management under uncertainty. While large language models (LLMs) excel on static medical benchmarks, methods to optimize the full sequence of clinical decisions remain underdeveloped. We present ResidencyRL, a reinforcement learning (RL) method for training clinical artificial intelligence (AI) agents through simulated multi-turn clinical encounters (up to 60 dialogue turns and 8 tool calls per trajectory). ResidencyRL pairs the policy agent with LLM simulators capable of complex, adversarial behaviors, training against a structured reward aligned to diagnostic accuracy, management quality, communication, documentation, and safety. On held-out evaluations, the ResidencyRL agent improves diagnostic accuracy by 7.0% under adversarial conditions (88.0% vs. 81.0%) and reduces missed red flag rates by 31%, demonstrating rigorous mitigation of premature closure. Blinded expert clinicians validated these gains, preferring the trained agent in 87.6% of side-by-side comparisons. The procedural competencies transfer to unseen benchmarks: the agent outperforms the base model across all six clinical axes of the AMIE multi-visit benchmark, and shows consistent directional improvements on AgentClinic and CRAFT-MD. Our findings demonstrate that sequential clinical decision-making can be effectively learned through multi-turn RL in simulation, yielding robust, generalizable capabilities, paving the way towards clinical mastery. Prospective validation with real-world workflows remains necessary to establish clinical utility.
Updated: 2026-10-09 17:41:01
标题: 《ResidencyRL：在模拟临床环境中的强化学习》
摘要: 在医学教育中，医师通过住院医师培训将学术知识转化为临床专长：在成千上万次的接触中进行多年的培训，获得多样化的反馈来源并逐渐提高自主性。许多临床推理依赖于患者接触，这是一个对话过程，临床医生在其中获取病史，完善诊断假设，并在不确定性下进行管理决策。虽然大型语言模型（LLMs）在静态医学基准上表现出色，但优化完整的临床决策序列的方法仍未得到充分发展。我们提出了ResidencyRL，这是一种通过模拟多轮临床接触（每个轨迹可达到60个对话轮和8次工具调用）进行临床人工智能（AI）代理训练的强化学习（RL）方法。ResidencyRL将策略代理与能够进行复杂对抗行为的LLM模拟器配对，针对与诊断准确性、管理质量、沟通、文档记录和安全性一致的结构化奖励进行训练。在留置评估中，ResidencyRL代理在对抗条件下将诊断准确性提高了7.0％（88.0％对81.0％），并将红旗遗漏率降低了31％，表明对早期关闭进行了严格的缓解。蒙眼专家医生验证了这些收益，在87.6％的旁边对比中更喜欢经过训练的代理。程序能力转移到看不见的基准测试中：代理在AMIE多次访问基准测试的所有六个临床轴上均优于基础模型，并在AgentClinic和CRAFT-MD上显示了一致的方向性改进。我们的发现表明，通过模拟中的多轮RL可以有效学习顺序临床决策，产生强大的通用能力，为达到临床掌握铺平道路。需要通过真实世界工作流程的前瞻性验证来建立临床实用性。
更新时间: 2026-10-09 17:41:01
领域: cs.AI,cs.CL
Task Context Shapes Component Roles in Language Models
Large language models reuse the same attention heads across tasks, but identifying these shared heads does not explain what they do differently in each task. A natural source of inspiration is task-dependent interactions among shared brain regions, which motivate a network view of language-model computation. Here, we compare fixed model components across tasks using identical graph facts, statements specifying which directed connections are present. We measure how interventions affect attention heads and the activation state in later layers (influence), as well as the answer (importance). Task changes redistribute the influence in five open-weight models spanning four families and 1.5-14 billion parameters. In Gemma-2-9B and Qwen3-14B, an activation edit changes an early internal activation vector and switches which fact the model reads while the written instruction stays fixed. The same heads then respond to different graph facts and influence later components differently. For each prompt, we construct two linear response maps, one before the edit and one after it. Each map predicts a later head's response from a measured change in an earlier head. For a separate factual intervention, the map for the corresponding condition reduces mean squared prediction error by 53.6% in Gemma and 5.0% in Qwen compared with the average of the two maps. These findings show that task context changes both which graph facts the shared components respond to and how their outputs affect later computation. A network account may help explain how models use shared parameters to perform different computations. Measurements of task-dependent influence could also guide interventions intended to alter one behavior while preserving others.
Updated: 2026-10-09 17:40:53
标题: 任务背景塑造语言模型中的组件角色
摘要: 大型语言模型在任务之间重复使用相同的注意头，但是识别这些共享头并不能解释它们在每个任务中的不同之处。任务依赖的相互作用在共享的大脑区域之间是一个自然的灵感来源，这激发了对语言模型计算的网络视图。在这里，我们使用相同的图形事实比较跨任务的固定模型组件，这些事实指定了存在哪些定向连接。我们测量干预如何影响注意头和后续层的激活状态（影响），以及答案（重要性）。任务变化重新分配了跨四个系列和15-14亿参数的五个开放权重模型中的影响。在Gemma-2-9B和Qwen3-14B中，激活编辑改变了早期的内部激活向量，并切换了模型读取的事实，而书面指令保持不变。然后，相同的头对不同的图形事实作出响应，并以不同方式影响后续组件。对于每个提示，我们构建了两个线性响应映射，一个是在编辑之前，一个是在编辑之后。每个映射从早期头部的测量变化预测后续头部的响应。对于单独的事实干预，与两个映射的平均值相比，相应条件的映射在Gemma中将平均平方预测误差降低了53.6％，在Qwen中将其降低了5.0％。这些发现表明，任务背景改变了共享组件对哪些图形事实作出响应以及它们的输出如何影响后续计算。网络账户可能有助于解释模型如何使用共享参数执行不同的计算。任务依赖的影响的测量也可以指导旨在改变一个行为而保留其他行为的干预。
更新时间: 2026-10-09 17:40:53
领域: cs.CL,cs.AI
RATED: Reliability-Aware AI Agent Tool Retrieval via Execution Dynamics
As AI agents increasingly rely on external tools, accurate retrieval is essential for reliable behavior. Existing methods optimize semantic relevance but largely ignore tool availability during execution. To address this gap, we propose RATED (Reliability-Aware Tool Retrieval via Execution Dynamics), a retrieval framework that combines robust semantic matching with dynamic reliability adaptation. RATED employs a frozen dual-encoder with contrastive fine-tuning for semantic matching, and a lightweight contextual bandit module that re-ranks candidates based on real-time execution feedback. This design enables rapid adaptation to runtime tool failures without expensive model retraining. We introduce EcoAgent, a benchmark of 5,172 real-world tools and over 210,000 query-tool pairs for offline retrieval and dynamic reliability evaluation. Experimental results demonstrate that RATED achieves a 52.2% relative MRR improvement over the ToolLLM API retriever on real-world queries, and reduces fault selection rate from 86.2% to 2.9% under severe API outage scenarios with a 30% failure rate, while incurring negligible latency overhead of less than 1 ms over encode-only inference. The code and dataset are available at https://github.com/Elliotes/EcoAgent.
Updated: 2026-10-09 17:38:11
标题: RATED: 通过执行动态进行可靠性感知的人工智能代理工具检索
摘要: 随着人工智能代理越来越依赖外部工具，准确的检索对于可靠的行为至关重要。现有的方法优化语义相关性，但在执行过程中很大程度上忽视了工具的可用性。为了解决这一问题，我们提出了RATED（通过执行动态的可靠性感知工具检索），这是一个检索框架，它将稳健的语义匹配与动态的可靠性调整相结合。RATED采用了一个带有对比微调的冻结双编码器用于语义匹配，以及一个轻量级的上下文匪徒模块，根据实时执行反馈对候选进行重新排序。这种设计使得系统能够快速适应运行时工具故障，而无需昂贵的模型重新训练。我们介绍了EcoAgent，这是一个包含5,172个真实世界工具和超过210,000个查询-工具对的离线检索和动态可靠性评估基准。实验结果表明，RATED在真实世界查询中相对于ToolLLM API检索器实现了52.2%的相对MRR改进，并在API严重中断情况下将故障选择率从86.2%降低到2.9%，故障率为30%，同时产生的延迟开销不到1毫秒。代码和数据集可在https://github.com/Elliotes/EcoAgent获取。
更新时间: 2026-10-09 17:38:11
领域: cs.AI,cs.IR
The Illusion of AI Expertise Under Uncertainty: A Probabilistic Paradigm to Interpret Results Under Ground Truth Variation
Benchmarking the capabilities of AI systems, including Large Language Models (LLMs) and Vision Models, typically ignores the impact of uncertainty in the underlying ground truth answers from experts. This ambiguity is not just limited to human preferences, but is also consequential even in safety critical domains such as medicine where uncertainty is pervasive. In this paper, we introduce a probabilistic paradigm to theoretically explain how high certainty in ground truth answers is almost always necessary for even an expert to achieve high scores. In contrast, in datasets with high variation in ground truth answers, the expert performance drops substantially resulting in almost no difference between a random labeller and an expert. This characteristic also manifests when comparing models, where uncertainty obfuscates differences between poor and high performing models. Therefore, ignoring uncertainty in ground truth evaluation data can result in the misleading conclusion that a non-expert has similar performance to that of an expert. Given ground truth answer variability and commonly used F1 and accuracy metrics, we bring forth a simple and explainable probabilistic coin toss model to estimate expected accuracyand expected F1 and compare the estimated scores with those that experts (human or model) achieve across multiple domains -- including radiology and models such as GPT-5 and Gemini-3. The results lead to the recommendation that stratification by the probability of the ground truth answer becomes critical when expert performance is relatively low. Under stratified evaluation, performance comparison becomes more reliable in high certainty bins, mitigating the effect of the key confounding factor --- uncertainty.
Updated: 2026-10-09 17:38:00
标题: 在不确定性下的AI专业知识幻觉：一种概率范式来解释地面真相变化下的结果
摘要: 将AI系统的能力进行基准测试，包括大型语言模型（LLMs）和视觉模型，通常忽略了专家提供的基本真实答案的不确定性影响。这种模糊性不仅仅局限于人类偏好，甚至在医学等安全关键领域中也具有重要意义，因为不确定性是普遍存在的。在本文中，我们引入了一个概率范式，从理论上解释了即使对于专家来说，高确定性的基本真实答案几乎总是必要的才能取得高分。相比之下，在基本真实答案变化较大的数据集中，专家表现大幅下降，导致随机标记者和专家之间几乎没有区别。当比较模型时，这种特征也表现出来，不确定性使得质量较差和高性能模型之间的差异变得模糊。因此，忽略基本真实评估数据中的不确定性可能导致误导性结论，即非专家的表现与专家相似。考虑到基本真实答案的变化和常用的F1和准确性指标，我们提出了一个简单且可解释的概率抛硬币模型，用于估计预期准确性和预期F1，并将估计的分数与专家（人类或模型）在多个领域（包括放射学和GPT-5和Gemini-3等模型）中的表现进行比较。结果表明，在专家表现相对较低时，通过基本真实答案的概率分层变得至关重要。在分层评估下，性能比较在高确定性区间变得更加可靠，从而减轻了关键混杂因素-不确定性的影响。
更新时间: 2026-10-09 17:38:00
领域: cs.AI
X-AgentRed: Red-Teaming Personal Agents in Evolving Multi-Service Ecosystems
Personal agents increasingly connect to services across the web and applications by using tools. As the effects of their actions accumulate and interact across services, unintended or harmful outcomes can arise when agents fulfill benign user requests. Proactively identifying these compositional failures requires systematically exploring diverse cross-service paths. Existing red-teaming methods often simply prompt LLMs to identify potentially harmful paths from task and environment descriptions or decompose harmful objectives specified by humans or LLMs into seemingly benign action combinations. However, generation is biased toward patterns learned during training, potentially favoring familiar tool combinations and leaving novel cross-service interactions underexplored. To address this, we propose X-AgentRed, a self-evolving red-teaming framework that discovers harmful tool call paths through a graph of tool dependencies and expands and refines them using execution feedback. By analyzing each tool's effects, potential harms, and preconditions for execution, the red-teaming agent constructs a directed graph that captures how one tool's effects can satisfy another's execution preconditions. Recipes sampled from this graph are instantiated as realistic scenarios in which agents can produce harmful outcomes. Execution feedback enables self-improvement by diversifying paths that produce harm through tool substitution and refining preconditions at blocked steps in unsuccessful paths to enable further execution. Across four frontier agentic models, X-AgentRed achieves $1.77\times$ the harmful path rate of STAC and explores a more diverse set of tools, yielding $2.99\times$ the tool coverage on average.
Updated: 2026-10-09 17:34:44
标题: X-AgentRed: 在不断演变的多服务生态系统中进行红队行动的个人代理程序
摘要: 个人代理越来越多地通过工具连接到网络和应用程序中的服务。随着它们的行为效果在服务之间积累和互相作用，当代理满足良性用户请求时可能会出现意外或有害的结果。积极地识别这些组合失败需要系统地探索不同的跨服务路径。现有的红队方法通常只是促使LLM从任务和环境描述中识别潜在有害路径，或将由人类或LLM指定的有害目标分解为看似良性的动作组合。然而，生成过程倾向于在训练期间学习的模式，可能偏向熟悉的工具组合，并使新颖的跨服务交互得不到充分探索。为了解决这个问题，我们提出了X-AgentRed，一个自我进化的红队框架，通过工具依赖关系图发现有害的工具调用路径，并利用执行反馈扩展和细化它们。通过分析每个工具的效果、潜在危害和执行前提条件，红队代理构建了一个有向图，捕捉一个工具的效果如何满足另一个的执行前提条件。从这个图中采样出的配方被实例化为现实情景，代理可以在其中产生有害的结果。执行反馈通过通过工具替换多样化产生危害的路径，并通过在不成功的路径中阻塞步骤的细化前提条件来进一步执行，从而实现自我改进。在四个前沿代理模型中，X-AgentRed实现了STAC的1.77倍有害路径率，并探索了更多样化的工具，平均工具覆盖率提高了2.99倍。
更新时间: 2026-10-09 17:34:44
领域: cs.AI
Unified Convergence Theory of Stochastic and Variance-Reduced Cubic Newton Methods
We study stochastic Cubic Newton methods for solving general possibly non-convex minimization problems. We propose a new framework, which we call the helper framework, that provides a unified view of the stochastic and variance-reduced second-order algorithms equipped with global complexity guarantees. It can also be applied to learning with auxiliary information. Our helper framework offers the algorithm designer high flexibility for constructing and analyzing the stochastic Cubic Newton methods, allowing arbitrary size batches, and the use of noisy and possibly biased estimates of the gradients and Hessians, incorporating both the variance reduction and the lazy Hessian updates. We recover the best-known complexities for the stochastic and variance-reduced Cubic Newton, under weak assumptions on the noise. A direct consequence of our theory is the new lazy stochastic second-order method, which significantly improves the arithmetic complexity for large dimension problems. We also establish complexity bounds for the classes of gradient-dominated objectives, that include convex and strongly convex problems. For Auxiliary Learning, we show that using a helper (auxiliary function) can outperform training alone if a given similarity measure is small.
Updated: 2026-10-09 17:33:13
标题: 随机和方差减小的三次牛顿方法的统一收敛理论
摘要: 我们研究了用于解决一般可能非凸最小化问题的随机三次牛顿方法。我们提出了一个新的框架，称为辅助框架，提供了一种统一的视角，该视角涵盖了具有全局复杂性保证的随机和方差减少的二阶算法。它还可以应用于带有辅助信息的学习。我们的辅助框架为算法设计者提供了高度灵活性，用于构建和分析随机三次牛顿方法，允许任意大小的批处理，并使用噪声和可能有偏差的梯度和Hessian估计，结合方差减少和懒惰Hessian更新。在噪声的弱假设下，我们恢复了随机和方差减少的三次牛顿方法的最佳已知复杂性。我们理论的一个直接结果是新的懒惰随机二阶方法，显著改善了大维问题的算术复杂性。我们还为梯度主导目标类建立了复杂度界限，其中包括凸和强凸问题。对于辅助学习，我们表明，如果给定的相似度度量小，使用一个辅助（辅助函数）可以优于单独训练。
更新时间: 2026-10-09 17:33:13
领域: math.OC,cs.LG
Approximation of differential entropy in Bayesian optimal experimental design
Bayesian optimal experimental design provides a principled framework for selecting experimental configurations that maximize obtained information. In this work, we consider settings where the likelihood entropy is independent of the design or can be evaluated explicitly, which applies, e.g. to additive noise models with design-independent or explicitly computable noise entropy. This reduces the problem to maximum entropy estimation, where the nested integral defining the expected information gain collapses into a single integral. Our study is motivated by large-scale inference problems, such as inverse problems, where the computational cost is dominated by expensive likelihood evaluations. Under additive Gaussian noise model, we propose a computational approach in which the evidence density is approximated by a Monte Carlo or quasi-Monte Carlo surrogate, while the differential entropy is evaluated using standard methods without additional likelihood evaluations. We prove that this strategy achieves convergence rates that are comparable to, or better than, state-of-the-art methods for full expected information gain estimation, particularly when the cost of entropy evaluation is negligible. Our approach relies only on mild smoothness of the forward map and avoids stronger technical assumptions required in earlier work. We also present numerical experiments, which confirm some of our theoretical findings.
Updated: 2026-10-09 17:31:27
标题: 在贝叶斯最优实验设计中的差分熵近似
摘要: 贝叶斯最优实验设计提供了一个有原则的框架，用于选择最大化获得信息的实验配置。在这项工作中，我们考虑了似然熵独立于设计或可以明确评估的情形，例如适用于具有设计独立或可明确计算的噪声熵的加性噪声模型。这将问题简化为最大熵估计，其中定义期望信息增益的嵌套积分坍缩为单一积分。 我们的研究受到大规模推断问题的启发，例如逆问题，其中计算成本主要由昂贵的似然评估所主导。在加性高斯噪声模型下，我们提出了一种计算方法，其中证据密度通过蒙特卡洛或拟蒙特卡洛替代来近似，而微分熵则使用标准方法进行评估，无需额外的似然评估。我们证明，这种策略实现的收敛速度可与或优于全期望信息增益估计的最先进方法，特别是当熵评估成本可以忽略时。我们的方法仅依赖于前向映射的轻微光滑性，并避免了早期工作中需要更强技术假设的情况。我们还展示了一些数值实验，证实了我们的一些理论发现。
更新时间: 2026-10-09 17:31:27
领域: stat.ML,cs.LG,math.NA,stat.CO
RanSOM: Second-Order Momentum with Randomized Scaling for Constrained and Unconstrained Optimization
Second-order momentum corrects the lag of momentum behind the current gradient with a Hessian-vector product (HVP), but existing corrections need either a Lipschitz Hessian or extra forward and backward passes at an auxiliary point. We show that randomizing the step length removes this trade-off: if the step is exponentially distributed with mean $η$, a Stein-type identity makes $η$ times one HVP at the new iterate an estimate of the gradient change that is unbiased in expectation over the step, and this HVP shares its forward pass with the next gradient. The resulting method, RanSOM-E, works with normalized, sign, and Muon-type steps, and a Beta-distributed variant, RanSOM-B, keeps Frank-Wolfe iterates feasible. For objectives with Lipschitz gradient and heavy-tailed gradient and Hessian noise with tail indices $p, q \in (1,2]$, RanSOM-E attains the optimal oracle-complexity exponents for every pair $(p,q)$, including $p \neq q$, without a Lipschitz Hessian or clipping ($\mathcal{O}(\varepsilon^{-3})$ for $p = q = 2$). On CIFAR-10 with ResNet-20, we detect no accuracy difference from existing second-order momentum methods, and its mean update time is lower than that of the auxiliary-point method.
Updated: 2026-10-09 17:30:50
标题: RanSOM：具有随机缩放的二阶动量用于受约束和无约束优化
摘要: 二阶动量通过Hessian-vector product (HVP)纠正了动量在当前梯度后面的滞后，但现有的纠正方法需要一个Lipschitz Hessian或者在辅助点进行额外的前向和后向传递。我们展示了通过随机化步长可以消除这种折衷：如果步长呈指数分布，均值为$η$，则Stein类型的恒等式使得新迭代中$η$倍的一个HVP成为对梯度变化的估计，在步长期望上是无偏的，而且这个HVP与下一个梯度共享其前向传递。由此产生的方法，RanSOM-E，适用于标准化、符号和Muon类型的步长，而Beta分布的变体RanSOM-B使得Frank-Wolfe迭代可行。对于具有Lipschitz梯度和重尾梯度以及Hessian噪声的目标函数，尾指数为$p, q \in (1,2]$，RanSOM-E在每对$(p,q)$中实现了最佳的Oracle复杂度指数，包括$p \neq q$，而无需Lipschitz Hessian或剪辑（对于$p=q=2$为$\mathcal{O}(\varepsilon^{-3})$）。在CIFAR-10上使用ResNet-20，我们没有发现与现有的二阶动量方法相比的准确性差异，而且其平均更新时间低于辅助点方法。
更新时间: 2026-10-09 17:30:50
领域: math.OC,cs.LG
Stochastic Compositional Optimization via Hybrid Momentum Frank--Wolfe
Stochastic compositional optimization minimizes $F(f(x), x)$ over a compact convex set when the inner map $f$ is available only through noisy samples. We propose Hybrid Momentum Stochastic Frank--Wolfe, a projection-free method for outer functions $F$ that are jointly convex and Lipschitz but not necessarily differentiable, such as maxima, CVaR, and norms. The method tracks the Jacobian and the value of $f$ with momentum, using a Taylor correction for the value, and passes the tracked linearization to a generalized linear minimization oracle. Under finite-variance noise, it attains an expected generalized Frank--Wolfe gap of order $K^{-1/4}$ for non-convex objectives and, when the linearized models minorize the objective, a last-iterate primal error of order $K^{-1/3}$. The analysis extends to heavy-tailed noise without clipping, with separate moment orders for values and Jacobians. With an exact oracle, the method reduces to the deterministic method of Vladarean et al. (2023).
Updated: 2026-10-09 17:27:46
标题: 随机复合优化：通过混合动量Frank-Wolfe算法
摘要: 随机组合优化在仅通过嘈杂样本获取内部映射$f$时，通过在紧凸集上最小化$F(f(x), x)$来实现。我们提出了混合动量随机Frank--Wolfe方法，这是一种无投影的方法，适用于外部函数$F$同时具有凸性和Lipschitz性，但不一定可微，例如最大值、CVaR和范数。该方法使用动量跟踪$f$的雅可比矩阵和值，使用泰勒修正值，并将跟踪的线性化传递给广义线性最小化预测器。在有限方差噪声下，对于非凸目标，该方法达到了期望的广义Frank--Wolfe差距阶数为$K^{-1/4}$，当线性化模型较小化目标时，最后迭代的原始误差为$K^{-1/3}$。该分析扩展到不裁剪的重尾噪声，值和雅可比矩阵具有单独的矩阵顺序。在具有精确预测器的情况下，该方法简化为Vladarean等人(2023年)的确定性方法。
更新时间: 2026-10-09 17:27:46
领域: math.OC,cs.LG
Prediction Is Not a Sufficient Memory State
A memory that reproduces every current prediction may still have forgotten how to learn from the next observation. We characterize this distinction for exact cumulative learning and attention. For smooth convex residual losses with unrestricted real features and labels, a continuous recurrent state of fixed dimension exists exactly when the loss is polynomial. With ordinary, unweighted observations, degree $D$ requires precisely $r=\binom{d+D}{D}-1$ coordinates; even among histories with one fixed current predictor, exactly $r-d$ coordinates can remain necessary for continuation. Nonpolynomial losses have unbounded requirements on such equal-prediction histories. Bounded scalar logistic examples retain this obstruction after fixing any finite derivative jet. For ridge regression, the hidden evidence is the design matrix: the exact state costs $dp+d(d+1)/2$, versus $dp$ for the predictor. On a well-conditioned zero-prediction family, optimal $B$-bit compression has average one-update error of order $d^{-1}2^{-2B/s_d}$, $s_d=d(d+1)/2$, even with arbitrary randomized codes. Kernel attention gives a second exact closure classification. The results identify what future updates reveal, rather than proposing a new regression algorithm or asserting trained-model gains.
Updated: 2026-10-09 17:27:06
标题: 预测不是足够的记忆状态
摘要: 一个能够重现每次当前预测的记忆可能已经忘记了如何从下一个观察中学习。我们对精确的累积学习和注意力进行了表征。对于具有无限制实特征和标签的平滑凸残差损失，当损失为多项式时，确实存在一个固定维度的连续循环状态。对于普通的，非加权的观察，度为$D$的需要精确地$r=\binom{d+D}{D}-1$个坐标；即使在具有一个固定当前预测器的历史中，也确实有$r-d$个坐标可能是继续进行所必需的。非多项式损失对这种相等预测历史有无界的要求。有界的标量逻辑例子在固定任何有限导数阶时仍保留这种障碍。对于岭回归，隐藏证据是设计矩阵：精确状态的成本为$dp+d(d+1)/2$，而预测器的成本为$dp$。在一个良好条件的零预测家族中，最佳的$B$比特压缩平均一个更新错误的顺序为$d^{-1}2^{-2B/s_d}$，$s_d=d(d+1)/2$，即使使用任意随机代码。核心注意力提供了第二个精确的闭包分类。这些结果识别了未来更新所揭示的内容，而不是提出新的回归算法或声称经过训练的模型的收益。
更新时间: 2026-10-09 17:27:06
领域: cs.LG
Partially Informed Stochastic Optimization via Zeroth-Order Residual Tracking
Many optimization problems provide direct access to only part of the gradient, while the remaining component must be inferred from function evaluations. This structure arises in performative prediction, performative reinforcement learning, and bilevel optimization with a black-box lower level. We introduce Partially Informed Stochastic Optimization (PISO), which combines the available gradient component with a running estimate of the residual. By centring each zeroth-order observation on this estimate, PISO makes the variance due to random probing depend on the tracking error rather than the residual's magnitude. This yields a randomized sketch iteration with an explicit contraction mechanism and a principled interpretation of momentum. For smooth nonconvex objectives, we establish convergence guarantees that depend on the residual's variation and initial magnitude, without requiring a uniform bound on the unavailable gradient component. With independently sampled function evaluations and Gaussian or spherical probes, PISO attains leading function-evaluation complexities of \(O(d^{2}\varepsilon^{-6})\) under Lipschitz gradients and \(O(d^{2}\varepsilon^{-5})\) under an additional Lipschitz-Hessian assumption. These rates require only one probe direction, two function samples, and one known-component sample per update. The analysis also quantifies how batching trades observations for fewer deployments. Across five settings, PISO variants improve on the tested baselines with comparable oracle access in three, reach a comparable solution faster in bilevel routing, and perform best among methods without response-model derivatives in performative reinforcement learning.
Updated: 2026-10-09 17:20:41
标题: 部分信息随机优化通过零阶残差跟踪
摘要: 许多优化问题只提供梯度的部分信息，剩余部分必须通过函数评估进行推断。这种结构出现在表现预测、表现强化学习和具有黑盒子下层的双层优化中。我们引入了部分信息随机优化（PISO），它将可用的梯度部分与残差的运行估计结合起来。通过将每个零阶观察值集中在这个估计值上，PISO使由于随机探测而产生的方差取决于跟踪误差而不是残差的大小。这导致了一个带有明确收缩机制和动量合理解释的随机化草图迭代。对于光滑的非凸目标，我们建立了收敛保证，这些保证依赖于残差的变化和初始幅度，而不需要对不可用梯度部分进行统一限制。通过独立采样的函数评估和高斯或球形探测，PISO在Lipschitz梯度下达到了\(O(d^{2}\varepsilon^{-6})\)的领先功能评估复杂度，并在额外的Lipschitz-Hessian假设下达到了\(O(d^{2}\varepsilon^{-5})\)。这些速率仅需要一个探测方向，两个函数样本和一个已知组件样本每次更新。分析还 quantifies了如何通过批处理交换观察结果以减少部署。在五个设置中，PISO变体改进了测试基线，在三个设置中具有可比较的oracle访问权限，而在双层路由中更快达到可比较的解决方案，并且在表现强化学习中表现最佳，是没有响应模型导数方法中的最佳方法。
更新时间: 2026-10-09 17:20:41
领域: cs.LG
Global Consistency Repair for Natural-Language Claim Sets
NLP systems increasingly aggregate claims from retrieved passages, documents, and generated reasoning steps. Yet current factuality methods mostly verify claims individually or compare pairs of claims, leaving open the question of whether the entire aggregated claim set can be retained coherently. This misses a fundamental failure mode: claims may be locally plausible and pairwise compatible while still being jointly inconsistent. We formalize global consistency repair: given an inconsistent natural-language claim set, preserve the largest coherent subset while localizing the compact sources of inconsistency. We introduce QXR, an adaptive diagnosis-and-repair framework that treats an LLM as a noisy set-level consistency judge, localizes minimal conflicting subsets, and repairs the claim set by deleting a small hitting set of claims. Across LLM judges and benchmarks derived from VitaminC, FEVER, and conflicting-evidence RAG, QXR preserves substantially more valid claims than direct all-at-once judging while maintaining high precision. These results suggest that factuality in claim aggregation systems should be treated as a set-level repair problem, not only as local verification.
Updated: 2026-10-09 17:18:50
标题: 自然语言主张集的全球一致性修复
摘要: 自然语言处理系统越来越多地从检索到的段落、文档和生成的推理步骤中聚合主张。然而，当前的真实性方法大多是逐个验证声明或比较声明对，未解决整个聚合声明集能否保持一致的问题。这忽略了一个基本的失败模式：声明可能在局部上是合理的，两两之间是兼容的，但在联合上却是不一致的。我们形式化了全局一致性修复：给定一个不一致的自然语言声明集，保留最大的连贯子集，同时定位出不一致性的紧凑来源。我们引入了QXR，一个自适应的诊断和修复框架，将LLM视为一个嘈杂的集合级别一致性评判器，定位最小的冲突子集，并通过删除一小部分命中的声明修复声明集。在基于VitaminC、FEVER和冲突证据RAG的LLM评判器和基准上，QXR保留了比直接一次性判断更多的有效声明，同时保持高精度。这些结果表明，在主张聚合系统中，事实应被视为一个集合级别的修复问题，而不仅仅是本地验证。
更新时间: 2026-10-09 17:18:50
领域: cs.AI
Mixture of Many Zero-Compute Experts: A High-Rate Quantization Theory Perspective
This paper uses classical high-rate quantization theory to provide new insights into mixture-of-experts (MoE) models for regression tasks. Our MoE is defined by a segmentation of the input space to regions, each with a single-parameter expert that acts as a constant predictor with zero-compute at inference. Motivated by high-rate quantization theory assumptions, we assume that the number of experts is sufficiently large to make their input-space regions very small. This lets us study the approximation error of our MoE model class: (i) for one-dimensional inputs, we formulate the test error and its minimizing segmentation and experts; (ii) for multidimensional inputs, we formulate an upper bound for the test error and study its minimization. Moreover, we consider the learning of the expert parameters from a training dataset, given an input-space segmentation, and formulate their statistical learning properties. This leads us to theoretically and empirically show how the tradeoff between approximation and estimation errors in MoE learning depends on the number of experts.
Updated: 2026-10-09 17:16:42
标题: 许多零计算专家的混合：高速量化理论视角
摘要: 本文利用经典的高速率量化理论，为回归任务的专家混合（MoE）模型提供了新的见解。我们的MoE通过将输入空间分割成区域来定义，每个区域都有一个单参数专家，其在推理时作为一个具有零计算的恒定预测器。受高速率量化理论假设的启发，我们假设专家数量足够大，使它们的输入空间区域非常小。这使我们能够研究我们的MoE模型类的逼近误差：（i）对于一维输入，我们制定了测试误差及其最小化分割和专家；（ii）对于多维输入，我们制定了测试误差的上限，并研究其最小化。此外，我们考虑从训练数据集中学习专家参数，给定输入空间分割，并制定它们的统计学习特性。这使我们能够在理论和实证上展示MoE学习中逼近和估计误差之间的权衡如何取决于专家数量。
更新时间: 2026-10-09 17:16:42
领域: cs.LG
Near-Minimax-Optimal Rates for Asynchronous TD Learning with Markovian Data
We study the last iterate of standard importance-weighted temporal-difference (TD) learning from a single trajectory of a finite Markov decision process. The trajectory follows a behavior policy $b$ with an irreducible, possibly periodic state chain; a target policy $π$ covered by $b$ is evaluated with action importance ratios bounded by $\varrho$. Write $H=(1-γ)^{-1}$, with $μ_{\min}$ the minimum stationary probability and $\mathfrak h_\star$ the largest expected hitting time of the reverse behavior state chain. We prove that the last iterate achieves sup-norm error at most $\varepsilon$ with high probability after $\widetilde O(\varrho H^3/(μ_{\min}\varepsilon^2)+\mathfrak h_\star)$ transitions for $0<\varepsilon\le1$. The statistical term retains the cubic effective-horizon dependence of synchronous TD, while the initialization cost is additive in $\mathfrak h_\star$. Setting $b=π$ and $\varrho=1$ recovers the on-policy rate. For aperiodic behavior chains, $\mathfrak h_\star=\widetilde O(t_{\rm mix}/μ_{\min})$ recovers a mixing-time bound. The proof combines an anchored local Poisson equation in reverse time to control fluctuations without a mixing-time factor, a compensation argument for successive hitting costs, and an expected cumulative propagation mass bound for decreasing step sizes. Given the policy ratios and their bound, an empirical state-count rule attains the same rate without prior knowledge of model parameters. The framework covers nonnegative affine importance-weighted prediction updates, including successor features, transition-dependent discounting, and coupled finite-mode predictions. A perturbation estimate yields a time-uniform guarantee for online return-variance estimation. A three-state, two-action construction gives matching minimax lower bounds for both terms up to logarithmic factors on the considered parameter classes.
Updated: 2026-10-09 17:16:04
标题: 异步TD学习在马尔可夫数据下的近最小最优速率
摘要: 我们研究了从有限马尔可夫决策过程的单个轨迹中标准重要性加权时序差分（TD）学习的最后迭代。该轨迹遵循一个行为策略$b$，具有一个不可约的、可能是周期性状态链；一个由$b$覆盖的目标策略$π$通过动作重要性比$\varrho$进行评估。定义$H=(1-γ)^{-1}$，其中$μ_{\min}$是最小静态概率，$\mathfrak h_\star$是反向行为状态链的最大期望击中时间。我们证明，在经过$\widetilde O(\varrho H^3/(μ_{\min}\varepsilon^2)+\mathfrak h_\star)$次转换后，最后一次迭代以高概率至多达到$\varepsilon$的上限误差。统计项保留了同步TD的立方有效视野依赖性，而初始化成本在$\mathfrak h_\star$中是可加的。当$b=π$且$\varrho=1$时，恢复了同策略速率。对于非周期性行为链，$\mathfrak h_\star=\widetilde O(t_{\rm mix}/μ_{\min})$恢复了混合时间上界。证明结合了反向时间中的锚定局部泊松方程来控制波动而不需要混合时间因子，连续击中成本的补偿论证，以及针对减小步长的期望累积传播质量上界。在给定策略比率及其上限的情况下，经验状态计数规则可达到相同速率，无需了解模型参数。该框架涵盖了非负仿射重要性加权预测更新，包括后继特征、转移相关折扣和耦合有限模式预测。一个扰动估计提供了在线返回方差估计的时间一致保证。一个三状态、两动作的构造在考虑的参数类上给出了匹配的极小值下界，直到对数因子。
更新时间: 2026-10-09 17:16:04
领域: stat.ML,cs.LG
SURF: Steering the Scalarization Weight to Uniformly Traverse the Pareto Front
Scalarization is widely used in multi-objective optimization owing to its simplicity and scalability. In many applications, the goal is to generate solutions that represent diverse user preferences, ideally with uniform coverage of the Pareto front (PF). However, uniformly sampling scalarization weights usually induces non-uniform coverage of the PF. We explain this mismatch through the geometry of the scalarization path. As the weight varies, the corresponding solutions traverse the PF at a generally nonuniform speed. Integrating this speed gives cumulative arc length. After normalization by the total length, it defines a cumulative distribution function (CDF) on weights whose inverse selects points at equal arc-length intervals. Building on this insight, we propose SURF (Sampling Uniformly along the Pareto Front). For structured problems, including bi-objective bandits, we derive analytical expressions for this CDF map and the resulting PF-aware weight sampling rule. For general problems, SURF alternates between CDF reconstruction and weight sampling. Theoretically, we show that under provable conditions for bi-objective LS, SURF's CDF reconstruction error converges linearly to an $O(N^{-2})$ discretization floor. Empirically, experiments on bandits, multi-objective-gymnasium, and multi-objective LLM alignment demonstrate that SURF efficiently achieves more uniform PF coverage than baselines.
Updated: 2026-10-09 17:13:29
标题: SURF：引导标量化权重以均匀遍历帕累托前沿
摘要: 标量化在多目标优化中被广泛使用，因其简单性和可扩展性。在许多应用中，目标是生成代表不同用户偏好的解决方案，理想情况下应具有帕累托前沿（PF）的均匀覆盖。然而，均匀采样标量化权重通常会导致PF的非均匀覆盖。我们通过标量化路径的几何性质来解释这种不匹配。随着权重的变化，相应的解决方案以一般非均匀的速度遍历PF。将这种速度积分得到累积弧长。经过总长度的归一化后，它定义了一个权重上的累积分布函数（CDF），其逆选择在相等弧长间隔上的点。基于这一洞察力，我们提出了SURF（沿帕累托前沿均匀采样）。对于结构化问题，包括双目标赌博机，我们为这个CDF映射和由此产生的PF感知权重采样规则导出了解析表达式。对于一般问题，SURF在CDF重建和权重采样之间交替进行。理论上，我们证明在双目标LS的可证条件下，SURF的CDF重建误差会线性收敛到$O(N^{-2})$的离散化底线。在实证上，对赌博机、多目标体育馆和多目标LLM对齐的实验表明，SURF比基线能够更有效地实现更均匀的PF覆盖。
更新时间: 2026-10-09 17:13:29
领域: cs.LG,math.OC,stat.ML
RPP: A Certified Poisoned-Sample Detection Framework for Backdoor Attacks under Dataset Imbalance
Deep neural networks are highly susceptible to backdoor attacks, yet most defense methods to date rely on balanced data, overlooking the pervasive class imbalance in real-world scenarios that can amplify backdoor threats. This paper presents the first in-depth investigation of how the dataset imbalance amplifies backdoor vulnerability, showing that (i) the imbalance induces a majority-class bias that increases susceptibility and (ii) conventional defenses degrade significantly as the imbalance grows. To address this, we propose Randomized Probability Perturbation (RPP), a certified poisoned-sample detection framework that operates in a black-box setting using only model output probabilities. For any inspected sample, RPP determines whether the input has been backdoor-manipulated, while offering provable within-domain detectability guarantees and a probabilistic upper bound on the false positive rate. Extensive experiments on five benchmarks (MNIST, SVHN, CIFAR-10, TinyImageNet and ImageNet10) covering 10 backdoor attacks and 12 baseline defenses show that RPP achieves significantly higher detection accuracy than state-of-the-art defenses, particularly under dataset imbalance. RPP establishes a theoretical and practical foundation for defending against backdoor attacks in real-world environments with imbalanced data.
Updated: 2026-10-09 17:09:10
标题: RPP: 一种针对数据集不平衡下后门攻击的认证中毒样本检测框架
摘要: 深度神经网络极易受到后门攻击的影响，然而迄今为止大多数防御方法依赖于平衡数据，忽视了现实场景中普遍存在的类别不平衡问题，这可能会加剧后门威胁。本文首次深入研究了数据集不平衡如何放大后门脆弱性，结果表明：（i）不平衡导致了多数类偏差，增加了易受攻击的可能性；（ii）随着不平衡程度的增加，传统的防御方法明显失效。为了解决这个问题，我们提出了随机概率扰动（RPP），这是一个在黑盒设置中仅使用模型输出概率的可认证毒样本检测框架。对于任何检查的样本，RPP确定输入是否被后门操纵，同时提供可证明的领域内可检测保证和假阳性率的概率上限。对五个基准数据集（MNIST、SVHN、CIFAR-10、TinyImageNet和ImageNet10）进行了大量实验，涵盖10种后门攻击和12种基准防御方法，结果显示RPP在数据集不平衡情况下实现了显著更高的检测准确性，特别是在实验中。RPP为在具有不平衡数据的现实环境中抵御后门攻击奠定了理论和实践基础。
更新时间: 2026-10-09 17:09:10
领域: cs.CR,cs.CV,cs.LG
Second-Order Optimization with Asynchronous Curvature
Second-order methods make fast progress per iteration, but forming and factorizing the Hessian often dominates their wall-clock time. We study \emph{Split-Client}, a task-parallel design in which a curvature client factorizes Hessians in the background while a gradient client takes cubic-regularized steps with the most recently published matrix, so that curvature cost appears as a delay rather than as blocking time. For delayed exact Hessians, we prove a first-order stationarity bound that depends on the average age of the curvature rather than its maximum, using an overlap count and a weighted Hardy inequality that allows any bounded-error initial matrix; the analysis extends to subsampled and uniformly inexact curvature. Under a fixed-cost model, the resulting wall-clock bound matches the scaling of optimally tuned Lazy Hessian without choosing a reuse period. In experiments on two LIBSVM datasets, an implementation decreases the objective earlier in wall-clock time than Lazy and Vanilla cubic Newton, with the largest gains at intermediate accuracy.
Updated: 2026-10-09 17:05:28
标题: 使用异步曲率的二阶优化
摘要: 二阶方法在每次迭代中取得快速进展，但形成和分解Hessian矩阵通常占据了大部分时间。我们研究了“Split-Client”，这是一种任务并行设计，其中一个曲率客户端在后台对Hessian矩阵进行因式分解，而一个梯度客户端则使用最近发布的矩阵进行三次正则化步骤，使曲率成本表现为延迟而非阻塞时间。对于延迟的精确Hessian矩阵，我们证明了一个一阶稳定性界限，该界限取决于曲率的平均年龄而非最大值，使用了重叠计数和加权Hardy不等式，允许任何有界误差的初始矩阵；该分析扩展到子采样和统一不精确的曲率。在固定成本模型下，得到的墙钟界限与最佳调整的Lazy Hessian的比例相匹配，而无需选择重用周期。在两个LIBSVM数据集上的实验中，一种实现在墙钟时间较早地降低了目标函数，比Lazy和Vanilla三次牛顿法更快，而在中等精度下获得了最大的收益。
更新时间: 2026-10-09 17:05:28
领域: math.OC,cs.LG
Decision-Focused Learning for Cyber-Physical Risk-Aware Power Grid Operations
Power grid cyber-physical security requires intrusion detection methods that account for the operational consequences of detection errors. However, the existing intrusion detection approaches primarily focus on the predictive performance of learning algorithms, while downstream operational costs are often optimized independently, which can lead to inferior operational performance. To help alleviate this problem, we propose a novel optimization-learning cyber-physical-risk-informed framework that communicates iteratively to minimize power grid operational costs. This framework translates machine learning prediction errors into power grid operational cyber-physical-risk costs. Experimental results on two power grid operation datasets, Industrial Control System (ICS) Cyber Attack and Sherlock, show that the proposed approach improves power grid operation costs compared with the traditional approach.
Updated: 2026-10-09 17:02:30
标题: 面向决策的学习在网络物理风险感知电网运营中的应用
摘要: 电网网络物理安全需要入侵检测方法，考虑到检测错误的操作后果。然而，现有的入侵检测方法主要集中在学习算法的预测性能上，而下游操作成本通常独立优化，这可能导致较差的操作性能。为了帮助缓解这个问题，我们提出了一种新颖的优化学习的网络物理风险信息框架，通过迭代通信来最小化电网操作成本。该框架将机器学习预测错误转化为电网操作的网络物理风险成本。在两个电网操作数据集工业控制系统（ICS）网络攻击和神探上的实验结果表明，所提出的方法相比传统方法改善了电网操作成本。
更新时间: 2026-10-09 17:02:30
领域: cs.CR
Particle-based Generalised Stochastic Optimisation
We develop a class of general diffusion-based stochastic particle optimisation methods for loss functions with intractable gradients. Specifically, we consider problems in which the loss gradient is an integral with respect to a parameter-dependent distribution, a structure that includes training generative models, fine-tuning, and learning latent-variable models. We introduce mean-field dynamics and its interacting-particle approximations, which contain several existing algorithms as special cases and provide a route to constructing new methods. Under well-posedness and joint contractivity assumptions, we prove exponential convergence and show that the continuous-time particle system admits a non-asymptotic error bound. We illustrate our framework by developing higher-order Langevin and momentum optimisers and evaluating them on maximum marginal-likelihood estimation and energy-based-model training.
Updated: 2026-10-09 17:01:27
标题: 基于粒子的广义随机优化
摘要: 我们开发了一类基于扩散的随机粒子优化方法，用于具有难以处理梯度的损失函数。具体来说，我们考虑损失梯度是关于参数相关分布的积分的问题，这种结构包括训练生成模型、微调和学习潜变量模型。我们引入平均场动力学及其相互作用粒子的近似，其中包含几种现有算法作为特例，并提供构建新方法的途径。在良定性和联合收缩性假设下，我们证明指数收敛，并展示连续时间粒子系统具有非渐近误差上界。我们通过开发高阶 Langevin 和动量优化器，并在最大边缘似然估计和基于能量的模型训练上评估它们来说明我们的框架。
更新时间: 2026-10-09 17:01:27
领域: stat.ML,cs.LG,stat.CO
Finite-Sample Guarantees for Fully Discrete Conditional Diffusion Models
We establish finite-sample guarantees for conditional diffusion models learned from data and implemented with finitely many denoising steps. Our approach aligns the score-training objective with the sampling grid, allowing statistical estimation and numerical discretization errors to be analyzed together. Under a smooth conditional density model, we derive high-probability bounds on the generated distribution's error and quantify the required number of score evaluations. For every smoothness index $β>0$, a suitable grid yields averaged conditional Kullback--Leibler (KL) divergence $\widetilde O_p(n^{-2β/(d_x+d_y+2β)})$ relative to the original target with or without early stopping. Both the full and early-stopped samplers attain average target total-variation (TV) error $\widetilde O_p(n^{-β/(d_x+d_y+2β)})$. Here $n$ is the sample size, $d_x,d_y$ are the data and conditioning dimensions, and errors are averaged over the conditioning variable. The total-variation bound matches the rate established in existing conditional distribution-estimation theory, up to logarithmic factors.
Updated: 2026-10-09 16:59:18
标题: 离散条件扩散模型的有限样本保证
摘要: 我们为从数据学习并通过有限次去噪步骤实现的条件扩散模型建立了有限样本保证。我们的方法将评分训练目标与采样网格对齐，允许一起分析统计估计和数值离散化误差。在平滑条件密度模型下，我们推导了生成分布误差的高概率界限，并量化了所需的评分次数。对于每个平滑度指数$β>0$，合适的网格相对于原始目标（无论是否提前停止）产生平均条件Kullback-Leibler（KL）散度$\widetilde O_p(n^{-2β/(d_x+d_y+2β)})$。完整的和提前停止的采样器都达到了平均目标总变差（TV）误差$\widetilde O_p(n^{-β/(d_x+d_y+2β)})$。这里$n$是样本大小，$d_x,d_y$是数据和条件维度，误差是在条件变量上平均的。总变差界限与现有条件分布估计理论中建立的速率相匹配，直到对数因子。
更新时间: 2026-10-09 16:59:18
领域: stat.ML,cs.LG
Inference-Time Steering Across Scaffolds and Epitopes for Sample-Efficient De Novo Nanobody Design
Generative molecular design typically relies on over-generation and post-hoc filtering. Inference-time steering methods like Feynman-Kac (FK) steering improve efficiency by biasing trajectories toward high-reward candidates, yet they optimize within a single fixed configuration. In de novo nanobody design, upfront commitment to conditioning choices like scaffold identity or target epitope creates a bottleneck. Since yield varies across these conditions, steering unpromising setups wastes compute. We recast generative condition selection as an inference-time resource-allocation problem. We propose two dynamic budget allocators: a joint Global Sequential Monte Carlo (SMC) formulation that steers and allocates compute across configurations concurrently, and a decoupled approach adapting Batched Successive Elimination (BaSE). Empirically, these dynamic allocators increase the yield of designable candidates per GPU-hour nearly 5$\times$ without sacrificing diversity across distinct therapeutic targets and generative configurations.
Updated: 2026-10-09 16:54:05
标题: 对于高效De Novo纳米体设计的推理时间跨越支架和表位
摘要: 生成式分子设计通常依赖于过度生成和事后过滤。像费曼-卡克（FK）导向这样的推理时间导向方法通过将轨迹偏向高奖励候选者来提高效率，然而它们在单个固定配置内进行优化。在全新的纳米体设计中，对蛋白质框架身份或靶标表位等条件的预先承诺会造成瓶颈。由于产量在这些条件之间变化，导向不利设置会浪费计算资源。我们将生成式条件选择重新构建为推理时间资源分配问题。我们提出了两种动态预算分配器：一个联合全局顺序蒙特卡罗（SMC）公式，同时在不同配置之间引导和分配计算资源，以及一种适应批处理连续消除（BaSE）的分离方法。从经验上看，这些动态分配器在不牺牲不同治疗靶点和生成配置之间的多样性的情况下，每个GPU小时可增加可设计候选者的产量近5倍。
更新时间: 2026-10-09 16:54:05
领域: cs.LG
Vision2Code: A Multi-Domain Benchmark for Evaluating Image-to-Code Generation
We study whether vision-language models (VLMs) can reconstruct structured, graphics-based images by writing executable code to generate them. Existing benchmarks on image-to-code generation are generally limited in the domains they cover, rely on reference source code for evaluation, and do not provide metrics to quantify domain-specific issues. To address this, we introduce Vision2Code, a multi-domain, reference-code-free benchmark containing 539 test examples from 15 different datasets, spanning charts and plots, geometry, graphs, scientific diagrams, documents and 3D spatial scenes. Evaluation is done by executing the generated programs and comparing their rendered outputs to the source images using dataset- and domain-specific rubrics for a VLM judge. We find that the final scores obtained in this manner have substantially higher correlation with human ratings than those obtained via generic VLM rubrics or standard pixel-based, embedding-based or perceptual similarity metrics. We further demonstrate the reliability of our evaluation under repeated sampling from the VLM judges, varying rubric weights, and in the presence of controlled errors. Our results highlight the variability across domains of existing image-to-code generation models, with accuracy on charts and graphs being significantly higher than other domains. Lastly, we demonstrate the utility of our scoring method for rejection sampling of generated programs for fine-tuning, improving Qwen3.5-9B by 16% and Phi-4-MM by 28% compared to their base checkpoints. Our code and data are publicly available at https://image2code.github.io/vision2code/.
Updated: 2026-10-09 16:51:01
标题: Vision2Code：用于评估图像生成代码的多领域基准测试
摘要: 我们研究了视觉-语言模型（VLMs）是否能通过编写可执行代码来生成结构化的基于图形的图像。现有的图像到代码生成的基准通常在其涵盖的领域上有限，依赖于用于评估的参考源代码，并且不提供用于量化特定领域问题的指标。为了解决这个问题，我们引入了Vision2Code，这是一个多领域、无需参考代码的基准，包含来自15个不同数据集的539个测试示例，涵盖图表和图形、几何、图形、科学图表、文档和3D空间场景。评估是通过执行生成的程序，并使用数据集和特定领域的规则与源图像的渲染输出进行比较来进行的，用于VLM评分。我们发现，通过这种方式获得的最终得分与人类评分之间的相关性远高于通过通用VLM规则或标准像素、嵌入或感知相似性指标获得的得分。我们进一步展示了我们的评估在重复抽样、变化的规则权重以及存在控制错误的情况下的可靠性。我们的结果突显了现有图像到代码生成模型在不同领域之间的差异，图表和图形的准确性显著高于其他领域。最后，我们展示了我们的评分方法在拒绝抽样生成的程序进行微调、改进Qwen3.5-9B和Phi-4-MM的效用，分别比它们的基准检查点提高了16%和28%。我们的代码和数据公开可在https://image2code.github.io/vision2code/上获取。
更新时间: 2026-10-09 16:51:01
领域: cs.CV,cs.LG
Returning to Development States: When Full Sandbox Snapshots Help Coding Agents
Coding agents often make useful progress before they fail, but retry-from-scratch discards the machine state created along the way. This makes failure recovery an interesting search problem: if intermediate environment states preserve meaningful progress, restoring and branching from them could outperform repeatedly restarting from the initial state. We implement Go-Explore-style search over full Daytona snapshots and find that higher and more efficient solve rates depend mainly on progress that sits outside the repository, such as databases, services, and other runtime state. Our results suggest that full-snapshot search is most effective when long-horizon coding work spans multiple services whose state cannot be reconstructed from a single diff.
Updated: 2026-10-09 16:50:19
标题: 回归发展状态：全沙箱快照如何帮助编码代理
摘要: 编码代理通常在失败之前取得有用的进展，但重新从头开始会丢弃沿途创建的机器状态。这使得失败恢复成为一个有趣的搜索问题：如果中间环境状态保留了有意义的进展，那么从这些状态恢复和分支可能会优于反复从初始状态重新启动。我们实现了Go-Explore风格的搜索，对完整的Daytona快照进行搜索，并发现更高效的解决率主要取决于存储库之外的进展，例如数据库、服务和其他运行时状态。我们的结果表明，当长期编码工作跨越多个服务时，完整快照搜索效果最好，这些服务的状态无法从单个差异重建。
更新时间: 2026-10-09 16:50:19
领域: cs.LG,cs.SE
Studying Comics at Scale with Artificial Intelligence: The PAGE Framework and Three Studies of American Comics and Japanese Manga
Comics are one of the great visual records of the twentieth century: hundreds of thousands of pages in which a mass readership saw its leaders, its wars and its heroes. Most of this record has never been studied as a whole; comics scholarship proceeds one book at a time. We argue that artificial intelligence, including generative models, can change this to the benefit of the medium itself. We propose PAGE, a framework that makes artificial intelligence an instrument for comics research. It parses every scan into panels, text and a bibliographic record; annotates each page with a vision-language model that follows a written codebook, as a human coder would; grounds samples of these readings in human annotation; and explains change across an archive, with estimates controlled for publisher and genre. We demonstrate the framework on 3,929 American comic books (1938--1956) and 87 volumes of Japanese manga, in three studies, each resting on a different capability. The first pairs text search with visual reading to recover how a mass medium portrayed political power. The second measures the page's visual form across an industry and recovers the hand of individual artists. The third follows who speaks across two comics traditions, checked against human annotation. Beyond research, the same instruments can make archives searchable by what is drawn, open comics to blind and low-vision readers, and help preserve fragile collections and credit uncredited artists. We outline a research agenda in which artificial intelligence strengthens the study, preservation and reach of the medium.
Updated: 2026-10-09 16:49:29
标题: 用人工智能研究大规模漫画：PAGE框架和三个关于美国漫画和日本漫画的研究
摘要: 漫画是20世纪伟大的视觉记录之一：数以十万计的页面展示了大众读者看到的领导人、战争和英雄。大部分这些记录从未被作为一个整体进行研究；漫画学术研究是一本一本地进行。我们认为，包括生成模型在内的人工智能可以改变这一状况，从而使媒介本身受益。我们提出了PAGE框架，使人工智能成为漫画研究的工具。该框架将每个扫描解析为面板、文本和文献记录；用一个遵循书写代码的视觉语言模型为每一页进行注释，就像人类编码者一样；将这些阅读样本基于人类注释进行比较；并解释存档中的变化，根据出版商和流派进行控制。我们在三个研究中展示了这一框架，涵盖了3,929本美国漫画书（1938-1956年）和87册日本漫画。每个研究都依赖于不同的能力。第一项研究将文本搜索与视觉阅读相结合，以恢复大众媒体如何描绘政治权力。第二项研究测量了整个行业的页面视觉形式，并恢复了个别艺术家的手笔。第三项研究跟踪两种漫画传统中的发言者，经过人类注释的核查。除了研究之外，同样的工具可以使存档可通过所绘内容进行搜索，让盲人和低视力读者能够阅读漫画，并有助于保护脆弱的收藏品并给未署名的艺术家以认可。我们概述了一个研究议程，其中人工智能加强了对媒介的研究、保护和影响。
更新时间: 2026-10-09 16:49:29
领域: cs.AI,cs.CL,cs.CY
RoboMME-Interference: Benchmarking Robot Memory Under Interference
Robots deployed in realistic settings will accumulate experience across many sessions and tasks over their deployment. The robot's tasks may often require it to remember information from multiple sessions ago, making long-context robot memory important for real-world deployments. However, most robot-memory benchmarks today are based on single episodes or a short context. To measure how current robot memory systems perform on longer sessions with more distractions, we introduce RoboMME-Interference, a cross-session benchmark built on RoboMME (Dai et al., 2026). For each query episode, we construct a session history using the query's relevant prior demonstration followed by a controlled number of unrelated sessions, which we provide to the VLA as memory and measure accuracy. Running RoboMME's released memory-augmented $π_{0.5}$ variants unmodified through this benchmark, we find that while perceptual memory variants improve success when given the history without any distractors, they decay strongly and steadily as unrelated sessions accumulate. The subgoal variants, which read the history with a vision-language model and pass written subgoals to the policy, improve less at their best but hold more of that improvement as distractors accumulate. Adding a retrieval step to the strongest perceptual variant, which selects the section of history most visually similar to the robot's current view and passes only that section to the policy, restores its no-distractor success rate at every interference level. With this release, we emphasize the importance of long-context memory and robustness to interference and show that current systems largely fail on such capabilities. The project page, videos, code, and data are at https://robotmemorybench.com.
Updated: 2026-10-09 16:43:07
标题: RoboMME-Interference：在干扰下对机器人内存性能进行基准测试
摘要: 在现实环境中部署的机器人将在它们的部署过程中积累许多会话和任务的经验。机器人的任务通常需要它记住多个会话前的信息，因此长期上下文的机器人记忆对于实际部署至关重要。然而，今天大多数机器人记忆基准都是基于单个事件或短期上下文。为了衡量当前机器人记忆系统在更长时间的会话和更多干扰下的表现，我们引入了RoboMME-Interference，这是一个基于RoboMME（Dai等人，2026年）构建的跨会话基准。对于每个查询事件，我们使用查询相关的先前演示后跟一定数量的无关会话构建一个会话历史，将其提供给VLA作为记忆并测量准确性。在这个基准测试中，我们未经修改地运行RoboMME发布的增强记忆$π_{0.5}$变种，我们发现，尽管感知记忆变种在没有任何干扰的情况下提高了成功率，但它们在无关会话积累时会急剧而稳定地衰减。通过视觉-语言模型阅读历史并将书面子目标传递给策略的子目标变种在最佳状态下改善较少，但在干扰积累时保持了更多的改进。将最强的感知变种添加检索步骤，选择与机器人当前视图最相似的历史部分并仅将该部分传递给策略，可在每个干扰水平下恢复其无干扰成功率。通过这个发布，我们强调了长期上下文记忆和对干扰的稳健性的重要性，并展示了当前系统在这些能力上很大程度上失败。项目页面、视频、代码和数据位于https://robotmemorybench.com。
更新时间: 2026-10-09 16:43:07
领域: cs.RO,cs.AI,cs.LG
Specialists Learn Faster: Knowledge Organization in Pre-training Shapes Knowledge Update Efficiency
An ideal (language) model must continually update its knowledge. Existing work largely focuses on making these updates efficient after training. We investigate how the organization of time-varying knowledge during pre-training shapes the efficiency of subsequent updates. Models trained on chronologically ordered knowledge specialize in the most recently encountered knowledge; we call them specialists. In contrast, models trained on the same knowledge in randomly shuffled order perform well across all represented time periods; we call them generalists. Surprisingly, specialists require far fewer update steps than generalists to acquire new knowledge. To explain this advantage, we derive a law from first principles that relates update cost to the model's optimization dynamics and that closely predicts empirical results. Guided by these findings, we design a strategy for organizing pre-training data that preserves stable historical knowledge while enabling efficient updates to current knowledge. We also show that the benefits of temporal knowledge organization extend to real-world pre-trained models and existing knowledge update tools, and generalize to other knowledge orderings. Project page: https://horizon-lab.org/specialist/
Updated: 2026-10-09 16:42:33
标题: 专家学得更快：预训练中的知识组织塑造了知识更新效率
摘要: 一种理想的（语言）模型必须不断更新其知识。现有的工作主要集中在如何在训练后使这些更新变得有效率。我们研究了在预训练期间时间变化的知识组织如何影响后续更新的效率。按时间顺序训练的模型专门研究最近遇到的知识；我们称之为专家。相反，按随机顺序训练相同知识的模型在所有代表的时间段表现良好；我们称之为通才。令人惊讶的是，专家比通才需要更少的更新步骤来获取新知识。为了解释这一优势，我们从第一原则推导出一条关于更新成本与模型优化动态相关的定律，并且该定律与实证结果密切相关。受到这些发现的指导，我们设计了一种组织预训练数据的策略，既保留了稳定的历史知识，又使得对当前知识的更新更加高效。我们还展示了时序知识组织的好处延伸到现实世界的预训练模型和现有的知识更新工具，并且泛化到其他知识排序方式。项目页面：https://horizon-lab.org/specialist/
更新时间: 2026-10-09 16:42:33
领域: cs.LG
On the Design Space of Continuous Diffusion Language Models
Continuous diffusion language models (CDLMs) promise to bring the benefits of continuous flow modelling to language, including accelerated sampling and controllable generation. The field has grown rapidly, and the capabilities of CDLMs have begun to rival those of discrete diffusion. Yet, unlike for continuous modalities such as images or videos, our grasp of the levers that matter when training CDLMs remains limited. In this work, we focus on two of these levers, namely self-conditioning and embedding geometry, and shed light on some of the mechanisms behind them, which in turn suggest cheaper and more effective designs. We begin by showing that self-conditioning corrects a train/inference mismatch by exposing the model to inference-like states that are otherwise rare during training. These gains can be recovered without an extra forward pass by a two-level interpolant, coined $\mathrm{DoubleCat}$, which improves TinyGSM accuracy more than threefold. Self-conditioning also aids in recognising tokens dominated by large-norm Gaussians$\unicode{x2013}$an issue known as token identifiability. To address it directly, we introduce discrete reveals to $\mathrm{DoubleCat}$, yielding $\mathrm{SemiCat}$, which achieves discrete diffusion-level performance on TinyGSM, reaching $49.94\%$ with standard training hyperparameters and a tuned sampler. At the $1.7$B scale, $\mathrm{SemiCat}$ outperforms discrete diffusion on four of six likelihood-based evaluations, and surpasses one-hot flows on unconditional generation. On embedded spaces, we find that cross-entropy drives embeddings towards a geometry whose decoding error rate closely matches the one-hot one. We briefly explore a range of further options, such as hybrid corruption processes and noise schedules, including asynchronous ones. We believe this exploration can help the community better understand and advance the design space of CDLMs.
Updated: 2026-10-09 16:41:32
标题: 关于连续扩散语言模型的设计空间
摘要: 连续扩散语言模型（CDLMs）承诺将连续流建模的好处带入语言领域，包括加速采样和可控生成。该领域发展迅速，CDLMs的能力已经开始与离散扩散相媲美。然而，与图像或视频等连续模态不同，我们对训练CDLMs时重要的杠杆的理解仍然有限。在这项工作中，我们关注其中的两个杠杆，即自我调节和嵌入几何，揭示了它们背后的一些机制，从而提出了更便宜和更有效的设计。我们首先展示，自我调节通过让模型暴露于训练过程中罕见的推理状态，纠正了训练/推理不匹配。通过一个被称为$\mathrm{DoubleCat}$的两级插值器，可以在不进行额外前向传递的情况下恢复这些收益，该插值器将TinyGSM的准确性提高了三倍以上。自我调节还有助于识别受大范数高斯主导的令牌$\unicode{x2013}$这是一个被称为令牌可识别性的问题。为了直接解决这个问题，我们在$\mathrm{DoubleCat}$中引入了离散透露，产生了$\mathrm{SemiCat}$，在TinyGSM上实现了离散扩散级别的性能，使用标准训练超参数和调整的采样器，达到了$49.94\%$。在$1.7$B规模上，$\mathrm{SemiCat}$在六个基于似然的评估中超过了离散扩散四个，并在无条件生成上超过了单热流。在嵌入空间中，我们发现交叉熵驱动嵌入向一个几何形状靠近，其解码错误率与单热错误率密切匹配。我们简要探讨了一系列进一步的选择，如混合损坏过程和噪声时间表，包括异步的时间表。我们相信这种探索可以帮助社区更好地理解和推进CDLMs的设计空间。
更新时间: 2026-10-09 16:41:32
领域: cs.LG
SpatialASM: Internalizing Active Mentalization for Spatial Reasoning
When solving complex spatial reasoning tasks, ordinary humans naturally revisit the scenes from task-relevant viewpoints and mentally reconstruct the spatial layouts from those viewpoints. But can Large Vision-Language Models (LVLMs) proactively think in the same way? We term this ability Active Spatial Mentalization (ASM), an intrinsic component of visual intelligence that remains substantially underdeveloped in current LVLMs. To bridge this gap, we introduce SpatialASM, a scalable framework to internalize ASM into LVLMs through training-time spatial simulation. SpatialASM employs an allocentric novel-view generator to ground hypothetical viewpoints in visual evidence and produces two complementary forms of supervision: i) explicit ASM trajectories that teach LVLMs when and how to adaptively invoke view planning and imagination during spatial reasoning, and ii) automatically-constructed perspective-taking tasks that further strengthens viewpoint-conditioned scene imagination. On seven benchmarks spanning relatively perception-oriented spatial understanding to more challenging tasks requiring multi-image, cross-view and viewpoint-transformation reasoning, SpatialASM yields performance gain on LVLMs of four LVLM backbones in spatial reasoning while retaining their original general vision capabilities. The experimental results demonstrate that our approach improves average accuracy across the seven spatial benchmarks. And the synthesis of ASM trajectories provides a scalable pathway to alleviate the scarcity of spatial Chain-of-Thought (CoT) supervision for eliciting active viewpoint exploration and mental scene simulation.
Updated: 2026-10-09 16:39:50
标题: SpatialASM：内化空间推理的主动心智化
摘要: 在解决复杂的空间推理任务时，普通人自然会从与任务相关的视角重新审视场景，并从这些视角中心理重建空间布局。但是，大型视觉语言模型（LVLMs）是否可以主动以相同的方式思考？我们将这种能力称为主动空间心理化（ASM），这是视觉智能的一个内在组成部分，在当前的LVLMs中仍然存在显著的不足。为了弥合这一差距，我们引入了SpatialASM，这是一个可扩展的框架，通过训练时的空间模拟将ASM内化到LVLMs中。SpatialASM采用一种allocentric新视角生成器，将假设视角与视觉证据联系起来，并产生两种互补形式的监督：i）明确的ASM轨迹，教导LVLMs何时以及如何在空间推理过程中适应性地调用视角规划和想象力；ii）自动构建的视角取向任务，进一步加强了视角条件的场景想象。在涵盖相对感知导向的空间理解到需要多图像、跨视图和视角转换推理的更具挑战性任务的七个基准测试中，SpatialASM在空间推理中为四个LVLM骨干提供了性能增益，同时保留其原始的通用视觉能力。实验结果表明，我们的方法提高了七个空间基准测试的平均准确率。ASM轨迹的合成提供了一条可扩展的路径，以缓解空间推理的空间思维链（CoT）监督的稀缺性，促进主动视角探索和心理场景模拟。
更新时间: 2026-10-09 16:39:50
领域: cs.CV,cs.AI
MC-EvoBench: Auditing Experience-Driven Self-Improvement in Minecraft Agents
A central challenge in embodied recursive self-improvement is enabling agents to convert interaction history into capabilities that persist beyond one world and transfer to later tasks. A saved memory or successful final action does not by itself establish that a reusable capability was acquired; evaluation must separate capability change from carried-over resources, unequal source exposure, and early execution failures. We construct MC-EvoBench as a controlled Minecraft benchmark with a fixed catalog of 19 registered evolution chains, including 12 primary evaluation chains and seven calibration or probe chains, together with three graph-aligned paired probe families for transfer. The benchmark represents each task as a task-state graph, defines stage-level environment verifiers, assigns matched source and target roles, resets the target world, and permits only a declared agent artifact to cross the source-target boundary. For each probe, no-prior, related, unrelated, and shuffled source conditions are evaluated under matched environment-step budgets and five seeds; we report TSR, WGP, and VMP and retain provenance records. The benchmark reveals a consistent direct-capability gap between controller and planner-backed variants, while related experience yields no stable transfer: it does not improve either tool probe, its small wood-species gain hinges on one seed, and all intervals include zero. Stage audits further show that early wood-collection and crafting bottlenecks often leave later capabilities unobserved. MC-EvoBench thus offers a controlled testbed for durable, transferable embodied capabilities. Code and data are available at https://github.com/xzw-ustc/mc-bench-open-source
Updated: 2026-10-09 16:37:39
标题: MC-EvoBench：在Minecraft代理中审计经验驱动的自我改进
摘要: 在具体化递归自我改进过程中的一个核心挑战是使代理能够将交互历史转化为超越一个世界并转移到后续任务的能力。保存的记忆或成功的最终动作本身并不能确立已获得可重复使用能力；评估必须将能力变化与携带的资源、不平等的源曝光和早期执行失败区分开来。我们构建了MC-EvoBench作为一个带有固定目录的受控Minecraft基准测试，其中包括19个注册进化链，包括12个主要评估链和七个校准或探针链，以及三个图对齐的配对探针家族用于转移。该基准测试将每个任务表示为一个任务状态图，定义了阶段级环境验证器，分配了匹配的源和目标角色，重置了目标世界，并且仅允许声明的代理工件跨越源-目标界线。对于每个探针，在匹配的环境-步骤预算和五个种子下评估无先验、相关、不相关和洗牌源条件；我们报告TSR、WGP和VMP并保留来源记录。该基准测试揭示了控制器支持和规划支持变体之间的一致的直接能力差距，而相关经验并不产生稳定的转移：它不会改善任何工具探针，其小型木种获益取决于一个种子，所有区间都包括零。阶段审计进一步显示早期的木材收集和制作瓶颈通常会导致后来的能力无法观察到。因此，MC-EvoBench为持久的、可转移的具体化能力提供了一个受控测试平台。代码和数据可在https://github.com/xzw-ustc/mc-bench-open-source获取。
更新时间: 2026-10-09 16:37:39
领域: cs.RO,cs.AI
An introduction to local differential privacy protocols using block designs
The design of protocols for local differential privacy (or LDP) has been a topic of considerable research interest in recent years. LDP protocols utilise the randomised encoding of outcomes of an experiment using a transition probability matrix (TPM). Several authors have observed that balanced incomplete block designs (BIBDs) provide nice examples of TPMs for LDP protocols. Indeed, it has been shown that such BIBD-based LDP protocols provide optimal estimators. In this primarily expository paper, we give a detailed introduction to LDP protocols and their connections with block designs. We prove that a subclass of LDP protocols known as pure LDP protocols are equivalent to $(r,λ)$-designs (which contain balanced incomplete block designs as a special case). An unbiased estimator for an LDP scheme is a left inverse of the transition probability matrix. We show that the optimal estimators for BIBD-based TPMs are precisely those obtained from the Moore-Penrose inverse of the corresponding TPM. We also review some existing work on optimal LDP protocols in the context of pure protocols.
Updated: 2026-10-09 16:35:53
标题: 使用分块设计介绍本地差分隐私协议
摘要: 近年来，本地差分隐私（或LDP）协议的设计已成为一个相当受关注的研究课题。LDP协议利用随机编码实验结果的转移概率矩阵（TPM）。一些作者观察到平衡不完全区组设计（BIBD）为LDP协议提供了很好的TPM示例。事实上，已经证明基于BIBD的LDP协议提供了最佳估计量。 在这篇主要是阐释性的论文中，我们详细介绍了LDP协议及其与区组设计的联系。我们证明了LDP协议的一个子类，即纯LDP协议等同于$(r,λ)$-设计（其中平衡不完全区组设计是一个特例）。一个LDP方案的无偏估计器是转移概率矩阵的左逆。我们表明基于BIBD的TPM的最佳估计器恰好是通过相应TPM的Moore-Penrose逆获得的。我们还回顾了一些关于纯协议背景下的最佳LDP协议的现有工作。
更新时间: 2026-10-09 16:35:53
领域: math.CO,cs.CR
Same Query, Different Answer: Run-to-Run Variance in Semantic Query Processing
Semantic query engines ask a large language model (LLM) to judge individual rows, and benchmarks compare these engines on the quality of a few executions at temperature 0. We measure how much another execution changes. We re-execute the text and tabular workloads of SemBench, LRO-Bench, and UDA-Bench through their own harnesses, log every one of 1.2 million model calls, and reanalyze SemBench's published executions. At temperature 0, between 0.07% and 10.2% of rows change their operator output over five executions, depending on the model, and up to 24.0% on one task. Plans amplify these changes exponentially in the number of rows an answer exposes, as a possible-worlds model predicts: on UDA-Bench, a few percent of extracted values change, but about two thirds of the answers do. In SemBench's published results, 21% of text results change quality between executions, including on engines whose plans are deterministic. Several per-query system comparisons, and 16 of 28 tests of LRO-Bench's implementation recommendations, are not settled by the executions the benchmarks run. Changing rows are rows on which models disagree with each other; identical requests within one execution disagree as often as across executions; and one execution's decision margin or reasoning length predicts which rows change. Voting reduces the variance but does not remove it, and a calibrated cascade keeps its guarantee while the rows it returns change. We derive reporting rules for benchmarks and design rules for engines.
Updated: 2026-10-09 16:34:13
标题: 相同查询，不同答案：语义查询处理中的运行到运行差异
摘要: 语义查询引擎要求一个大型语言模型（LLM）来评判单个行，并且基准测试比较这些引擎在温度为0时的几次执行质量。我们衡量另一次执行的变化程度。我们重新执行SemBench、LRO-Bench和UDA-Bench的文本和表格工作负载，通过它们自己的测试框架记录120万个模型调用，并重新分析SemBench的公开执行。在温度为0时，根据模型不同，5次执行中有0.07%至10.2%的行更改其操作符输出，其中一个任务最高达24.0%。计划在回答暴露的行数上呈指数级增长这些变化，正如一个可能世界模型所预测的：在UDA-Bench上，一小部分提取值会发生变化，但约三分之二的答案会发生变化。在SemBench的公开结果中，21%的文本结果在执行之间质量有所改变，包括计划是确定性的引擎。几个每查询系统比较，以及LRO-Bench实施建议的28项测试中的16项，并未由基准测试运行的执行解决。更改的行是模型之间存在分歧的行；在一个执行中相同的请求与跨执行同样经常产生分歧；一个执行的决策边际或推理长度预测哪些行会发生变化。投票减少了方差，但并未完全消除，一个校准的级联保持其担保，同时返回的行发生变化。我们提出了基准测试的报告规则和引擎的设计规则。
更新时间: 2026-10-09 16:34:13
领域: cs.DB,cs.AI
Shared Canonical Component Removal Beyond Expert Views
In clinical practice, experts often begin by converting raw inputs into a set of expert features before making a decision. While black-box models trained on raw input can outperform models trained on expert views, it remains unclear what predictive signal the raw model exploits beyond what the expert view captures. We introduce an audit framework to quantify this gap using Shared Canonical Component Removal (SCCR). Unlike standard residualization or concept erasure, SCCR selectively removes only those directions strongly shared between learned raw and expert embeddings. Evaluations on synthetic data, coronary artery disease (CAD) detection, antidepressant treatment dropout prediction, and skin lesion classification using SkinCon annotations show that SCCR identifies the ways in which expert views are insufficient and recovers residual predictive signal while minimizing redundancy with the expert view.
Updated: 2026-10-09 16:32:12
标题: 超越专家观点的共享规范组件去除
摘要: 在临床实践中，专家们经常在做出决策之前将原始输入转换为一组专家特征。尽管基于原始输入训练的黑匣模型可能优于基于专家视图训练的模型，但依然不清楚原始模型利用了专家视图未捕获的预测信号。我们引入了一个审计框架来量化这种差距，使用共享标准组件去除（SCCR）。与标准的残差化或概念消除不同，SCCR选择性地仅去除在学习的原始和专家嵌入之间强烈共享的方向。 对合成数据、冠状动脉疾病（CAD）检测、抗抑郁药治疗退出预测以及使用SkinCon注释进行皮肤病变分类的评估表明，SCCR识别了专家视图不足之处，并在最小化与专家视图的冗余的同时恢复了残留的预测信号。
更新时间: 2026-10-09 16:32:12
领域: cs.LG
Beyond State-Wise Mirror Descent: Offline Policy Optimization with Parametric Policies
We investigate the theoretical aspects of offline reinforcement learning (RL) under general function approximation. While prior works (e.g., Xie et al., 2021) have established the theoretical foundations of learning a good policy from offline data via pessimism, existing algorithms that are computationally tractable (often in an oracle-efficient sense), such as PSPI, only apply to finite and small action spaces. Moreover, these algorithms rely on state-wise mirror descent and require actors to be implicitly induced from the critic functions, failing to accommodate standalone policy parameterization which is ubiquitous in practice. In this work, we address these limitations and extend the theoretical guarantees to parameterized policy classes over large or continuous action spaces. When extending mirror descent to parameterized policies, we identify contextual coupling as the core difficulty, and show how connecting mirror descent to natural policy gradient leads to novel analyses, guarantees, and algorithmic insights, including a surprising unification between offline RL and imitation learning.
Updated: 2026-10-09 16:28:54
标题: 超越状态逐步镜像下降：具有参数化策略的离线策略优化
摘要: 我们研究了在一般函数逼近下离线强化学习（RL）的理论方面。尽管先前的研究（例如Xie等人，2021年）已经建立了通过悲观主义从离线数据中学习良好策略的理论基础，但现有的算法（通常在有关Oracle的意义上是可计算的），如PSPI，仅适用于有限且小的动作空间。此外，这些算法依赖于状态智能镜像下降，并且需要从评论功能隐式诱导演员，无法适应在实践中普遍存在的独立政策参数化。在这项工作中，我们解决了这些限制，并将理论保证扩展到大型或连续动作空间上的参数化策略类。在将镜像下降扩展到参数化策略时，我们确定了上下文耦合作为核心难题，并展示了如何将镜像下降与自然政策梯度相连接，从而导致新的分析，保证和算法洞察，包括离线RL和模仿学习之间的令人惊讶的统一。
更新时间: 2026-10-09 16:28:54
领域: cs.LG,cs.AI
TabMF: Learning Dataset Representations with Tabular Foundation Models
Representations of supervised datasets support higher-level machine-learning tasks such as dataset retrieval and distribution-shift monitoring. However, existing approaches typically rely either on hand-crafted meta-features or on dataset encoders trained specifically for representation learning. We introduce TabMF, a framework that repurposes pretrained tabular foundation models (TFMs) as encoders of supervised datasets. TabMF equips a predictive TFM with an architecture-aware dataset-level readout and adapts the model end-to-end using contrastive post-training on stochastic views of synthetic supervised datasets. We instantiate TabMF on TabICLv2 and TabPFN-2.5 and evaluate the resulting representations across six tasks on unseen real datasets. Simple frozen TFM readouts yield competitive representations, indicating that predictive pretraining already induces substantial dataset-level structure. The two TabMF encoders obtain the strongest aggregate performance among the evaluated learned and classical dataset representations, with particularly strong results in dataset retrieval, compact benchmark construction, and distribution-shift severity assessment.
Updated: 2026-10-09 16:28:45
标题: TabMF：使用表格基础模型学习数据集表示
摘要: 监督数据集的表示支持更高级别的机器学习任务，如数据集检索和分布转移监控。然而，现有方法通常依赖于手工制作的元特征或专门用于表示学习的数据集编码器。我们介绍了TabMF，这是一个利用预训练的表格基础模型（TFMs）作为监督数据集编码器的框架。TabMF使用具有体系结构感知的数据集级读取的预测TFM，并使用对比后训练在合成监督数据集的随机视图上端到端地调整模型。我们在TabICLv2和TabPFN-2.5上实例化了TabMF，并在未见的真实数据集上评估了生成的表示。简单的冻结TFM读取产生竞争性表示，表明预测性预训练已经引入了相当大的数据集级结构。两个TabMF编码器在评估的学习和经典数据集表示中获得了最强的总体性能，在数据集检索、紧凑基准构建和分布转移严重性评估方面表现尤为突出。
更新时间: 2026-10-09 16:28:45
领域: cs.LG
Dream to Remember: Continual Steering of Generalist Robot Policies with World Models
Generalist robot policies are trained with imitation learning and are bounded by the quality and coverage of their demonstrations. Reinforcement learning can improve and extend them, either by fine-tuning the policy weights or by training a small module to steer the input noise of a frozen generative policy. When tasks arrive sequentially, both approaches suffer from catastrophic forgetting. Fine-tuning on a task degrades prior tasks, and steering only shifts forgetting to the shared steering module. We propose Dream to Remember (D2R), which continually steers a frozen generalist policy toward new tasks, while retaining its earlier skills. As the steering module learns a new task, a frozen pretrained world model rehearses prior tasks through imagined rollouts, anchoring the module to its earlier behavior without a replay buffer or new data. When the world model is faithful enough, D2R learns the new task entirely in imagination. Otherwise, it learns the task in the environment while using the world model only to rehearse prior tasks. On CALVIN, D2R improves eight tasks in sequence entirely in imagination, raising average success from 63.7% to 74.8%, while LoRA fine-tuning on the same sequence drops it to 11.7%. With a frozen off-the-shelf generalist policy and world model, $π_0$ and A2World, D2R learns unseen LIBERO-90 tasks in the environment while keeping $π_0$'s performance on the LIBERO suites it was fine-tuned on, even though the world model is not accurate enough for policy optimization. Finally, on a real Franka robot, D2R improves three tasks in sequence entirely in imagination while retaining the earlier improvements.
Updated: 2026-10-09 16:28:01
标题: 记忆中的梦想：通过世界模型对通用机器人策略进行持续引导
摘要: 广义机器人策略是通过模仿学习训练的，并受到演示的质量和覆盖范围的限制。强化学习可以改进和扩展这些策略，可以通过微调策略权重或训练一个小模块来引导冻结生成策略的输入噪声来实现。当任务依次到达时，这两种方法都会遭受灾难性遗忘的影响。在一个任务上的微调会降低先前任务的性能，而仅进行引导会将遗忘转移到共享的引导模块上。我们提出了Dream to Remember (D2R)，它不断地引导一个冻结的广义策略朝向新任务，同时保留其先前的技能。当引导模块学习一个新任务时，一个冻结的预训练世界模型通过想象中的实验来排练先前的任务，将模块锚定到其先前的行为，而无需重放缓冲区或新数据。当世界模型足够忠实时，D2R会完全在想象中学习新任务。否则，它会在环境中学习任务，同时仅使用世界模型来排练先前的任务。在CALVIN上，D2R完全在想象中改进了八个任务的顺序，将平均成功率从63.7%提高到74.8%，而在相同顺序上进行的LoRA微调使成功率降至11.7%。通过冻结的现成的广义策略和世界模型π0和A2World，D2R在环境中学习了未见过的LIBERO-90任务，同时保持了π0在其微调的LIBERO套件上的表现，即使世界模型不够准确进行策略优化。最后，在一个真实的Franka机器人上，D2R完全在想象中改进了三个任务的顺序，同时保留了先前的改进。
更新时间: 2026-10-09 16:28:01
领域: cs.RO,cs.CV,cs.LG
Privacy-Preserving Intent Fulfilment and Assurance for 6G RAN
Intent-based network management automates 6G service lifecycles, but existing fulfilment and assurance approaches require traffic inspection, per-flow tracking, or vendor-internal telemetry. These requirements conflict with privacy constraints in multi-tenant networks and vendor opacity in O-RAN deployments. We present an architecture that provisions resources from declared intent categories without traffic inspection and verifies fulfilment using only cell-level standardised PM counters at the O1 interface. A data-processing inequality bounds resource-allocation leakage about traffic content by $\log_2 K$ bits for $K$ categories. Intent-traffic unlinkability and node-opaque verification hold by construction. If verification requires at least $q$ cells per category among $M$ cells, the ceiling becomes $\log_2 \lfloor M/q \rfloor$ bits. This bound covers the category/resource channel, not total inference from PM counters. When individual records add state-independent noise to the category aggregate, the category-mean verifier is at least as powerful as any per-agent verifier. We map the architecture to the 3GPP intent lifecycle and O-RAN Non-RT RIC. On production PM data from four operator networks, the category-mean score has higher AUC on three networks and comparable AUC on the fourth under category-wide replacement, while per-cell screening suits sparse faults. These injection-based experiments do not establish the theorem's observation premise. Within-period cell co-movement limits a category mean to the evidence of 6 to 14 independent cells, and category-level verifiers need per-category calibration.
Updated: 2026-10-09 16:27:20
标题: 隐私保护意图实现和保证对于6G无线接入网
摘要: 基于意图的网络管理自动化了6G服务生命周期，但现有的实现和保证方法需要流量检查、流追踪或供应商内部遥测。这些要求与多租户网络中的隐私约束和O-RAN部署中的供应商不透明度相冲突。我们提出了一种架构，可以从声明的意图类别中配置资源，而无需流量检查，并且仅使用O1接口上的单元级标准化PM计数器对实现进行验证。数据处理不等式通过$\log_2 K$位来限制关于流量内容的资源分配泄漏，其中$K$为类别数。意图流量的不可连接性和节点不透明验证通过构建实现。如果验证在每个类别的$M$个单元中至少需要$q$个单元，则上限变为$\log_2 \lfloor M/q \rfloor$位。这个界限涵盖了类别/资源通道，而不是从PM计数器中的总推断。当个体记录向类别总数添加与状态无关的噪声时，类别均值验证器至少与任何基于代理的验证器一样强大。我们将这个架构映射到3GPP意图生命周期和O-RAN非实时RIC。在四个运营商网络的生产PM数据上，类别均值分数在三个网络上具有更高的AUC，并在第四个网络上具有可比的AUC，在类别范围内进行替换，而每个单元筛选适用于稀疏故障。这些基于注入的实验并不建立定理的观察前提。周期内单元的共同移动限制了一个类别均值至6到14个独立单元的证据，类别级验证器需要进行每个类别的校准。
更新时间: 2026-10-09 16:27:20
领域: cs.CR,cs.NI
Spanning Diffusion and Flow-based Posterior Sampling via Exact Gaussian Conditioning
Flow-matching and diffusion models are increasingly used as priors for linear inverse problems, however, most methods are bespoke to one class of trajectory prior. Plug-and-play methods have been proposed, however these implicitly decorrelate the guided trajectory, resulting in broad priors that can struggle when data is sparse. This work proposes $β$-Flow: a general plug-and-play method indexed by a correlation parameter $β$ that generalises over flow, diffusion, and decorrelated priors, and a confidence parameter $κ$ that controls how strongly the path distribution is enforced. Each step requires only one forward evaluation of a denoising network, followed by a Gaussian conditioning step. This step is an exact Kalman update that is not dependent on network gradients. We show that many existing guided samplers can be interpreted as instances or approximations of our method, each implicitly using a different trajectory prior and operator in place of the Kalman gain. From this viewpoint, we provide a qualitative comparison of methods in terms of their implicit biases and assumptions. We demonstrate consistently state-of-the-art level performance across 7 inverse problems over 4 datasets in terms of PSNR and SSIM. Moreover, we demonstrate the utility of this prior control, by showing that confident high correlations (strict priors) work best when observations are less informative, while low correlation is preferred when observations are more informative, reflecting the complementary roles of the prior and likelihood in Bayesian inference.
Updated: 2026-10-09 16:24:23
标题: 通过精确高斯条件化跨越扩散和基于流的后验采样
摘要: 流匹配和扩散模型越来越被用作线性逆问题的先验，然而，大多数方法都是专门针对一类轨迹先验设计的。虽然已经提出了即插即用的方法，但这些方法隐含地使引导轨迹失相关，导致广泛的先验在数据稀疏时可能会遇到困难。本文提出了$β$-Flow：一种通用的即插即用方法，由相关参数$β$索引，该方法通用于流动、扩散和失相关的先验，并由控制路径分布强制性的置信参数$κ$。每个步骤只需要一次去噪网络的前向评估，然后是一个高斯调节步骤。这一步骤是一个精确的卡尔曼更新，不依赖于网络梯度。我们展示了许多现有的引导采样器可以被解释为我们方法的实例或近似，每个方法隐含地使用不同的轨迹先验和算子来代替卡尔曼增益。从这个视角出发，我们在隐含偏差和假设方面提供了方法的定性比较。我们在4个数据集上展示了7个逆问题的PSNR和SSIM方面的始终领先水平的性能。此外，我们展示了这种先验控制的实用性，通过显示在观测信息较少时，自信的高相关性（严格的先验）效果最好，而在观测信息更充分时，低相关性更受欢迎，反映了贝叶斯推断中先验和似然的互补作用。
更新时间: 2026-10-09 16:24:23
领域: cs.LG,cs.CV
QTestBench: A Test Framework for Evaluating LLM-Generated Quantum Code
Large language models (LLMs) have shown strong results in classical code generation and are increasingly used to write the unit tests that safeguard that code. Whether they can do the same for quantum software is less clear, because quantum programs are probabilistic, depend on numerical tolerances, and are judged by domain-specific checkers, so a well-formed test can still disagree with the benchmark's verdict. Existing quantum benchmarks measure whether a model produces a correct solution, but not whether its tests can distinguish correct implementations from faulty ones. We introduce QTestBench, a framework that evaluates LLM-generated quantum test suites against independent benchmark checkers, never treating the tests themselves as ground truth. For each task, QTestBench prompts models to produce syntactic, semantic, and behavioral tests, injects candidate solutions at runtime, and compares each test verdict with the checker's label under three validation cohorts, each answering a distinct diagnostic question. We benchmark five LLMs on eight PennyLane QHack challenges and ten Qiskit HumanEval tasks. Generated suites almost always execute, but discrimination lags behind: the best generator reaches a test quality of 0.71 on QHack and 0.34 on Qiskit HumanEval, so executability alone overstates test quality. Oracle-verified synthetic mutants stabilize bug-detection estimates when natural faulty solutions are scarce, as on three of ten Qiskit tasks that yield none. QTestBench offers a reproducible, cross-framework benchmark for the trustworthiness of LLM-generated quantum tests.
Updated: 2026-10-09 16:23:46
标题: QTestBench：用于评估LLM生成的量子代码的测试框架
摘要: 大型语言模型（LLMs）已经在经典代码生成方面展现出了强大的结果，并且越来越多地被用于编写保护代码的单元测试。它们是否能够在量子软件中做到同样的效果尚不清楚，因为量子程序是概率性的，依赖于数值容差，并且由领域特定的检查器评判，因此一个良好构建的测试仍然可能与基准的判定不符。现有的量子基准测试衡量一个模型是否产生了正确的解决方案，但并不衡量它的测试是否能够区分正确的实现和错误的实现。我们介绍了QTestBench，一个评估LLM生成的量子测试套件与独立基准检查器的框架，绝不将测试本身视为绝对真理。对于每个任务，QTestBench促使模型生成语法、语义和行为测试，运行时注入候选解决方案，并在三个验证组中比较每个测试的判决与检查器的标签，每个组回答一个不同的诊断问题。我们在八个PennyLane QHack挑战和十个Qiskit HumanEval任务上对五个LLMs进行了基准测试。生成的测试套件几乎总是可以执行，但辨别力滞后：最好的生成器在QHack上达到了0.71的测试质量，在Qiskit HumanEval上达到了0.34，因此仅仅可执行性可能会夸大测试质量。在自然错误解决方案稀缺的情况下，经过Oracle验证的合成突变体可以稳定bug检测估计，例如在十个Qiskit任务中的三个不产生错误解决方案的情况下。QTestBench为LLM生成的量子测试的可信度提供了一个可重复的、跨框架的基准测试。
更新时间: 2026-10-09 16:23:46
领域: cs.LG,cs.CL,quant-ph
MPMMine: Better Benchmarks for Constraint Acquisition
Constraint Acquisition (CA) and related research on the validation and enhancement of Mathematical Programming (MP) models using domain knowledge artifacts are currently limited by inadequate benchmarks. The existing benchmarks are loosely organized, provide inconsistent types and formats of artifacts for individual problems, are designed for benchmarking solvers rather than CA algorithms, include narrow classes of problems, or omit crucial details for the reproducibility of experiments and fair comparison of CA methods. We develop a benchmark dataset, a supporting software package, and a benchmarking tool crafted for the consistent, systematic, and comprehensive assessment of CA algorithms, bridging the gap in algorithmic development and experimental infrastructure. This work presents MPMMine, a benchmark suite designed to assess algorithms that discover, validate, and enhance MP models using diverse artifacts. The development of MPMMine is guided by design rules supporting consistency, standardization, completeness, extensibility, openness, and version control. It adopts a uniform structure and relies on open formats. It provides alternative formulations of MP models per problem, tens of parameter instantiations per MP model, and thousands of exemplary solutions and non-solutions in discrete, continuous, and mixed domains, alongside natural-language descriptions to support text-to-model methods. The mpmmine-py Python package provides an application programming interface for accessing the dataset artifacts. The MPMMine Leaderboard is a~tool that systematically evaluates existing CA algorithms using MPMMine under consistent conditions. It leverages open source implementations, runs them in isolated containers adapting input/output to their needs to build an extensive, interactive scoreboard.
Updated: 2026-10-09 16:22:45
标题: MPMMine：更好的约束获取基准
摘要: Constraint Acquisition（CA）及相关研究目前受到不足的基准数据的限制，这些基准数据用于验证和增强使用领域知识工具的数学规划（MP）模型。现有的基准数据组织松散，为各个问题提供不一致的类型和格式的工具，设计用于基准测试求解器而不是CA算法，包括狭窄类别的问题，或者省略实验可重现性和CA方法公平比较的关键细节。 我们开发了一个基准数据集、一个支持软件包和一个专为一致、系统和全面评估CA算法而设计的基准工具，弥合了算法开发和实验基础设施之间的差距。 这项工作介绍了MPMMine，一个旨在使用各种工具评估发现、验证和增强MP模型的算法的基准套件。MPMMine的开发受设计规则的指导，支持一致性、标准化、完整性、可扩展性、开放性和版本控制。它采用统一结构，并依赖于开放格式。它为每个问题提供MP模型的替代公式，每个MP模型提供数十个参数示例化，以及在离散、连续和混合领域中数千个示例解和非解，同时提供自然语言描述以支持文本到模型方法。mpmmine-py Python软件包提供了一个应用程序编程接口，用于访问数据集工具。MPMMine排行榜是一个工具，通过一致条件下使用MPMMine系统评估现有的CA算法。它利用开源实现，在隔离容器中运行它们，适应它们的输入/输出需求以构建一个广泛的、互动的计分板。
更新时间: 2026-10-09 16:22:45
领域: cs.AI,cs.CE
Diversity-Aware Ensemble of Quantum-Classical Models via Combinatorial Fusion
Variational quantum circuits (VQCs) are highly sensitive to circuit architecture, yet most quantum machine learning pipelines select a single best design and discard the remaining candidates. This paper addresses whether the structural diversity across quantum circuit search spaces can be used to produce stronger predictions through model fusion. We apply Combinatorial Fusion Analysis (CFA), a model fusion framework that selects and combines multiple scoring systems using rank-score function (RSF) and cognitive diversity (CD), to a pool of 29 architecturally diverse classical models and quantum circuits. Using a diversity-aware model selection method, we identify subsets of maximally diverse models and combine them with both score and rank combination. Experiments on four binary classification benchmarks covering different domains demonstrate that CFA-based hybrid classical-quantum fusion consistently outperform the best individual model on every dataset considered with improvements up to 6.03% in accuracy. When comparing with relevant literatures, the CFA framework further achieves superior performance across all four datasets.
Updated: 2026-10-09 16:20:02
标题: 多元意识集成量子-经典模型的组合融合
摘要: 变分量子电路（VQCs）对电路架构非常敏感，然而大多数量子机器学习流程选择一个最佳设计并丢弃其余候选方案。本文探讨了量子电路搜索空间中的结构多样性是否可以通过模型融合产生更强的预测能力。我们应用了组合融合分析（CFA），这是一个模型融合框架，使用等级分数函数（RSF）和认知多样性（CD）选择和结合多个评分系统，应用于一个包含29种结构多样的经典模型和量子电路的池中。通过一种考虑多样性的模型选择方法，我们识别出最大多样性模型的子集，并使用得分和等级组合将它们结合起来。在涵盖不同领域的四个二元分类基准上进行的实验表明，基于CFA的混合经典-量子融合在每个考虑的数据集上始终优于最佳单个模型，准确率提高高达6.03%。与相关文献比较时，CFA框架在所有四个数据集上进一步实现了更优异的性能。
更新时间: 2026-10-09 16:20:02
领域: quant-ph,cs.LG
The Spec Growth Engine: Spec-Anchored, Code-Coupled, Drift-Enforced Architecture for AI-Assisted Software Development
AI coding agents dramatically accelerate implementation speed but introduce two structural failure modes that existing spec-driven approaches do not fully solve: (1) context explosion -- the agent must reason over an entire repository at once, degrading output quality as the context window fills; and (2) silent spec-code drift -- code evolves, the specification does not, and the divergence becomes invisible until it is costly to repair. We present the Spec Growth Engine, a lightweight framework that addresses both failure modes through a machine-readable spec graph whose nodes carry explicit contract/design separation, a Spine context assembler that scopes agent context to an ownership path, a vertical-slice growth protocol that enforces hardest-first ordering, and a drift gate that makes spec-code divergence a blocking merge condition. The design synthesises well-established software engineering principles (Parnas information hiding, C4, ADRs, Walking Skeleton, Reflexion Models, Fitness Functions) into a lean, code-coupled, machine-enforced whole -- without the overhead of heavy-weight frameworks such as RUP or MDA.
Updated: 2026-10-09 16:19:41
标题: “Spec Growth Engine：基于规范锚定、代码耦合和漂移强制架构的AI辅助软件开发”
摘要: AI编码代理显著加快了实施速度，但引入了两种现有基于规范的方法未能完全解决的结构性故障模式：（1）上下文爆炸 - 代理必须一次性地对整个存储库进行推理，随着上下文窗口填满，输出质量会下降；以及（2）静默规范-代码漂移 - 代码发展，规范不变，分歧变得看不见，直到修复成本高昂为止。 我们提出了Spec Growth Engine，这是一个轻量级框架，通过一个机器可读的规范图来解决这两种故障模式，其中节点携带显式的合同/设计分离，Spine上下文组装器将代理上下文范围限制为所有权路径，一个垂直切片增长协议强制执行最难的优先顺序，以及一个漂移门，使规范-代码分歧成为阻碍合并的条件。 该设计将已经建立的软件工程原则（Parnas信息隐藏，C4，ADR，Walking Skeleton，反射模型，适应函数）综合到一个精简的、代码耦合的、机器强制执行的整体中 - 而无需像RUP或MDA这样的重量级框架的开销。
更新时间: 2026-10-09 16:19:41
领域: cs.SE,cs.AI
FORTE: A Production Framework for Feature Evaluation and Deprecation in Industrial Recommender Systems
Production recommender systems have feature-lifecycle asymmetry: adding new features must produce gains on some metric, while feature removal must not regress any metric. In practice, feature deprecation is hard to prioritize as a feature can fail to be removed if it interacts with other features, its importance changes daily, it cannot be evaluated before production models change or data drift occurs, it is in contention for A/B test space with other launches, it causes regressions in an online A/B test, or it is consumed by multiple models that need to coordinate its deprecation. We present FORTE, a model-agnostic production framework for discovering, validating, and removing features. Its efficient perturbation implementation allows daily feature evaluation. To account for interactions, it evaluates candidate feature sets constructed from all the low-rank features through grouped ablation and its offline results are sufficiently predictive of online A/B tests to allow deprecations to proceed to online testing with confidence. Lineage tracing identifies all downstream consumer models to realize the full savings of removing a feature from the training set. We validate FORTE over 2 large-scale recommender systems and find that it identifies deprecable feature sets which do not cause online regressions, while the impact on efficiency depends on model characteristics. On a search-relevance model, FORTE resulted in 3 nested candidate sets of 32, 48, and 59 features. Online guardrails rejected the largest treatment. The 48-feature set was selected for deprecation, removing 43.3% of the model's features. On a large ranking model, FORTE identified 8 candidates for removal. Retraining and A/B testing confirmed no regression. Its launch reduced the trainable parameters by 29.9%, improved serving throughput by 19.8-25.5%, and reduced annual training costs by hundreds of thousands of dollars.
Updated: 2026-10-09 16:18:39
标题: FORTE：工业推荐系统中特征评估和弃用的生产框架
摘要: 生产推荐系统存在特征生命周期的不对称性：添加新特征必须对某些指标产生增益，而删除特征则不得导致任何指标的退化。在实践中，特征废弃很难优先考虑，因为如果一个特征与其他特征互动、其重要性每天都在变化、在生产模型变化或数据漂移发生之前无法评估、与其他推出项目争夺A/B测试空间、在在线A/B测试中导致退化，或者被多个需要协调其废弃的模型所使用，那么该特征可能无法被移除。我们提出了FORTE，这是一个针对生产环境的模型无关框架，用于发现、验证和移除特征。其高效的扰动实现允许每日对特征进行评估。为了考虑互动，它通过分组消蚀评估由所有低秩特征构建的候选特征集，其离线结果足以预测在线A/B测试，从而确保废弃的进行在线测试时有信心。谱系追踪确定所有下游消费模型，以实现从训练集中移除特征的全部节省。我们在两个大规模的推荐系统上验证了FORTE，并发现它识别出可废弃的特征集，不会导致在线退化，而其对效率的影响取决于模型特征。在一个搜索相关性模型上，FORTE得出了3个嵌套的候选特征集，分别包含32、48和59个特征。在线防护栏拒绝了最大的处理。48个特征集被选定用于废弃，删除了模型43.3%的特征。在一个大型排名模型上，FORTE确定了8个可移除的候选特征。重新训练和A/B测试确认没有退化。其推出使可训练参数减少了29.9%，提高了服务吞吐量19.8-25.5%，并将年度培训成本降低数十万美元。
更新时间: 2026-10-09 16:18:39
领域: cs.LG,cs.AI
Constrained Probabilistic Numerics: Gaussian Process Solvers for Linear PDEs under Linear Inequality Constraints
Gaussian process solvers for linear partial differential equations (PDEs) return posteriors rather than point estimates, but they condition only on equality information: the PDE residual, boundary data, and measurements. Many solutions of interest also obey inequality constraints, including nonnegativity, monotonicity, and absence of arbitrage, which equality conditioning cannot enforce. We prove that conditioning a Gaussian process prior jointly on linear equality observations and a finite-dimensional inequality event is equivalent to conditioning sequentially in terms of Borel measures on the solution space. The physics-informed posterior is therefore a valid prior for the inequality update at virtual points. The combined law is non-Gaussian yet computable through truncated-Gaussian moments, and neither the PDE conditioning nor the inequality update requires modification. We evaluated the construction on a Poisson benchmark, option-price surfaces, and air-quality fields. Under incomplete boundary data, a monotonicity constraint lowered the error of a two-dimensional Poisson reconstruction by $63\%$. On a real S&P 500 option chain thinned to $20$ retained quotes, no-arbitrage constraints reduced held-out extrapolation error by $22\%$ and cut the calendar-arbitrage rate from $10\%$ to $1\%$. On four real air-quality snapshots, nonnegativity reduced a Gaussian moment-based sub-zero-tail proxy by $74\%$ at the sparsest monitor count. Changes in mean accuracy and constraint diagnostics differed across data regimes. The probability of the noisy virtual constraint event under the equality posterior measures agreement with the specified bounds before the update. The construction turns admissibility from a post-processing step into conditioning information for probabilistic PDE solvers.
Updated: 2026-10-09 16:18:11
标题: 受限概率数值方法：在线性不等式约束下解线性PDE的高斯过程求解器
摘要: 高斯过程求解器用于线性偏微分方程（PDE）返回后验而不是点估计，但它们仅在等式信息上进行条件化：PDE残差、边界数据和测量。许多感兴趣的解也遵守不等式约束，包括非负性、单调性和无套利性，等式条件化无法强制执行这些约束。我们证明，在高斯过程先验上同时对线性等式观测和有限维不等式事件进行条件化等价于根据解空间上的波雷尔测度进行顺序条件化。因此，基于物理的后验是不等式更新虚拟点的有效先验。组合法则是非高斯的，但通过截断高斯矩可计算，既不需要修改PDE条件化也不需要修改不等式更新。我们在泊松基准、期权价格曲面和空气质量场上评估了这种构造。在边界数据不完整的情况下，单调性约束将二维泊松重建的错误降低了63％。在一个实际的S＆P 500期权链上，稀疏到20个保留报价，无套利约束将持有的外推误差降低了22％，并将日历套利率从10％降低到1％。在四个实际的空气质量快照中，非负性将基于高斯矩的次零尾部代理在最稀疏的监测计数处降低了74％。均值准确性和约束诊断的变化在数据区域之间有所不同。在更新之前，基于等式后验的噪声虚拟约束事件的概率与指定的边界一致。这种构造将可接受性从后处理步骤转变为概率PDE求解器的条件信息。
更新时间: 2026-10-09 16:18:11
领域: stat.ML,cs.LG,math.NA
Adaptive Variance Reduction without Knowing Oracle Smoothness
Variance reduction improves the convergence rate of stochastic non-convex optimization under the average smoothness condition. Existing adaptive variance reduction methods are especially designed for average smoothness and do not adapt to standard smoothness. In this paper, we study how one algorithm can adapt to these two settings and the noiseless setting without knowing the smoothness constants or the variance bound. Given the gradient budget $T$, we develop a two-phase method. The first phase compares two independent gradients and switches to normalized momentum when their difference exceeds a threshold. This phase provides both a candidate with standard smooth and noiseless guarantees and also ensures a bounded expected function gap for the second variance reduction phase. The resulting rates are $O(T^{-1/3})$ under average smoothness, $O(T^{-1/4})$ under standard smoothness, and $O(T^{-1/2})$ without noise. When $T$ is unknown, we propose a method with two paths that attain the corresponding rates with logarithmic factors in the stochastic settings. Here, we use a variance reduction path to deal with average smoothness and employ the AdaGrad-Norm for standard smoothness and noiseless setting. Neither method restarts from the initial point or requires a growing number of oracle calls in each round.
Updated: 2026-10-09 16:13:42
标题: 不知道Oracle平滑度的自适应方差缩减
摘要: 方差减少改善了在平均光滑条件下随机非凸优化的收敛速度。现有的自适应方差减少方法特别设计用于平均光滑，不适应标准光滑。在本文中，我们研究了一个算法如何适应这两种设置和无噪声设置，而不知道光滑常数或方差上限。给定梯度预算$T$，我们开发了一种两阶段方法。第一阶段比较两个独立的梯度，并在它们的差异超过阈值时切换到归一化动量。这个阶段不仅提供了一个具有标准光滑和无噪声保证的候选者，还确保了第二个方差减少阶段的有界期望函数间隙。结果的速率在平均光滑条件下为$O(T^{-1/3})$，在标准光滑条件下为$O(T^{-1/4})$，在无噪声条件下为$O(T^{-1/2})$。当$T$未知时，我们提出了一种具有两个路径的方法，在随机设置中以对数因子达到相应速率。在这里，我们使用方差减少路径处理平均光滑，并在标准光滑和无噪声设置中使用AdaGrad-Norm。两种方法都不需要从初始点重新开始，也不需要在每一轮中调用不断增加的预言者数量。
更新时间: 2026-10-09 16:13:42
领域: math.OC,cs.LG
WoundAmbit: Bridging State-of-the-Art Semantic Segmentation and Real-World Wound Care
Chronic wounds affect a large population, particularly the elderly and diabetic patients, who often exhibit limited mobility and co-existing health conditions. Automated wound monitoring via mobile image capture can reduce in-person physician visits by enabling remote tracking of wound size. Semantic segmentation is key to this process, yet wound segmentation remains underrepresented in medical imaging research. To address this, we benchmark state-of-the-art deep learning models from general-purpose vision, medical imaging, and top methods from public wound challenges. For a fair comparison, we standardize training, data augmentation, and evaluation, conducting cross-validation to minimize partitioning bias. We also assess real-world deployment aspects, including generalization to an out-of-distribution wound dataset, computational efficiency, and interpretability. Additionally, we propose a reference object-based approach to convert AI-generated masks into clinically relevant wound size estimates and evaluate this, along with mask quality, for the five best architectures based on physician assessments. Overall, the transformer-based TransNeXt showed the highest levels of generalizability. Despite variations in inference times, all models processed at least one image per second on the CPU, which is deemed adequate for the intended application. Interpretability analysis typically revealed prominent activations in wound regions, emphasizing focus on clinically relevant features. Expert evaluation showed high mask approval for all analyzed models, with VWFormer and ConvNeXtS backbone performing the best. Size retrieval accuracy was similar across models, and predictions closely matched expert annotations. Finally, we demonstrate how our AI-driven wound size estimation framework, WoundAmbit, is integrated into a custom telehealth system.
Updated: 2026-10-09 16:07:10
标题: WoundAmbit：连接最先进的语义分割技术和现实世界的伤口护理
摘要: 慢性伤口影响着大量人群，尤其是老年人和糖尿病患者，他们通常表现出行动受限和伴随的健康状况。通过移动图像捕捉进行自动化伤口监测可以通过远程跟踪伤口大小来减少亲自就诊医生的次数。语义分割是这一过程的关键，然而伤口分割在医学影像研究中仍然未被充分重视。为了解决这个问题，我们对来自通用视觉、医学影像以及公开伤口挑战赛中的顶级方法的最新深度学习模型进行了基准测试。为了公平比较，我们标准化了训练、数据增强和评估，进行了交叉验证以最小化分区偏差。我们还评估了现实世界的部署方面，包括对分布外伤口数据集的泛化、计算效率和可解释性。此外，我们提出了一种基于参考对象的方法，将AI生成的掩模转换为临床相关的伤口大小估计，并基于医师评估对五种最佳架构进行了评估。总的来说，基于Transformer的TransNeXt显示出最高的泛化水平。尽管推理时间有所变化，但所有模型在CPU上每秒至少处理一张图像，这被认为对于预期应用是足够的。可解释性分析通常显示出在伤口区域的显著激活，强调对临床相关特征的关注。专家评估显示，所有分析模型的掩模批准率都很高，其中VWFormer和ConvNeXtS骨干表现最好。各模型的尺寸检索准确性相似，并且预测与专家注释非常接近。最后，我们展示了我们的AI驱动的伤口大小估算框架WoundAmbit如何集成到自定义远程医疗系统中。
更新时间: 2026-10-09 16:07:10
领域: cs.CV,cs.AI
Randomized Momentum Transport with Cheap Predictors
Momentum methods such as STORM and second-order momentum track the gradient by adding a transport correction to momentum, at the cost of an extra oracle call per iteration. We introduce randomized momentum transport, which evaluates the correction with probability $q$ and otherwise uses a cheap prediction, with importance weighting that preserves the conditional mean and adds $(q^{-1}-1)$ times the prediction loss to the variance. Under transport-bias, second-moment, and residual-control assumptions, linear minimization oracle (LMO) updates, including normalized, sign, and exact spectral (Muon) directions, retain an $O(T^{-1/3})$ stationarity bound for each fixed $q>0$ with a single-gradient initialization. For unnormalized STORM, some $q<1$ lowers the leading coefficient of our sufficient gradient budget if and only if the gradient-difference noise modulus exceeds the smoothness constant, by at most about $10\%$. An online diagonal predictor admits a regret-to-convergence guarantee. At approximately equal oracle cost, querying a quarter of the corrections lowers the final training loss by $18$ to $43\%$ on MNIST1D and raises the test accuracy of Hessian-vector transport on CIFAR-10 by $1.5$ points.
Updated: 2026-10-09 16:04:53
标题: 用廉价预测器实现随机动量传输
摘要: 动量方法，如STORM和二阶动量，通过在动量中添加运输校正来跟踪梯度，每次迭代需要额外的Oracle调用。我们引入了随机动量传输，它以概率$q$评估校正，否则使用廉价的预测，通过保持条件均值和向方差添加$(q^{-1}-1)$倍的预测损失的重要性加权。在传输偏差、二阶矩和残差控制假设下，线性最小化Oracle（LMO）更新，包括归一化、符号和精确谱（Muon）方向，在每个固定的$q>0$下保持$O(T^{-1/3})$的稳定性界，只需单梯度初始化。对于未归一化的STORM，如果且仅如果梯度差噪声模大于平滑常数，一些$q<1$会降低我们充分的梯度预算的主导系数，最多约为$10\%$。在线对角预测器可以保证收敛性遗憾。在大约相同的Oracle成本下，查询四分之一的校正可以将最终训练损失降低$18$至$43\%$，并将Hessian-向量传输在CIFAR-10上的测试准确率提高了$1.5$个点。
更新时间: 2026-10-09 16:04:53
领域: cs.LG,math.OC
Optimal Algorithms for Nonsmooth Nonconvex Stochastic Optimization with Constraints
We present new algorithms for stochastic optimization problems in which the objective is neither smooth nor convex, subject to a convex constraint set. Our algorithms return a $(δ,\varepsilon)$-Goldstein stationary point, as recently defined for constrained problems by Liu et al. (2024), with improved complexities, and are the first to achieve optimal convergence rates for this task in several settings. In particular, we design projection-based and projection-free algorithms that achieve a $\widetilde O(δ^{-1}\varepsilon^{-3})$ rate using stochastic gradients, which we show to be optimal, and a $\widetilde O(dδ^{-1}\varepsilon^{-3})$ rate using zeroth-order queries, which is also optimal in its dependence on the dimension $d$ and improves over prior work. Notably, our lower bound is given in terms of the Frank-Wolfe gap and holds even for smooth problems, which is of independent interest. These results show that in nonconvex stochastic optimization, the convergence rate is degraded neither by the presence of constraints nor by nonsmoothness. Our analysis is based on a reduction to online linear optimization, extending a related technique from unconstrained optimization.
Updated: 2026-10-09 16:04:13
标题: 具有约束的非光滑非凸随机优化的最优算法
摘要: 我们提出了一种新的算法，用于处理随机优化问题，其中目标既不是光滑的也不是凸的，但受到一个凸约束集的限制。我们的算法返回一个$(δ,\varepsilon)$-Goldstein稳定点，最近由Liu等人（2024年）在约束问题中定义，具有改进的复杂度，并且在几种情况下是首次实现这一任务的最佳收敛速率。具体来说，我们设计了基于投影和无投影的算法，使用随机梯度可以实现$\widetilde O(δ^{-1}\varepsilon^{-3})$的速率，我们证明这是最优的，使用零阶查询可以实现$\widetilde O(dδ^{-1}\varepsilon^{-3})$的速率，这在对维度$d$的依赖方面也是最优的，并且优于以前的工作。值得注意的是，我们的下界是以Frank-Wolfe间隙来表示的，即使对于光滑问题，这个下界也是成立的，这是具有独立意义的。这些结果表明，在非凸随机优化中，收敛速度既不会因为存在约束也不会因为不光滑而降低。我们的分析基于在线线性优化的简化，扩展了与无约束优化相关的技术。
更新时间: 2026-10-09 16:04:13
领域: math.OC,cs.LG
Your Discrete Graph Diffusion Model is Secretly Memorizing
Discrete graph diffusion models (GDMs) corrupt the graph structure by adding noise independently to nodes and edges under a Markovian noise model. In the unconditional setting without node features, an equivariant denoiser trained on a single graph is able to generate novel graphs and sample-level metrics such as edge overlap and WL kernel similarity find no evidence of memorization. In this work, we challenge this view and show for the first time that graph diffusion models indeed exhibit memorization. Analyzing the reverse diffusion trajectories of GDMs trained on two-block stochastic block models (SBMs), we find a strong train-data bias: the model recovers a corrupted training graph almost perfectly compared to a held-out test graph. We term this conditional memorization, as it surfaces only when the reverse trajectory is conditioned on a corrupted training graph and is not apparent during inference. It coincides with a growing generalization gap, where the training loss improves as the loss on held-out graphs degrades, and it cannot be mitigated by early stopping. The onset of memorization always precedes the emergence of community structure in the generated graphs, thus every checkpoint earlier produces structureless graphs. This behavior holds across all node sizes we test.
Updated: 2026-10-09 16:04:06
标题: 你的离散图扩散模型正在暗中记忆
摘要: 离散图扩散模型（GDMs）通过在马尔可夫噪声模型下独立地向节点和边添加噪声来破坏图结构。在无节点特征的无条件设置中，一个在单个图上训练的等变去噪器能够生成新的图，并且样本级指标如边重叠和WL核相似性没有记忆的证据。在这项工作中，我们挑战了这种观点，并首次展示了图扩散模型确实表现出记忆现象。通过分析在两块随机块模型（SBMs）上训练的GDMs的逆扩散轨迹，我们发现存在严重的训练数据偏差：与留出测试图相比，模型几乎完美地恢复了一个受损的训练图。我们将这种现象称为条件记忆，因为它只在逆扩散轨迹受损的训练图条件下出现，并且在推断过程中并不明显。它与日益增加的泛化差距相一致，即在留出图上的损失恶化时，训练损失却有所改善，并且无法通过提前停止进行缓解。记忆的开始总是在生成图中的社区结构出现之前，因此每个较早的检查点产生的图都是没有结构的。这种行为在我们测试的所有节点大小范围内都成立。
更新时间: 2026-10-09 16:04:06
领域: cs.LG,cs.AI
UkrMeme: Multimodal Explainable Persuasion Detection in Ukrainian Memes
Memes play a central role in modern online communication, especially in contexts of political conflict where visual and textual cues combine to convey persuasive messages. We study persuasive technique detection in Ukrainian memes using vision-language models (VLMs). We introduce UkrMeme, a dataset of 200 organically produced Ukrainian wartime memes annotated for persuasion techniques, and PTM-UA, a Ukrainian translation of a subset of the Propaganda 950 benchmark that preserves original visual content while translating meme text. We evaluate VLMs in a zero-shot prompting setup across image, text, and multimodal inputs using hierarchical F1 metrics. Finally, we compare Layer-wise Relevance Propagation (LRP) and model-generated token attributions and analyze the semantic categories of evidence emphasized by each explanation method.
Updated: 2026-10-09 16:03:27
标题: 乌克兰迷因：多模式可解释的说服检测
摘要: 梗在现代在线交流中发挥着核心作用，尤其是在政治冲突的背景下，视觉和文本提示相结合传达有说服力的信息。我们使用视觉-语言模型（VLMs）研究乌克兰梗中的说服技巧检测。我们介绍了UkrMeme数据集，其中包含200个有机制作的乌克兰战时梗，标注了说服技巧，以及PTM-UA，将Propaganda 950基准的子集翻译为乌克兰语，保留原始视觉内容并翻译梗文本。我们使用分层F1指标在图像、文本和多模态输入之间进行零-shot提示设置来评估VLMs。最后，我们比较了层级相关传播（LRP）和模型生成的标记属性，并分析了每种解释方法强调的证据语义类别。
更新时间: 2026-10-09 16:03:27
领域: cs.AI,cs.CL,cs.CV
LLMail-Inject: A Dataset from a Realistic Adaptive Prompt Injection Challenge
Indirect prompt injection attacks exploit a fundamental weakness of LLMs: executing unauthorized tasks from untrusted sources. Despite the practical significance of this attack vector, existing benchmarks rely on small sets of template-based attacks (100-1,000 examples) that fail to capture the diversity and adaptiveness of real-world adversaries. Consequently, benchmarks are saturated and overfit to static attacks. We introduce LLMail-Inject, a large-scale benchmark of 208,095 distinct human-written attack prompts collected through a public red-teaming challenge with 839 participants. LLMail-Inject evaluates the attack's complete chain of execution: prompts must be retrieved from an email database, evade production-level defenses, and trigger unauthorized tool calls with correct arguments. Evaluation reveals that models that achieve 5% attack success rate (ASR) on existing benchmarks fail more on LLMail-Inject: Meta SecAlign--specifically trained against prompt injection--still yields 32% ASR, GPT-4o exhibits 58% ASR, and GPT-5 mini 47% ASR. We further demonstrate that attacks transfer across domains, achieving 37% ASR on proof-of-concept coding tasks and 28-45% ASR on file management tasks. We release our challenge platform and curated evaluation subsets at multiple difficulty tiers and the full dataset with rich annotations.
Updated: 2026-10-09 16:01:12
标题: LLMail-Inject：来自真实适应性提示注入挑战的数据集
摘要: 间接提示注入攻击利用LLM的一个基本弱点：从不受信任的来源执行未经授权的任务。尽管这种攻击向量具有实际重要性，但现有基准依赖于小型模板攻击集（100-1,000个示例），无法捕捉现实世界对手的多样性和适应性。因此，基准已经饱和并过度拟合于静态攻击。我们引入了LLMail-Inject，一个由208,095个不同的人类编写的攻击提示组成的大规模基准，通过一个公开的红队挑战与839名参与者收集。LLMail-Inject评估了攻击的完整执行链：提示必须从电子邮件数据库中检索，规避生产级防御，并触发带有正确参数的未经授权的工具调用。评估显示，在现有基准上实现5%攻击成功率（ASR）的模型在LLMail-Inject上失败更多：专门针对提示注入进行训练的Meta SecAlign仍然产生32%的ASR，GPT-4o表现出58%的ASR，而GPT-5 mini则为47%的ASR。我们进一步证明攻击可以跨领域传输，在概念验证编码任务上实现37%的ASR，在文件管理任务上实现28-45%的ASR。我们发布了我们的挑战平台和精心策划的评估子集，具有多个难度层次，以及带有丰富注释的完整数据集。
更新时间: 2026-10-09 16:01:12
领域: cs.CR,cs.AI
ReplaySSM: Cache SSM Inputs, Not State
State-space models (SSMs) maintain a fixed-size recurrent state, but reading and writing it at every decoding step can bottleneck hybrid models. This eager state management causes large state I/O with low arithmetic intensity, leaves speculative decoding no cheap rollback, and feeds state quantization error into subsequent recurrent updates. We introduce ReplaySSM, which unrolls the recurrence to represent the state as a checkpoint and a buffer of recent SSM inputs. This separates output computation from checkpoint updates. By updating the checkpoint periodically, ReplaySSM avoids state writes on most steps and roughly halves the dominant state memory traffic. Keeping recent inputs also makes speculative rollback a buffer-pointer operation and enables output-only verification without materializing intermediate states. The same separation allows a low-precision checkpoint to serve output computation on most steps while a higher-precision checkpoint serves recurrent updates, so rounding error in the low-precision copy never enters later updates. We implement ReplaySSM in vLLM for Mamba-2 and Gated DeltaNet. At serving batch size 256, it accelerates standard decoding by $1.20$--$1.47\times$ over a tuned vLLM baseline across four hybrid models, and delivers about $1.8\times$ the throughput of vLLM's speculative decoding on the two 120B-class models. Separate state precisions also reduce output KL divergence and downstream accuracy losses compared with a shared low-precision checkpoint.
Updated: 2026-10-09 15:59:40
标题: ReplaySSM：缓存SSM输入，而不是状态
摘要: 状态空间模型（SSM）维护一个固定大小的循环状态，但在每个解码步骤中读取和写入该状态可能会成为混合模型的瓶颈。这种急切的状态管理导致大量的状态I/O和低算术强度，使得投机解码无法廉价回滚，并将状态量化误差传递到后续的循环更新中。我们引入了ReplaySSM，它将循环展开以表示状态为一个检查点和最近SSM输入的缓冲区。这将输出计算与检查点更新分开。通过定期更新检查点，ReplaySSM避免了大多数步骤上的状态写入，大致减少了主要的状态内存流量一半。保留最近的输入还使得投机回滚成为一个缓冲区指针操作，并且能够在不实现中间状态的情况下进行仅输出验证。相同的分离允许低精度检查点在大多数步骤上为输出计算提供服务，而更高精度的检查点则为循环更新提供服务，因此低精度副本中的舍入误差永远不会进入后续更新。我们在Mamba-2和Gated DeltaNet的vLLM中实现了ReplaySSM。在服务批量大小为256时，它在四个混合模型上比调整后的vLLM基线加速标准解码$1.20$--$1.47\times$，并且在两个120B级别的模型上交付了大约$1.8\times$ vLLM投机解码的吞吐量。分离的状态精度还与共享低精度检查点相比，减少了输出KL散度和下游准确性损失。
更新时间: 2026-10-09 15:59:40
领域: cs.LG,cs.AI
Geometry-Aware Multi-Objective Optimization: Norm-Constrained Oracles with Entropic Weight Updates
Multiple-gradient descent is typically formulated in Euclidean geometry, whereas matrix-aware optimizers use directions induced by other norms. We analyze a single-loop multi-objective method combining a norm-constrained linear minimization oracle with entropic updates of task weights. For any finite-dimensional norm pair, we establish an $O(T^{-1/2})$ ergodic stationarity rate in the deterministic setting, deriving the required trajectory gradient bound from smoothness and lower-boundedness in the unregularized case. With stochastic gradients and a bounded-gradient assumption, one persistent blended-gradient momentum buffer yields an $O(T^{-1/4})$ rate whose leading term has no explicit dependence on the number of tasks. We also establish static regret for the inner weighting objective on arbitrary bounded gradient sequences and, with entropy regularization and a weight floor, iteration-wise inner-gap control after burn-in while retaining ergodic stationarity, conditional on a trajectory gradient bound. The optimal blended gradient can be nonunique in spectral and sign geometries, motivating a selection-independent inner gap. Our analysis extends the spectral construction studied by concurrent MOON to general norms. Experiments compare the method with existing multi-task optimizers and examine its oracle, weighting, and momentum choices.
Updated: 2026-10-09 15:55:53
标题: 几何感知多目标优化：具有熵权重更新的规范约束神谕算法
摘要: 多梯度下降通常在欧几里得几何中制定，而矩阵感知优化器使用由其他范数诱导的方向。我们分析了一种单循环多目标方法，结合了约束范数线性最小化预言机和任务权重的熵更新。对于任何有限维范数对，在确定性设置中，我们建立了一个$O(T^{-1/2})$的遍历稳定速率，从而推导出在未正则化情况下所需的轨迹梯度界限。使用随机梯度和有界梯度假设，一个持续的混合梯度动量缓冲区产生了一个$O(T^{-1/4})$的速率，其主导项不显式依赖于任务数量。我们还在任意有界梯度序列上建立了内部加权目标的静态遗憾，并在经过燃烧阶段后保持遍历稳定性的情况下，使用熵正则化和权重下限来控制迭代内部差距，条件是存在轨迹梯度界限。最佳混合梯度可以在谱和符号几何中是非唯一的，这促使选择独立的内部差距。我们的分析将并发MOON研究的谱构造扩展到一般范数。实验比较了该方法与现有的多任务优化器，并检查了它的预言机、加权和动量选择。
更新时间: 2026-10-09 15:55:53
领域: math.OC,cs.LG
From Pixels to Privacy-Sensitive Features: An Information-Theoretic Framework for Reconstruction Bounds
What can an adversary recover from a model trained with DP-SGD? Its $(\varepsilon,δ)$ guarantee does not answer this question directly. We derive a distortion bound that can be computed before training and applies even to an adversary who observes the parameter trajectory during training, knows every other training record and the population distribution from which the training records are drawn. The analyst may choose the representation in which error is measured, allowing the bound to target the aspects of reconstruction that are privacy-sensitive in a given application. Our information-theoretic framework separates two contributions: the information about a record revealed by training, measured through conditional mutual information, and the uncertainty already present in the data distribution, measured through source entropy. Rate-distortion theory combines them into a lower bound on the expected error of every reconstruction attack. For DP-SGD, the leakage bound depends only on the sampling rates and noise multipliers, while the source entropy is lower-bounded from a sample with finite-sample confidence. An orthogonal block construction keeps the resulting bound informative for high-dimensional images. We instantiate the framework in pixel space and in LPIPS feature space, obtaining, to our knowledge, the first lower bound on reconstruction distortion measured directly by LPIPS. On a range of image datasets the bounds establish non-trivial unavoidable expected reconstruction error within heuristically recommended privacy-budget ranges and beyond. Experiments compare the bounds with strong reconstruction attacks and identify the principal sources of the remaining slack.
Updated: 2026-10-09 15:52:39
标题: 从像素到隐私敏感特征：用于重构界限的信息论框架
摘要: 对于用DP-SGD训练的模型，对手能够恢复什么？它的$(\varepsilon,δ)$保证不能直接回答这个问题。我们推导出一个畸变界限，可以在训练之前计算，并且适用于即使在训练过程中观察参数轨迹、了解每个训练记录以及从中抽样训练记录的总体分布的对手。分析人员可以选择衡量错误的表示形式，从而使边界针对在特定应用中涉及隐私的重建方面。我们的信息论框架将两个贡献分开：通过条件互信息衡量的训练过程中透露的记录信息，以及通过源熵衡量的数据分布中已经存在的不确定性。速率-畸变理论将它们结合在一起，形成对每次重建攻击的期望错误的下限。对于DP-SGD，泄漏界限仅取决于采样率和噪声倍增器，而源熵则从具有有限样本置信度的样本中下限。正交块构造使得得到的界限在高维图像上依然具有信息量。我们在像素空间和LPIPS特征空间中实例化了该框架，据我们所知，这是对由LPIPS直接衡量的重建畸变的第一个下限。在一系列图像数据集上，这些界限确定了在启发式建议的隐私预算范围内以及超出范围内的重建误差的非平凡不可避免期望。实验将界限与强重建攻击进行比较，并确定剩余松弛的主要来源。
更新时间: 2026-10-09 15:52:39
领域: stat.ML,cs.LG
DualQuant: Dual-scale Calibration-free Weight-and-Activation Quantization for LLMs
Low-bit inference on accelerators with native 4-bit support requires quantizing both weights and activations to realize the full throughput gains these formats provide. While post-training weight quantization has been extensively studied, activation quantization remains considerably more challenging and typically requires calibration data. In this work, we introduce DUALQUANT, a dual-scale, calibration-free post-training quantization framework that directly minimizes the weight-reconstruction Mean Squared Error (MSE) and, as an immediate by-product, yields per-channel activation rescalings prior to activation quantization. Our key observation is that the scaling factors obtained by minimizing weight-reconstruction error provide effective activation rescalings, enabling competitive activation quantization without calibration data. We formulate DUALQUANT as a non-differentiable weight-reconstruction problem and develop an efficient solver based on variable splitting and block coordinate descent, where each subproblem admits a closed-form update. DUALQUANT provides a unified algorithm supporting a wide range of formats, including the full MX family and NVFP4. To our knowledge, DUALQUANT is the first dual-scaling method to formally support MX and NV block formats. Across Llama and Qwen models evaluated on WikiText-2, C4, and zero-shot downstream tasks, DUALQUANT matches or exceeds the accuracy of calibration-based SmoothQuant across most models and formats, and substantially outperforms calibration-free baselines, with large gains for MX and NV block formats. We validate these gains with a real packed W4A4 inference path for Llama-3.1-8B on an RTX 4090. The resulting implementation closely matches the accuracy of fake quantization while delivering up to 2.61x memory reduction and up to 2.46x throughput speedup over BF16.
Updated: 2026-10-09 15:51:38
标题: DualQuant: 针对LLMs的双尺度无校准权重和激活量化
摘要: 具有本机4位支持的加速器上的低位推断需要对权重和激活进行量化，以实现这些格式提供的全速增益。尽管后训练权重量化已得到广泛研究，激活量化仍然相当具有挑战性，通常需要校准数据。在这项工作中，我们介绍了DUALQUANT，这是一个双尺度、无需校准的后训练量化框架，直接最小化权重重构均方误差（MSE），并且作为一个立即产生的副产品，在激活量化之前产生每个通道的激活重缩放。我们的关键观察是，通过最小化权重重构误差获得的缩放因子提供了有效的激活重缩放，从而实现了竞争性的激活量化而无需校准数据。我们将DUALQUANT制定为一个非可微的权重重构问题，并基于变量分裂和块坐标下降开发了一个高效的求解器，其中每个子问题都具有封闭形式的更新。DUALQUANT提供了一个统一的算法，支持一系列格式，包括完整的MX系列和NVFP4。据我们所知，DUALQUANT是第一个正式支持MX和NV块格式的双尺度方法。在WikiText-2、C4和零样本下游任务上评估的Llama和Qwen模型中，DUALQUANT在大多数模型和格式上与基于校准的SmoothQuant的精度相匹配或超过，并且大大优于无需校准的基线，在MX和NV块格式上获得了巨大的增益。我们在RTX 4090上验证了这些增益，实现了Llama-3.1-8B的实际打包W4A4推断路径。结果的实施与假量化的精度非常接近，同时在内存减少和吞吐量加速方面比BF16提高了高达2.61倍和2.46倍。
更新时间: 2026-10-09 15:51:38
领域: cs.LG,cs.AI
Shortcut Learning in Geospatial Foundation Model Embeddings for Cross-Disaster Damage Assessment
Geospatial foundation models can support disaster damage assessment, but high accuracy on familiar disasters may hide shortcut learning: prediction from event or scene characteristics rather than evidence that transfers to a new disaster. We study embeddings from TerraMind, Prithvi-EO-2.0, DOFA, and DINOv3 on the multi-disaster datasets xView2 and BRIGHT. Holding out an entire disaster, instead of randomly splitting image tiles, reduces mean balanced accuracy from 0.807 to 0.638 on xView2 and from 0.818 to 0.568 on BRIGHT. Four sets of tests examine what information supports prediction: event and acquisition context, before/after image inputs, building count, and correct image pairing. Pre-event images remain predictive, and a classifier given only the annotated number of buildings reaches balanced accuracy of 0.629 and 0.653. Combining before/after features performs best on average on xView2, but does not improve over pre-only features on BRIGHT. Shuffling image pairs within the same disaster and damage class causes modest losses. Motivated by these findings, we test whether feature transformations learned only from training disasters improve generalization to unseen disasters. These transformations improve performance, with supervised contrastive learning performing best on xView2 and PCA on BRIGHT. Our results highlight the importance of evaluation on unseen domains for assessing model robustness and show that simple feature-level transformations can improve generalization. Code and experiments can be found at https://github.com/alishibli97/gfm-shortcut-learning.
Updated: 2026-10-09 15:50:49
标题: 地理空间基础模型嵌入中的快捷学习，用于跨灾害损害评估
摘要: 地理空间基础模型可以支持灾害损害评估，但在熟悉的灾害上的高准确性可能会隐藏快捷学习：从事件或场景特征预测，而不是转移到新灾害的证据。我们研究了来自TerraMind、Prithvi-EO-2.0、DOFA和DINOv3的嵌入在多灾害数据集xView2和BRIGHT上。保留整个灾害，而不是随机拆分图像块，会使xView2的平均平衡准确率从0.807降至0.638，BRIGHT从0.818降至0.568。四组测试检查支持预测的信息：事件和获取上下文、前/后图像输入、建筑数量和正确的图像配对。预事件图像仍然具有预测性，仅给定建筑物注释数量的分类器达到0.629和0.653的平衡准确率。在xView2上，结合前/后特征的表现平均最佳，但在BRIGHT上与仅前特征相比没有改进。在同一灾害和损害类别内对图像对进行洗牌会导致适度损失。受这些发现的启发，我们测试了仅从训练灾害中学到的特征转换是否改进了对未见灾害的泛化。这些转换提高了性能，有监督对比学习在xView2上表现最佳，PCA在BRIGHT上表现最佳。我们的结果强调了对未见领域进行评估以评估模型鲁棒性的重要性，并显示简单的特征级转换可以改进泛化。可以在https://github.com/alishibli97/gfm-shortcut-learning找到代码和实验。
更新时间: 2026-10-09 15:50:49
领域: cs.CV,cs.LG
When Edge Independence Fails: Joint Graph Diffusion with Latent Sociability Priors
Discrete graph diffusion models corrupt graph structure through independent edge updates with a Markovian noise model. Combined with the exchangeability requirement on the output distribution, the terminal prior is necessarily an Erdős--Rényi (E-R) random graph. This has a fundamental consequence for unconditional graph generation without node features: in the large graph limit, any GNN denoiser applied to an E-R graph approximately outputs an E-R graph, making effective denoising difficult. In practice, generated graphs fail to recover meaningful node heterogeneity and higher-order structure. Previous models introduced specialized node features to compensate, framing this as an expressivity problem, while the role of the prior remained unexamined. We instead address the root cause by introducing the Latent Sociability Prior (LSP) and a join} diffusion process that co-evolves latent node sociability variables and graph structure, removing edge independence while preserving exchangeability. Unlike existing latent graph diffusion methods, the sociability mechanism requires no complex encoder: its sole role is to guide edge updates and preserve node heterogeneity throughout the trajectory. We validate our approach on multiple graph generation datasets and show that the proposed framework more accurately captures graph distributions than baseline methods without node features.
Updated: 2026-10-09 15:50:48
标题: 当边缘独立性失败：具有潜在社交性先验知识的联合图扩散
摘要: 离散图扩散模型通过具有马尔可夫噪声模型的独立边更新破坏了图结构。结合对输出分布的可交换性要求，终端先验必然是一个Erdős--Rényi（E-R）随机图。这对于无节点特征的无条件图生成具有基本影响：在大图限制下，任何应用于E-R图的GNN去噪器都会近似输出一个E-R图，使得有效去噪变得困难。在实践中，生成的图未能恢复有意义的节点异质性和高阶结构。先前的模型引入了专门的节点特征来补偿，将其视为一种表达能力问题，而先验的作用则未经检验。相反，我们通过引入潜在社交性先验（LSP）和一个共同的扩散过程来解决根本原因，这个过程共同演化潜在节点社交性变量和图结构，消除了边的独立性同时保持了可交换性。与现有的潜在图扩散方法不同，社交性机制不需要复杂的编码器：其唯一作用是引导边的更新并在整个轨迹中保留节点的异质性。我们在多个图生成数据集上验证了我们的方法，并表明所提出的框架比没有节点特征的基线方法更准确地捕捉了图分布。
更新时间: 2026-10-09 15:50:48
领域: cs.LG,cs.AI
Multi-Harness Reinforcement Learning for Coding Agents: Training Gains and Harness Dependence
Recent systems train coding agents with reinforcement learning (RL) on trajectories collected through multiple execution harnesses. However, gains under the training harnesses leave open whether the improvements transfer to unseen interfaces and how they depend on reward normalization. We present a controlled study that measures RL gains under fixed harnesses and compares the same models across harnesses. We train two Qwen3-8B RL variants from the same supervised fine-tuning (SFT) checkpoint using identical replay data. They differ only in whether rewards for each task are normalized within each harness (Within) or across harnesses (Cross). On SWE-bench Verified, we identify three main findings: (i) Multi-harness RL yields modest gains over the SFT baseline under both normalization rules. Across the four training harnesses, the estimated mean improvements are +0.48 and +0.77 percentage points for Within and Cross, respectively. (ii) Cross increases the fraction of trajectories with nonzero advantage but provides no consistent advantage over Within on an unseen harness across training seeds and evaluation budgets; neither rule significantly improves over SFT on this harness. (iii) Performance varies substantially more across harnesses than with RL training. After averaging over model variants, the highest and lowest source-harness solve rates differ by 7.13 percentage points, far exceeding the estimated training gains. Even after multi-harness RL, the observed performance differences across execution interfaces remain substantially larger than either the gains from training or the differences between normalization rules.
Updated: 2026-10-09 15:50:30
标题: 多束缰强化学习用于编码代理：训练效益和束缰依赖
摘要: 最近的系统使用强化学习（RL）在通过多个执行控制装置收集的轨迹上训练编码代理。然而，在训练控制装置下获得的收益并没有完全说明这些改进是否能转移到未见过的接口上，以及它们如何依赖奖励归一化。我们提出了一个受控研究，测量了在固定控制装置下的RL收益，并比较了相同模型在不同控制装置下的情况。我们使用相同的回放数据从相同的监督微调（SFT）检查点训练了两种Qwen3-8B RL变体。它们之间唯一的区别在于每个任务的奖励是在每个控制装置内（Within）还是跨控制装置（Cross）归一化。在SWE-bench Verified上，我们得出了三个主要发现：（i）多控制装置RL相对于SFT基线获得了适度的增益，无论是在哪种规则下归一化。在四个训练控制装置中，Within和Cross的估计平均改进分别为+0.48和+0.77个百分点。(ii) Cross增加了具有非零优势轨迹的比例，但在训练种子和评估预算下，与Within相比在未见过的控制装置上并没有稳定的优势；在这种控制装置上，两种规则都没有显著优于SFT。(iii) 性能在控制装置之间的差异远远大于RL训练的情况。在对模型变体进行平均后，最高和最低的源控制装置解决率相差7.13个百分点，远远超过估计的训练增益。即使经过多个控制装置的RL训练，观察到的执行接口之间的性能差异仍然远远大于训练的增益或归一化规则之间的差异。
更新时间: 2026-10-09 15:50:30
领域: cs.AI
How Well Do Prior-Data Fitted Networks Amortize Bayesian Inference?
Prior-data fitted networks (PFNs) are trained to approximate the Bayesian posterior predictive distribution (PPD) of a synthetic prior, yet results about the approximation quality are surprisingly thin. Taking an empirical risk minimization perspective, we decompose the deployment error of PFNs into prior, context, and amortization error, and measure the latter against a reference PPD under a realistic prior. Results show that PFNs generally recover the targeted PPD very well. Nevertheless, we find that parts of the residual amortization error persist, especially in \textit{rare} settings under the prior pretraining distribution, even when scaling the model size, which manifests in disagreement between independently pretrained PFNs and leaves uncertainty about the predictor at inference time. As a remedy, we study multiple types of Bayesian model averages and find that these can consistently reduce this uncertainty and improve predictions up to full-size TabICLv2 models.
Updated: 2026-10-09 15:49:24
标题: 之前数据拟合网络对贝叶斯推断的摊销效果如何？
摘要: Prior-data fitted networks（PFN）被训练来逼近一个合成先验的贝叶斯后验预测分布（PPD），然而关于逼近质量的结果却相当稀少。从经验风险最小化的角度出发，我们将PFN的部署错误分解为先验、上下文和摊销错误，并将后者与一个真实先验下的参考PPD进行比较。结果显示，PFN通常能很好地恢复目标PPD。然而，我们发现部分残余摊销错误仍然存在，尤其是在先验预训练分布下的\textit{稀有}情况下，即使扩大模型规模，这种错误仍然存在，这表现为独立预训练的PFN之间存在分歧，并在推断时留下了关于预测器的不确定性。为了解决这个问题，我们研究了多种类型的贝叶斯模型平均值，并发现这些方法可以始终降低这种不确定性，并提高预测性能，甚至可以提升到全尺寸的TabICLv2模型。
更新时间: 2026-10-09 15:49:24
领域: stat.ML,cs.LG
Hamiltonian locality testing and certification do not achieve the Heisenberg limit
We establish lower bounds for Hamiltonian property testing with access to the time-evolution operator but not its inverse. Each experiment may query the time-evolution operator multiple times, and distances between Hamiltonians are measured in the normalized Frobenius norm. In this model, we show that testing whether a Hamiltonian is $k$-local or $\varepsilon$-far from every $k$-local Hamiltonian requires $Ω(1/\varepsilon^2)$ total evolution time, matching the upper bound of Kallaugher and Liang (TQC'25). We also prove that testing whether an unknown Hamiltonian equals a target Hamiltonian or is $\varepsilon$-far from it requires $Ω(1/\varepsilon^2)$ total evolution time, matching the upper bound of Sinha and Tong (2025). These are the first lower bounds for natural problems in Hamiltonian learning and testing that rule out Heisenberg-limited scaling of $1/\varepsilon$. As a third result, we show that amplitude estimation to precision $\varepsilon$ requires $Ω(1/\varepsilon^2)$ total time evolution, recovering the result of Tang and Wright (QIP'26) in the continuous-time query model. All three results follow from the hardness of distinguishing the zero Hamiltonian from a suitably chosen ensemble of random Hamiltonians. We establish this hardness by adapting the continuous-time adversary method to forward Hamiltonian evolution.
Updated: 2026-10-09 15:43:09
标题: 哈密顿局部性测试和认证无法达到海森堡极限。
摘要: 我们建立了哈密顿性质测试的下界，其中可以访问时间演化算子，但不能访问其逆算子。每个实验可能多次查询时间演化算子，并且哈密顿量之间的距离以归一化的Frobenius范数进行测量。 在这个模型中，我们证明测试一个哈密顿量是否为$k$-局部的或者与每个$k$-局部哈密顿量$\varepsilon$远的问题需要$Ω(1/\varepsilon^2)$的总演化时间，与Kallaugher和Liang（TQC'25）的上界相匹配。我们还证明，测试一个未知的哈密顿量是否等于目标哈密顿量或者与其$\varepsilon$远的问题需要$Ω(1/\varepsilon^2)$的总演化时间，与Sinha和Tong（2025）的上界相匹配。这些是对于哈密顿学习和测试中自然问题的第一个下界，排除了Heisenberg极限尺度的$1/\varepsilon$。 作为第三个结果，我们证明到精度$\varepsilon$的幅度估计需要$Ω(1/\varepsilon^2)$的总时间演化，恢复了Tang和Wright（QIP'26）在连续时间查询模型中的结果。这三个结果都是基于区分零哈密顿量和一组随机哈密顿量的困难性而得出的。我们通过将连续时间对手方法调整为正向哈密顿演化来建立这种困难性。
更新时间: 2026-10-09 15:43:09
领域: quant-ph,cs.CC,cs.DS,cs.LG
Matched-Context Group Policy Optimization for LLM Agent Training
Training long-horizon language model agents requires assigning credit to intermediate decisions from sparse terminal feedback. Within a finite group of sibling rollouts, useful action comparisons depend on their context, evidence for both alternatives, and how that evidence enters learning. We propose Matched-Context Group Policy Optimization (MCGPO), a procedure for constructing weighted action preferences from already collected trajectories. MCGPO defines comparison contexts with pre-action state summaries and observed rubric progress, annotated by task rules. Two-action support selects among ordered exact and coarse matching rules. Fixed folds separate candidate decisions from the terminal outcomes used to orient their preferences, and support-dependent evidence weights scale an auxiliary pairwise objective alongside shared GraphGPO task learning. MCGPO reuses rollout data and each response's original history, preserving critic-free training. Across six environment-backbone settings with three training seeds each, MCGPO improves mean success over matched GraphGPO: by 0.47-1.46 percentage points on ALFWorld and WebShop with Qwen2.5-1.5B and 7B, and by 3.96 and 5.16 points on WebShop with Qwen3-4B and 8B. The ALFWorld gains build on baseline success above 92%, where remaining headroom is limited. Integrated success improves in three of four Qwen2.5 settings. Three-seed factorial controls support opposite-fold evidence and evidence weighting; joint state-rubric matching achieves the highest mean in all four representation-control settings.
Updated: 2026-10-09 15:41:37
标题: 匹配上下文的群体政策优化对LLM代理培训的影响
摘要: 训练长期语言模型代理需要为稀疏终端反馈中的中间决策分配信用。在有限的兄弟回合组内，有用的动作比较取决于它们的上下文、两个替代方案的证据以及该证据如何进入学习。我们提出了匹配上下文组策略优化（MCGPO），这是一种从已收集的轨迹中构建加权动作偏好的过程。MCGPO使用预先动作状态摘要和观察到的标准进展来定义比较上下文，由任务规则进行注释。双动作支持在有序的精确和粗略匹配规则之间进行选择。固定折叠将候选决策与用于定向其偏好的终端结果分开，并且支持相关证据权重在共享的GraphGPO任务学习旁边缩放一个辅助成对目标。MCGPO重复使用回合数据和每个响应的原始历史，保持无批评训练。在六个环境-主干设置中，每个设置有三个训练种子，MCGPO在ALFWorld和WebShop与Qwen2.5-1.5B和7B上比匹配的GraphGPO提高了0.47-1.46个百分点，并在WebShop与Qwen3-4B和8B上提高了3.96和5.16个点。ALFWorld的增益建立在基线成功率超过92%的基础上，剩余的潜力有限。在四个Qwen2.5设置中，集成成功率有所提高。三个种子因子控制支持相反折叠证据和证据加权；联合状态-标准匹配在四个表示控制设置中均获得最高平均值。
更新时间: 2026-10-09 15:41:37
领域: cs.LG,cs.AI
Studying the Effects of Task Diversity Using Small Mathematical Worlds
Basic mathematical tasks like modular addition have been a valuable tool for illuminating aspects of neural network behavior that are challenging to see in noisy, unverifiable, large-scale, real-world settings. These studies have typically focused on a single task. However, modern language models are exposed to a vast number of tasks despite being trained with a next-token prediction objective. In this work, we introduce two mathematical task ecosystems, IntWorld and PermWorld, built from distinct but related tasks on integers and permutations. Both are designed to simulate task diversity within a cohesive mathematical 'world' where tasks are distinct but related. Because the datasets are procedurally generated, we can precisely vary the number of tasks, allowing us to begin to disentangle the role of task variety from other factors in training. After describing the datasets, we provide some initial results obtained from these datasets which support the idea that increasing task diversity increases the richness of a model's internal representations.
Updated: 2026-10-09 15:39:42
标题: 研究使用小型数学世界研究任务多样性的影响
摘要: 基本的数学任务，如模块加法，已经成为揭示神经网络行为方面的有价值工具，这些方面在嘈杂、难以验证、大规模、现实世界环境中难以观察到。这些研究通常侧重于单一任务。然而，现代语言模型尽管以下一个标记预测目标进行训练，但暴露于大量任务之中。在本文中，我们介绍了两个数学任务生态系统，IntWorld和PermWorld，它们由整数和排列的不同但相关任务构建而成。两者都旨在模拟在一个统一的数学“世界”内的任务多样性，其中任务是独特但相关的。由于数据集是程序生成的，我们可以精确地改变任务数量，从而开始区分任务多样性与训练中的其他因素之间的作用。在描述数据集之后，我们提供了一些从这些数据集中获得的初步结果，支持增加任务多样性会增加模型内部表示的丰富性的想法。
更新时间: 2026-10-09 15:39:42
领域: cs.LG
Access Paths for Efficient Ordering with Large Language Models
In this work, we present the \texttt{LLM ORDER BY} semantic operator as a logical abstraction and conduct a systematic study of its physical implementations. First, we propose several improvements to existing semantic sorting algorithms and introduce a semantic-aware external merge sort algorithm. Our extensive evaluation reveals that no single implementation offers universal optimality on all datasets. From our evaluations, we observe a general scaling relationship between sorting cost and the ordering quality for comparison-based algorithms. Building on these insights, we design a budget-aware optimizer that utilizes heuristic rules, LLM-as-Judge evaluation, and consensus aggregation to dynamically select the near-optimal access path for LLM ORDER BY. In our extensive evaluations, our optimizer consistently achieves ranking accuracy on par with or superior to the best static methods across all benchmarks. We believe that this work provides foundational insights into the principled optimization of semantic operators essential for building robust, large-scale LLM-powered analytic systems.
Updated: 2026-10-09 15:39:40
标题: 大语言模型的高效排序访问路径
摘要: 在这项工作中，我们提出了\texttt{LLM ORDER BY}语义操作符作为一个逻辑抽象，并对其物理实现进行了系统研究。首先，我们对现有的语义排序算法进行了几项改进，并引入了一种语义感知的外部合并排序算法。我们的广泛评估表明，没有任何单一实现在所有数据集上都能提供普遍的最优性。从我们的评估中，我们观察到基于比较的算法的排序成本与排序质量之间存在一般的缩放关系。基于这些见解，我们设计了一个预算感知的优化器，利用启发式规则、LLM作为评判者的评估以及共识聚合来动态选择LLM ORDER BY的近似最优访问路径。在我们的广泛评估中，我们的优化器始终在所有基准测试中实现了与或优于最佳静态方法的排名准确性。我们相信这项工作为构建稳健、大规模LLM驱动的分析系统所必需的语义操作符的原则优化提供了基础性见解。
更新时间: 2026-10-09 15:39:40
领域: cs.DB,cs.AI,cs.IR
A Privileged Teacher That Only Scores Cannot Search: Critique-Conditioned On-Policy Self-Distillation for Reasoning
On-policy self-distillation promises to replace the sparse outcome signal of verifiable-reward RL with dense token-level credit: a teacher conditioned on the reference solution scores the student's own rollout, and the resulting distributional shift is distilled back into the policy. We investigate whether this scoring signal actually contains corrective credit. Using gold error spans in student rollouts, we test the two signals consumed by self-distillation objectives, the full-distribution KL and the token log-ratio, for two properties: whether they localize errors and whether they push probability away from erroneous student tokens. They exhibit neither of these properties nor can they reliably distinguish correct from incorrect rollouts within the same problem. We argue that this failure is structural: locating an error in a reasoning trace is itself a search problem, while a scoring teacher is never explicitly asked to perform that search. The same model, with the same privileged information, localizes errors substantially better when allowed to generate a critique. This forms the motivation for \emph{critique-conditioned self-distillation}: first generate a critique of the current rollout, then condition the scoring teacher on that critique. The resulting signal localizes errors, is corrective within error spans, and shifts probability toward tokens that initiate backtracking such as ``Wait'', ``But'', and ``Actually''. We further show that the critique is useful both at test time, where resuming a wrong rollout with the critique in context converts math problems the model cannot solve by sampling, and in cold-start GRPO, where it provides an advantage signal on prompts for which verifiable rewards alone provide none.
Updated: 2026-10-09 15:36:47
标题: 一个只会打分的特权老师无法探索：基于政策的自我蒸馏批判条件下的推理
摘要: 在政策自我蒸馏中，承诺用密集的标记级信用替代可验证奖励RL的稀疏结果信号：一个以参考解决方案评分为条件的教师评估学生自己的展开，然后将结果的分布偏移蒸馏回政策。我们研究了这种评分信号是否实际包含纠正信用。使用学生展开中的金标准错误范围，我们测试了自我蒸馏目标所消耗的两个信号，即完整分布KL和标记对数比，针对两个属性：它们是否定位错误以及它们是否将概率从错误的学生标记推开。它们都没有显示这些属性，也不能可靠地区分相同问题中的正确和不正确的展开。我们认为这种失败是结构性的：在推理追踪中定位错误本身就是一个搜索问题，而评分教师从未明确要求执行该搜索。当允许生成评论时，相同模型、相同特权信息在定位错误方面表现得更好。这构成了“评论条件自我蒸馏”的动机：首先生成当前展开的评论，然后将评分教师置于该评论的条件下。结果信号定位错误，对错误范围内具有纠正作用，并将概率转移到触发回溯的标记，如“等待”、“但是”和“实际上”。我们进一步展示了评论在测试时间和冷启动GRPO中的实用性，在测试时间，通过在上下文中使用评论恢复错误的展开，将模型无法通过抽样解决的数学问题转化为解决，而在冷启动GRPO中，评论为仅提供可验证奖励的提示提供了优势信号。
更新时间: 2026-10-09 15:36:47
领域: cs.AI,cs.CL,cs.LG
A Generalizable Machine Learning Framework for Half-Heusler Thermoelectrics
Machine learning (ML) has emerged as a powerful approach for accelerating the discovery of thermoelectric (TE) materials, yet many models suffer from limited generalizability. Here we present a robust ML framework for predicting the thermoelectric figure of merit (zT) in half-Heusler (hH) compounds, emphasizing prediction of previously unseen compositions. A principal component analysis (PCA)-based train/test splitting strategy is introduced to generate representative partitions of chemical space. Bayesian hyperparameter optimization is combined with correlation-based k-best feature selection for Random Forest, XGBoost, and Neural Network models, together with the interpretable SISSO symbolic-regression method. Relative to the full set of 118 input features, the optimized feature subsets reduce the test RMSE by 16-27%. SHAP and SISSO analyses identify temperature, A-site dopant concentration, and A-site heat of vaporization as the most important descriptors of zT. The framework is validated through prediction of previously unexplored compositions, recovery of literature-reported optimal zT values for established hH systems, and first-principles calculations for the new compounds CeCoBi and CeIrBi, which exhibit favorable electronic transport properties. The framework is further applied to high-throughput screening of the doped hH chemical space, covering about 2.46 million elemental substitution combinations across 631 thermodynamically screened parent matrices under stability constraints. This screening yields a concise set of promising candidates for future study. Overall, the work establishes a generalizable workflow for reliable data-driven thermoelectric materials discovery.
Updated: 2026-10-09 15:36:15
标题: 一个适用于半海氏热电材料的通用机器学习框架
摘要: 机器学习（ML）已经成为加速热电（TE）材料发现的强大方法，然而许多模型存在有限的泛化能力。在这里，我们提出了一个强大的ML框架，用于预测半Heusler（hH）化合物中的热电功率因子（zT），强调先前未见过的组成物的预测。引入了基于主成分分析（PCA）的训练/测试分割策略，以生成化学空间的代表性分区。贝叶斯超参数优化结合基于相关性的k最佳特征选择用于随机森林、XGBoost和神经网络模型，以及可解释的SISSO符号回归方法。与118个输入特征的完整集相比，经过优化的特征子集将测试RMSE减少了16-27%。SHAP和SISSO分析确定了温度、A位掺杂浓度和A位汽化热作为zT的最重要描述符。该框架通过预测以前未探索的组成物、恢复已建立的hH系统的文献报道的最佳zT值以及新化合物CeCoBi和CeIrBi的第一性原理计算来进行验证，这些化合物表现出有利的电子输运特性。该框架进一步应用于掺杂的hH化学空间的高通量筛选，覆盖了稳定性约束下的631个热力学筛选母基质中的约246万个元素替代组合。该筛选得出了一组有前途的候选物质，供未来研究使用。总的来说，这项工作建立了一个可靠的数据驱动热电材料发现的通用工作流程。
更新时间: 2026-10-09 15:36:15
领域: cond-mat.mtrl-sci,cs.LG
SmolWAM: A Latent World-Action Model for Robotic Manipulation on Edge Devices
World-action models (WAMs) predict a robot's actions together with the observations those actions will produce. However, the strongest models rely on video diffusion transformers that are far too computationally expensive for resource-constrained edge devices. We present SmolWAM, a latent WAM that we derive from the 2B-parameter Cosmos Policy by keeping its pretrained trunk and rebuilding its inference interface: single-frame visual latents, cached text features that preserve the trunk's attention sinks, 12 of 28 transformer blocks, and 1,008 tokens from which actions and future visual latents are read out in one rectified-flow step. Trained in 3.6 GPU-hours, SmolWAM succeeds in 95.5% of 400 LIBERO episodes at 57 ms and 2.3 TFLOP per 16-action chunk with 2.4 GiB of GPU memory, 11x faster than its teacher with 18.5x fewer FLOPs and 3.2x less memory, and its world head still forecasts scene motion.
Updated: 2026-10-09 15:34:33
标题: SmolWAM：用于边缘设备上机器人操作的潜在世界-动作模型
摘要: 世界行动模型（WAMs）预测机器人的动作以及这些动作将产生的观察结果。然而，最强大的模型依赖于视频扩散变压器，这对于资源受限的边缘设备来说计算代价太高。我们提出了SmolWAM，这是一个潜在的WAM，我们通过保留其预训练的主干并重新构建其推理接口来从2B参数的Cosmos策略中得到的：单帧视觉潜在、保留主干注意力汇聚的缓存文本特征、28个变压器块中的12个，以及1,008个令牌，这些令牌可以在一个修正流步骤中读取出动作和未来视觉潜在。在3.6个GPU小时内训练，SmolWAM在400个LIBERO episode中的95.5%中成功，每个16个动作块耗时57毫秒和2.3 TFLOP，GPU内存为2.4 GiB，比其教师快11倍，FLOP少18.5倍，内存少3.2倍，而其世界头部仍然可以预测场景运动。
更新时间: 2026-10-09 15:34:33
领域: cs.RO,cs.LG
Learning to Report Unsafe Tasks in a Multi-Agent Game
When a team of agents shares a reward for completed tasks, reporting and stopping unsafe work reduces the team's reward. Random audits that penalize silence can make universal reporting optimal without ensuring that a silent team learns to report. We study this problem in a game where any witness can stop a task by reporting. When a task's k witnesses share a policy and draw actions independently, the derivative of expected reward with respect to the shared silence probability at universal silence counts its benefit k times, while the comparison between universal reporting and silence counts it once. A fully shared policy can therefore be locally attracted to silence even when reporting is optimal. For any grouping into shared policies, we give an audit condition sufficient for exact policy-gradient updates to reach universal reporting and, outside boundary cases, necessary from near universal silence. In a balanced family, meeting the condition with prescribed margins costs k times as much for a fully shared policy as for one policy per role. We test these predictions with PPO on 24 witness graphs, starting from learned silence. Witness groups, which place co-witnesses in separate networks, reduce unsafe task completion by 33.59 percentage points relative to shuffled groups of the same sizes under the same audits (95% graph-bootstrap interval: 21.03-45.13). Only 9 of 48 witness-group runs achieve below 1% unsafe completion while retaining at least 90% legitimate completion. At the same audit budget, a fully shared network meets both thresholds in 0/48 runs with independent action draws and 48/48 with a common draw. On 24 further graphs with witness groups fixed, audits allocated by the condition reduce unsafe completion by 42.22 points relative to audits maximizing a static coalition deterrence ratio at the same budget.
Updated: 2026-10-09 15:34:05
标题: 学习在多智能体游戏中报告不安全任务
摘要: 当一组代理共享已完成任务的奖励时，报告和停止不安全工作会减少团队的奖励。对沉默进行惩罚的随机审计可以使普遍报告成为最佳选择，但并不能确保沉默的团队学会报告。我们在一个游戏中研究了这个问题，在这个游戏中，任何证人都可以通过报告来停止一个任务。当一个任务的k个证人共享政策并且独立地进行行动选择时，期望奖励对于共享沉默概率的导数，在普遍沉默时计算其收益k次，而普遍报告和沉默之间的比较只计算一次。因此，即使报告是最佳选择，一个完全共享的政策也可能在局部吸引到沉默。对于任何共享政策的分组，我们给出了一个审计条件，使得确切的政策梯度更新能够实现普遍报告，并且在边界情况之外，从几乎普遍沉默所需的条件。在一个平衡家庭中，以预定边界的成本对于一个完全共享的政策来说，要比每个角色一个政策的成本高出k倍。我们使用PPO在24个证人图上进行了这些预测的测试，从学习到的沉默开始。证人组将共同证人放置在不同的网络中，相对于相同大小的随机分组在相同审计下不安全任务完成减少了33.59个百分点（95%图随机区间：21.03-45.13）。48个证人组运行中只有9个达到了低于1%的不安全完成，同时至少保持90%的合法完成。在相同的审计预算下，一个完全共享的网络在独立行动选择时没有任何一个达到这两个阈值，而在共同选择时有48/48达到。在24个固定证人组的进一步图上，根据条件分配审计相对于在相同预算下最大化静态联盟威慑比例的审计，不安全完成减少了42.22个百分点。
更新时间: 2026-10-09 15:34:05
领域: cs.AI,cs.MA
SafeDiSTL: Decomposing Joint Constraints for Safe Multi-Agent Reinforcement Learning
Cooperative multi-agent tasks often impose constraints on a shared objective or collective safety. A team level violation cost function can measure this constraint outcome, but does not distinguish which agents' actions most strongly influence that violation. We present SafeDiSTL, a constrained multi-agent reinforcement learning framework that allows for credit assignment even with joint constraint measurements. We use Signal Temporal Logic (STL) robustness to measure episode-level violation through running-maximum cost increments. Smooth full-episode robustness and differentiable dynamics provide action sensitivities, which allocate each cost increment across agents at the same decision time. These sensitivities account for delayed effects when weighting agents' shares of each cost increment. We study the framework in cooperative payload transport, connected navigation, and networked traffic control tasks. SafeDiSTL achieves 1.6-5.1x the safe-success rate in transport, improves navigation safe-success by 11.1%, and delivers 50-68% more throughput in traffic when compared to constrained baselines. Results demonstrate a strong improvement of the learned policy in dealing with shared constraint while retaining the performance of unconstrained policies.
Updated: 2026-10-09 15:32:06
标题: SafeDiSTL：将联合约束分解用于安全的多智体强化学习
摘要: 合作式多智能体任务通常对共享目标或集体安全施加约束。团队级别的违规成本函数可以衡量这种约束结果，但不能区分哪些智能体的行动最大程度地影响了违规。我们提出了SafeDiSTL，这是一个受限多智能体强化学习框架，即使在联合约束测量下也可以进行信用分配。我们使用信号时间逻辑（STL）的强度来通过运行最大成本增量来衡量每一集违规。平滑的完整集强度和可微分动态提供了行动敏感性，这些敏感性在相同决策时间内将每个成本增量分配给智能体。这些敏感性考虑了在对智能体的每个成本增量进行加权时的延迟效应。我们在合作式负载运输、联合导航和网络交通控制任务中研究了该框架。与受限基线相比，SafeDiSTL在运输方面实现了1.6-5.1倍的安全成功率，将导航安全成功率提高了11.1%，并在交通方面提供了50-68%的吞吐量增加。结果表明，在处理共享约束时，学习策略的表现得到了显著改善，同时保持了不受限制策略的性能。
更新时间: 2026-10-09 15:32:06
领域: cs.LG,cs.AI
Orthros: Phase-Aware Heterogeneous Attention for Efficient Transformers
Efficient attention mechanisms, such as linear and sliding window attention, aim to reduce the computational complexity of Transformers from quadratic to linear. However, their degraded in-context recall often necessitates interleaving full attention layers, leaving the quadratic bottleneck unresolved in long-context processing. To address this, we introduce Orthros attention, which shifts the efficiency paradigm from structural allocation to phase-aware decoupling. As a unified module, it dynamically switches its computation pattern across different phases of sequence processing, enabling linear complexity prefilling and full attention decoding. Our design grants Orthros attention a lower asymptotic complexity compared to vanilla Transformers in prevalent long-context scenarios characterized by heavy prefilling and light decoding. Beyond theoretical formulation and complexity analysis, extensive empirical evaluations demonstrate Orthros's superiority across multiple dimensions. First, we validate its fundamental language modeling capabilities, demonstrating scalability across varying model sizes and strong competitiveness in comprehensive architectural comparisons. Second, Orthros exhibits exceptional conversion feasibility via a data-efficient Transformer-to-Orthros conversion, yielding performance that even surpasses the original Transformer. Finally, Orthros achieves a $21.9\times$ prefilling speedup over Transformers with end-to-end inference gains, while preserving long-context capabilities, as demonstrated by near-perfect 128K needle-in-a-haystack results and Transformer-level performance on real-world in-context recall tasks. Overall, these advantages in high performance, flexibility, and efficiency position Orthros as a promising candidate to succeed the Transformer in industrial applications.
Updated: 2026-10-09 15:30:08
标题: Orthros：面向效率变压器的相位感知异构注意力
摘要: Efficient attention mechanisms, such as linear and sliding window attention, aim to reduce the computational complexity of Transformers from quadratic to linear. However, their degraded in-context recall often necessitates interleaving full attention layers, leaving the quadratic bottleneck unresolved in long-context processing. To address this, we introduce Orthros attention, which shifts the efficiency paradigm from structural allocation to phase-aware decoupling. As a unified module, it dynamically switches its computation pattern across different phases of sequence processing, enabling linear complexity prefilling and full attention decoding. Our design grants Orthros attention a lower asymptotic complexity compared to vanilla Transformers in prevalent long-context scenarios characterized by heavy prefilling and light decoding. Beyond theoretical formulation and complexity analysis, extensive empirical evaluations demonstrate Orthros's superiority across multiple dimensions. First, we validate its fundamental language modeling capabilities, demonstrating scalability across varying model sizes and strong competitiveness in comprehensive architectural comparisons. Second, Orthros exhibits exceptional conversion feasibility via a data-efficient Transformer-to-Orthros conversion, yielding performance that even surpasses the original Transformer. Finally, Orthros achieves a $21.9\times$ prefilling speedup over Transformers with end-to-end inference gains, while preserving long-context capabilities, as demonstrated by near-perfect 128K needle-in-a-haystack results and Transformer-level performance on real-world in-context recall tasks. Overall, these advantages in high performance, flexibility, and efficiency position Orthros as a promising candidate to succeed the Transformer in industrial applications.
更新时间: 2026-10-09 15:30:08
领域: cs.LG,cs.AI
LAPT-Scale: Scaling Embodied Policies with Latent Action Pre-Training on Human Videos
Scaling generalist robotic policies is limited by the high cost of collecting embodied interaction data. Large-scale egocentric human videos offer a rich alternative, but they lack robot action labels, and existing action annotation methods are often noisy under real-world visual complexity and diverse manipulation dynamics. We introduce LAPT-Scale, a scalable latent action pre-training paradigm with two components: LAPT-Scale-LAM learns continuous latent actions by explaining frame-to-frame visual transitions in human videos and uses them as human-derived supervision, while LAPT-Scale-Policy trains a VLA policy through human-video latent-action pre-training, target-robot embodiment alignment, and task-specific post-training. Unlike prior vector-quantized LAMs, LAPT-Scale-LAM uses a continuous latent space to preserve fine-grained motion and interaction dynamics. We further combine perceptual, semantic, and optical-flow alignment signals to improve transition-level supervision beyond pixel reconstruction. Trained on 23.3K hours of egocentric video, LAPT-Scale-LAM provides implicit action supervision for policy pre-training, yielding consistent scaling gains and improving real-robot success from 63.3% to 85.0% as annotated human video scales from 1K to 5K hours. The learned policy also generalizes to unseen visual conditions.
Updated: 2026-10-09 15:29:20
标题: LAPT-Scale：通过在人类视频上进行潜在动作预训练来扩展具身体性的策略
摘要: 通用机器人策略的扩展受到采集实体交互数据成本高昂的限制。大规模的人类自我中心视频提供了一个丰富的替代方案，但它们缺乏机器人动作标签，现有的动作注释方法在现实世界的视觉复杂性和各种操作动态下通常存在噪声。我们引入了LAPT-Scale，这是一个可扩展的潜在动作预训练范式，包括两个组件：LAPT-Scale-LAM通过解释人类视频中帧与帧之间的视觉转换来学习连续的潜在动作，并将其用作人类衍生的监督，而LAPT-Scale-Policy通过人类视频潜在动作预训练、目标机器人体现对齐和特定任务后训练来训练VLA策略。与先前的矢量量化LAM不同，LAPT-Scale-LAM使用连续潜在空间来保留细粒度的运动和交互动态。我们进一步结合感知、语义和光流对齐信号，以改善超越像素重建的转换级监督。在训练了23.3K小时的自我中心视频后，LAPT-Scale-LAM为策略预训练提供了隐式动作监督，产生了一致的扩展收益，并将实际机器人成功率从63.3%提高到85.0%，而标注的人类视频规模从1K小时扩展到5K小时。学到的策略还能泛化到未见的视觉条件。
更新时间: 2026-10-09 15:29:20
领域: cs.CV,cs.AI,cs.RO
SeoulMMOD: A Large-Scale Multimodal Origin-Destination Flow Benchmark
Origin-destination (OD) flow forecasting supports urban mobility decisions by predicting the volume and patterns of human movement across a city. Most existing public OD benchmarks provide only a partial view of citywide mobility, covering a single transport service or at most two modes. With partial mode coverage, observed demand changes are difficult to separate into shifts across transport modes and changes in total travel. We introduce SeoulMMOD, a large-scale public benchmark for citywide multimodal mobility in Seoul. SeoulMMOD provides three years of hourly OD flows estimating total mobility for 6 urban travel modes across 25 districts and 426 sub-districts, together with travel-time, travel-distance, calendar, rainfall, and point-of-interest (POI) signals. Using SeoulMMOD, we benchmark 14 forecasting baselines under a common protocol and analyse the effects of spatial scaling, joint versus mode-separate training, and cross-year generalization. Results show that several spatio-temporal models that work at district level do not scale to sub-district OD forecasting, and that naive joint multi-mode training does not consistently improve accuracy. Together, the dataset and benchmark establish a reproducible testbed for multi-mode, multi-year OD forecasting at the city scale.
Updated: 2026-10-09 15:27:32
标题: SeoulMMOD：一个大规模的多模式出发地-目的地流量基准
摘要: 起点-终点（OD）流量预测通过预测跨越城市的人类移动量和模式来支持城市交通决策。大多数现有的公共OD基准仅提供了城市范围内交通的部分视图，涵盖了单一交通服务或最多两种模式。由于部分模式覆盖，观察到的需求变化很难分离出在交通模式之间的转变以及总出行量的变化。我们介绍了SeoulMMOD，这是首尔城市范围内多模式移动的大规模公共基准。SeoulMMOD提供了3年的每小时OD流量，估计了25个区域和426个子区域内6种城市出行模式的总出行量，同时提供了旅行时间、旅行距离、日历、降雨和兴趣点（POI）信号。利用SeoulMMOD，我们在一个共同的协议下对14种预测基线进行基准测试，并分析了空间缩放、联合训练与模式分开训练以及跨年度泛化的影响。结果显示，一些在区域级别有效的时空模型不适用于子区域OD预测，并且天真的联合多模式训练并不能始终提高准确性。总的来说，这个数据集和基准为城市范围内多模式、多年度OD预测建立了一个可复制的测试平台。
更新时间: 2026-10-09 15:27:32
领域: cs.LG
Dissecting Representation Structure in Vision Transformers: A Rigorous Architectural Study
Representation structure is crucial for understanding Vision Transformer (ViT) architectures and their generalization behavior. However, prior studies neither isolate nor analyze module-level features nor investigate how their interactions contribute to performance estimation. In this work, we conduct the first rigorous analysis of feature information across diverse architectural scales, empirically uncover the relationship between ViT representation and generalization behavior, and leverage these insights to guide efficient ViT design. Our contributions are fivefold: Across diverse architectural scales, 1) We identify feature collapse at initialization, which leads to redundancy, and propose a reduction scheme to mitigate this issue. 2) We quantify feature information using entropy and the minimum eigenvalue, demonstrating that these metrics serve as reliable indicators for generalization prediction. 3) We show that feature in the token space provides a more faithful representation than those in embedding space. 4) We discover an unexpected finding: features produced by linear submodules within ViT layers are critical for the prediction of generalization performance. 5) Our proposed proxy improves the correlation ranking by 18-48% over prior baselines and can effectively identify ViT architectures that achieve higher accuracy at lower or comparable computational cost.
Updated: 2026-10-09 15:22:48
标题: Vision Transformers中的表示结构解剖：一项严格的架构研究
摘要: Representation structure对于理解Vision Transformer (ViT)架构及其泛化行为至关重要。然而，先前的研究既未孤立分析模块级特征，也未调查它们的相互作用如何对性能估计有所贡献。在这项工作中，我们首次对不同架构规模下的特征信息进行了严格分析，从经验上揭示了ViT表示与泛化行为之间的关系，并利用这些见解指导高效的ViT设计。我们的贡献有五个方面：在不同的架构规模上，1)我们发现初始化时存在特征塌陷，导致冗余，并提出了一种减少方案来缓解这个问题。 2)我们使用熵和最小特征值量化特征信息，证明这些指标可以作为泛化预测的可靠指标。3)我们展示了token空间中的特征提供了比嵌入空间中更忠实的表示。 4)我们发现了一个意外的发现：ViT层内的线性子模块产生的特征对于泛化性能的预测至关重要。 5)我们提出的代理改进了18-48%的相关性排名，超过了先前的基线，并能有效地识别出在更低或可比的计算成本下实现更高准确度的ViT架构。
更新时间: 2026-10-09 15:22:48
领域: cs.CV,cs.LG
Amortized Counterfactual Prediction with Graph Neural Networks
Counterfactual inference asks how an observed outcome might have differed under different circumstances, such as whether a patient would have recovered had they received a different treatment. Such questions arise whenever decisions depend on the outcomes of alternatives that were not observed. Answering counterfactual queries generally requires knowledge of the underlying causal structure, which is often unavailable in practice. A common approach is therefore to first estimate the causal graph from observational data and then perform counterfactual inference. However, errors in the estimated causal graph can propagate to downstream counterfactual predictions. To address this challenge, we introduce the Counterfactual Graph Neural Process (CFGNP), an end-to-end approach that answers counterfactual queries directly from observational data without first estimating a causal graph. CFGNP uses Graph Neural Networks (GNNs) to exploit the dual-graph structure by restricting cross-world message passing to corresponding factual--counterfactual variables. Across synthetic and image-based benchmarks, CFGNP achieves stronger predictive performance than existing approaches, including on the German Loan benchmark, where it reduces MSE from $4.51 \times 10^{-2}$ to $0.19 \times 10^{-2}$. Furthermore, on synthetic causal graphs, a model trained only on 10-node graphs maintains strong predictive performance on graphs with up to 30 nodes, without retraining.
Updated: 2026-10-09 15:22:18
标题: 使用图神经网络进行摊销反事实预测
摘要: 反事实推断问的是在不同情况下观察到的结果可能会有何不同，比如，如果一个患者接受了不同的治疗，他们会否康复。这类问题在决策依赖于未观察到的替代方案的结果时经常出现。一般情况下回答反事实查询需要对潜在因果结构有所了解，但在实践中常常无法获取这种信息。因此，一个常见的方法是首先从观察数据中估计因果图，然后进行反事实推断。然而，估计的因果图中的错误可能会传播到下游的反事实预测。为了解决这一挑战，我们介绍了反事实图神经过程（CFGNP），这是一种端到端的方法，可以直接从观察数据中回答反事实查询而无需先估计因果图。CFGNP使用图神经网络（GNNs）利用双图结构，通过将跨世界消息传递限制在对应的实际-反事实变量之间。在合成和基于图像的基准测试中，CFGNP实现了比现有方法更强的预测性能，包括在德国贷款基准测试中，它将MSE从$4.51 \times 10^{-2}$减少到$0.19 \times 10^{-2}$。此外，在合成因果图上，一个只在10个节点图上训练的模型在最多30个节点的图上仍然保持着强大的预测性能，而无需重新训练。
更新时间: 2026-10-09 15:22:18
领域: cs.LG,cs.AI
GenCoord: Executable Commitments for Embodied LLM Agents under Private Information
Embodied language-model agents must coordinate interdependent actions while each holds private task information. In dialogue-based coordination, models generate messages and interpret replies to decide what each agent should do, spending inference before the team acts. We propose GenCoord, a decentralized protocol that makes the division of work itself the message. A proposal pairs the sender's skill path with its partner's continuation in Short DSL, a compact language for expressing joint work. Compatible paths form commitments linking the sender's delivery to the partner's work after verified receipt. We check fulfillment separately from original-task success. Complete requests also let a deterministic receiver proceed without another model call. In Minecraft, trained local 0.8B agents match the task success of controlled free-form communication while reducing median commitment time by 68.2%. The same-field full-key JSON control approaches free-form output length and delay, indicating that the amount generated matters even when messages are structured. With the same prompted DeepSeek model, both protocols complete nearly all tasks, but dialogue generates 2.8 times as many output tokens. Multistep commitments improve held-out-template success from 91.3% to 98.1% with fewer decisions. These results point to a design principle for embodied cooperation: make the division of work the message, and spend inference on choosing and revising the work.
Updated: 2026-10-09 15:20:03
标题: GenCoord：具体承诺在私人信息下的具体化LLM代理
摘要: 具身语言模型代理必须在每个持有私人任务信息的同时协调相互依赖的行动。在基于对话的协调中，模型生成消息并解释回复，以决定每个代理应该做什么，在团队行动之前花费推理。我们提出了GenCoord，一种分散化的协议，使工作分配本身成为消息。一项提议将发送者的技能路径与其伙伴在短DSL中的继续配对，这是一种用于表达共同工作的紧凑语言。兼容的路径形成承诺，将发送者的交付与伙伴收到后的工作联系在一起。我们将履行检查与原始任务成功分开。完整的请求还可以让确定性接收者在不进行另一个模型调用的情况下继续。在Minecraft中，经过训练的本地0.8B代理与受控的自由形式通信相比，在减少中位承诺时间68.2%的同时，实现了任务成功。相同领域的完整密钥JSON控制方法接近自由形式输出长度和延迟，表明生成的数量很重要，即使消息被结构化。使用相同提示的DeepSeek模型，两种协议几乎完成了所有任务，但对话生成的输出标记是后者的2.8倍。多步承诺将留存模板成功率从91.3%提高到98.1%，并减少了决策次数。这些结果指向了一项具身合作的设计原则：使工作分配成为消息，并将推理用于选择和修订工作。
更新时间: 2026-10-09 15:20:03
领域: cs.AI
Solving VeriContest with a Lean-Backed Rust Verifier
VeriContest is a benchmark of 1007 competitive-programming problems in Rust, each with a Verus specification, a judge-accepted solution, and a Verus proof. Its authors report that proof generation is the bottleneck for frontier models: given the specification and the code, the best model produces an accepted Verus proof for 13.95% of the problems on the first attempt. We report on solving the same proof-generation task with Rust-Prover, a verifier for Rust backed by Lean 4. The Verus specification and the Rust code are restated and translated into Lean, each specification becomes a theorem, and agents prove the theorems with Lean's kernel as the final check. All 1325 theorems of all 1007 problems were proved. 1259 of them were proved in one run of under 32 hours on Claude Opus 5.5, at a median of 3.2 minutes and $1.17 per proof, and 70% of them on the first iteration. The restated specifications were checked against the benchmark's test suites, and reviewed where no suite applies. None was wrong or weakened. The translated Lean programs were run on 21,413 of the benchmark's test cases and produced the same output as the Rust programs on every one. Across four Claude and four GPT models at five reasoning-effort settings, every current frontier model proves nearly all of a ten-theorem sample at every setting, and more effort raises the cost without raising the number of proofs. The cheapest Claude setting, Sonnet 5.5 at low effort, proves all of the 50 hardest theorems. We also rerun the benchmark's own Verus protocol with Claude Opus 5.5 on the 50 problems with the longest reference proofs. Opus 5.5 alone fails to prove one of them.
Updated: 2026-10-09 15:19:31
标题: 用支持Lean的Rust验证器解决VeriContest
摘要: VeriContest是一个由1007个竞争性编程问题组成的基准，使用Rust语言编写，每个问题都有一个Verus规范、一个经过评审接受的解决方案和一个Verus证明。其作者报告称，证明生成是前沿模型的瓶颈：在给定规范和代码的情况下，最好的模型在第一次尝试中为13.95%的问题生成了一个被接受的Verus证明。我们报告了使用Rust-Prover解决相同证明生成任务的情况，这是一个由Lean 4支持的Rust验证器。Verus规范和Rust代码被重新表述并翻译成Lean，每个规范都成为一个定理，代理人使用Lean的内核证明这些定理作为最终检查。所有1007个问题的1325个定理都得到了证明。其中1259个在Claude Opus 5.5的一个运行中在32小时内完成，中位数为3.2分钟，每个证明1.17美元，其中70%在第一次迭代中完成。重新表述的规范被检查并与基准测试套件进行比较，在没有适用套件的情况下进行审查。没有一个是错误的或被削弱的。翻译后的Lean程序在21413个基准测试用例上运行，并与每个测试用例上的Rust程序产生相同的输出。在五种推理努力设置下，每个当前的前沿模型在每个设置下几乎证明了所有的十个定理样本，更多的努力会提高成本而不会增加证明的数量。最便宜的Claude设置，Sonnet 5.5在低努力下，证明了所有50个最难的定理。我们还使用Claude Opus 5.5重新运行了基准的Verus协议，针对具有最长参考证明的50个问题。Opus 5.5独自一人未能证明其中一个问题。
更新时间: 2026-10-09 15:19:31
领域: cs.LO,cs.AI,cs.SE
MAGE: Multi-Agent Self-Evolution with Co-Evolutionary Knowledge Graphs
Self-evolving language-model agents must decide what to learn next and how to preserve what they have learned across iterations. Existing systems typically carry this cross-iteration knowledge as natural-language feedback, flat episodic memory, or implicit reinforcement signals, none of which cleanly supports a frozen weak backbone at inference time. This paper introduces MAGE (Multi-Agent Graph-guided Evolution), a framework that externalizes self-knowledge into a four-subgraph co-evolutionary knowledge graph. Its experience subgraph stores both teacher-written failure corrections and the learner's own past correct reasoning traces, which are retrieved as task-conditioned guidance for a frozen execution model. During evolution, the graph, a task-level search bandit, and a skill-level routing bandit are updated from the same reward stream, while the learner's backbone remains unchanged. We further provide structural analysis showing how append-only memory growth, bounded curriculum coverage, and task-filtered retrieval together support stable improvement of the retrieval substrate for frozen-learner evolution. Across nine benchmarks spanning mathematical reasoning, multi-hop and open-domain question answering, spatio-temporal analysis, financial numerical reasoning, medical multiple-choice, an open-world survival game, and web navigation, MAGE achieves strong performance against prompt-based frozen-backbone baselines. Ablations show that self-harvested success traces and teacher-written corrections are complementary, with success memories contributing most on reasoning-template-heavy tasks and corrective memories supporting harder composition and interaction settings. Code is available in https://github.com/RuiyiYang01/mage-alg
Updated: 2026-10-09 15:19:08
标题: MAGE：具有共同进化知识图的多智能体自我进化
摘要: 自我进化的语言模型代理必须决定下一步学习什么以及如何在迭代中保留已学到的知识。现有系统通常将跨迭代知识保存为自然语言反馈、扁平的记忆或隐式强化信号，但这些方式都不能清晰地支持推理时的冻结弱骨干。本文介绍了MAGE（多代理图引导进化）框架，它将自知识外化为一个包含四个子图的共同进化知识图。它的经验子图存储了教师编写的错误校正以及学习者自己过去的正确推理痕迹，这些痕迹根据任务条件作为冻结执行模型的指导被检索出来。在进化过程中，图、任务级搜索赌博机和技能级路由赌博机从相同的奖励流中更新，而学习者的骨干保持不变。我们进一步提供了结构分析，展示了仅追加式记忆增长、有限课程覆盖和任务过滤检索如何共同支持冻结学习者进化的稳定改善。在跨数学推理、多跳和开放领域问答、时空分析、金融数值推理、医学多项选择、开放世界生存游戏和网页导航等九项基准测试中，MAGE针对基于提示的冻结骨干基线取得了强大性能。消融实验表明，自收获的成功痕迹和教师编写的校正是互补的，成功记忆在推理模板密集的任务中起到了最大作用，而校正记忆支持更难的组合和交互设置。代码可在https://github.com/RuiyiYang01/mage-alg中找到。
更新时间: 2026-10-09 15:19:08
领域: cs.AI
Machine Unlearning for Federated Socially Fair k-Means Clustering
Federated clustering can support applications such as patient stratification across hospitals and customer segmentation across distributed financial datasets. Responsible AI calls for fair representation of demographic groups and support for data deletion, motivated in part by rights such as the GDPR's right to erasure. These goals interact: deleting records can change the balance between groups and alter the clustering of the remaining population. We study exact machine unlearning for federated socially fair $k$-means, whose objective is to minimize the largest group-average squared distance to the nearest center. Our approach keeps separate local summaries for each group, so summaries at unaffected institutions remain valid as global fairness weights change. For two groups, we prove an initialization approximation guarantee independent of client group proportions under ideal sampling, and we show that same-seed replay reproduces retraining exactly. Reuse pays off most when a whole client withdraws: across three datasets, replay is $35.8\times$ faster than retraining for the initialization-only model and remains $2.4\times$ and $1.7\times$ faster with one and two fair-refinement rounds, which improve worst-group cost. For individual record deletions, gains are modest (about $1.06\times$), and certifying cached assignments can cost more than it saves. These results map where exact unlearning can profitably reuse computation. Code and reproducibility materials are available at https://github.com/arareddy/mufsc.
Updated: 2026-10-09 15:16:48
标题: Machine Unlearning for Federated Socially Fair k-Means Clustering 机器去学习用于联合社会公平k均值聚类
摘要: 联邦聚类可以支持跨医院的患者分层和跨分布式金融数据集的客户分割等应用。负责任的人工智能要求公平地代表人口群体，并支持数据删除，部分受到《通用数据保护条例》中的消除权的启发。这些目标相互影响：删除记录可能改变各组之间的平衡，并改变剩余人口的聚类。我们研究了联邦社会公平$k$-means的精确机器遗忘，其目标是最小化距离最近中心的最大组平均平方距离。我们的方法为每个组保留单独的本地摘要，因此未受影响的机构的摘要在全局公平权重发生变化时仍然有效。对于两个组，我们证明了在理想抽样下独立于客户组比例的初始化近似保证，并展示了相同种子回放可以精确复制重新训练。当整个客户端退出时，重用效益最大：在三个数据集中，回放对于仅初始化模型比重新训练快$35.8$倍，并且在一个和两个公平细化轮中仍然比重新训练快$2.4$倍和$1.7$倍，这些轮改进了最坏组的成本。对于单个记录删除，收益较小（约$1.06$倍），而验证缓存的分配可能比节省的成本更高。这些结果展示了精确遗忘可以有利地重用计算的情况。代码和可重现材料可在https://github.com/arareddy/mufsc找到。
更新时间: 2026-10-09 15:16:48
领域: cs.LG,cs.DS
DeltaLLM: Delta-Key Sparse Attention for Edge LLMs
On-device LLM inference is increasingly important for edge platforms but remains challenging due to high computational and memory demands and tight resource constraints. Modern applications often involve long prompts. Although compact models reduce linear-layer costs through weight compression, attention remains largely unoptimized and becomes the dominant bottleneck, especially during prefilling. Existing attention pruning methods primarily target highly parallel hardware (e.g., GPUs) and long context windows, limiting their suitability for edge scenarios. We present DeltaLLM, a training-free LLM framework based on algorithm-hardware-dataflow co-design to accelerate prefilling attention. DeltaLLM introduces a causal-attention-aware K-only delta formulation that enables Delta-Key reuse across query heads under Grouped-Query Attention (GQA). A context-aware hybrid strategy preserves exact attention near the diagonal while applying delta approximation to nonlocal regions. For score-value multiplication, we propose a vectorized structured sparse online softmax (VSSOS) that converts fine-grained spatial score sparsity into hardware-friendly column-level skipping. A specialized vector-processor architecture and unified sparse/dense attention dataflow exploit the resulting sparsity. Evaluated on BitNet-b1.58-2B-4T and LLaMA3.2-1B-Instruct, DeltaLLM achieves up to 60% Delta-Key sparsity and 63% structured attention-score sparsity with negligible average accuracy loss. On the KV260 FPGA board, it delivers a 1.62x speedup over TeLLMe-v2 while reducing DSP and BRAM usage. These results show that DeltaLLM supports efficient edge LLM deployment and integration with existing pipelines without fine-tuning.
Updated: 2026-10-09 15:10:32
标题: DeltaLLM: 边缘LLM的Delta-Key稀疏注意力
摘要: 设备端LLM推理在边缘平台上变得越来越重要，但由于高计算和内存需求以及严格的资源限制而仍然具有挑战性。现代应用程序通常涉及长提示。尽管紧凑模型通过权重压缩减少了线性层的成本，但注意力仍然主要未经优化，并在预填充期间成为主要瓶颈，特别是在预填充期间。现有的注意力修剪方法主要针对高度并行的硬件（例如GPU）和长上下文窗口，限制了它们在边缘场景中的适用性。我们提出了DeltaLLM，这是一种基于算法-硬件-数据流共同设计的无需训练的LLM框架，用于加速预填充注意力。DeltaLLM引入了一种因果-注意力感知的K-only delta公式，使得Delta-Key在组合查询注意力（GQA）下能够在查询头之间重复使用。一种上下文感知的混合策略在对角线附近保留精确的注意力，同时在非局部区域应用delta近似。对于得分值乘法，我们提出了一种向量化结构化稀疏在线softmax（VSSOS），将细粒度空间得分稀疏性转化为硬件友好的列级跳过。专门设计的矢量处理器架构和统一的稀疏/密集注意力数据流利用了所产生的稀疏性。在BitNet-b1.58-2B-4T和LLaMA3.2-1B-Instruct上进行评估，DeltaLLM实现了高达60%的Delta-Key稀疏性和63%的结构化注意得分稀疏性，平均精度损失可以忽略不计。在KV260 FPGA板上，它比TeLLMe-v2实现了1.62倍的加速，并减少了DSP和BRAM的使用。这些结果表明DeltaLLM支持高效的边缘LLM部署，并与现有管道集成而无需微调。
更新时间: 2026-10-09 15:10:32
领域: cs.AI,eess.SP
Towards Understanding the Training of GFlowNets: A Theoretical Perspective
Generative Flow Networks (GFlowNets) learn policies that sample discrete objects in proportion to a positive reward. They are trained by minimizing balance objectives, whose exact satisfaction guarantees correct sampling. However, a comprehensive theory linking nonzero balance loss and approximate stationarity to sampling accuracy, especially under evolving sampling distributions and backward policies, is still lacking. To bridge this gap, we develop a unified theory for Flow Matching, Detailed Balance, and Trajectory Balance in this regime. We first bound terminal symmetric KL by the square root of the expected balance loss. Then, we show that approximate stationarity controls this distribution error up to a tangent-space residual measuring parameterization incompleteness. Finally, we establish finite-time convergence under fixed or controlled sampling and fixed, balance-trained, or separately updated backward policies. Under standard smoothness assumptions, with fixed or controlled sampling and a fixed or balance-trained backward policy, the expected loss is $O(\barε^2+T^{-1})$ and expected terminal symmetric KL is $O(\barε+T^{-1/2})$ under uniform coverage. For a separately updated backward policy, we characterize the sufficient update decay to approach the residual neighborhood and preserve these rates. These results connect balance objectives to sampling accuracy and clarify how parameterization, coverage, and moving components govern GFlowNet training.
Updated: 2026-10-09 15:10:02
标题: 朝着理解GFlowNets的训练：一个理论视角
摘要: Generative Flow Networks（GFlowNets）学习策略，根据正奖励比例采样离散对象。它们通过最小化平衡目标进行训练，确保正确采样。然而，尚缺乏将非零平衡损失和近似稳态与采样精度联系起来的综合理论，特别是在演变的采样分布和反向策略下。为了弥补这一差距，我们在这一领域开发了流匹配、详细平衡和轨迹平衡的统一理论。我们首先通过期望平衡损失的平方根来限制终端对称KL。然后，我们展示近似稳态控制这种分布误差，直到切线空间残差测量参数化不完整性。最后，我们在固定或受控采样和固定、平衡训练或单独更新的反向策略下建立有限时间收敛。在标准光滑性假设下，采用固定或受控采样和固定、平衡训练的反向策略，期望损失为$O(\barε^2+T^{-1})$，在均匀覆盖下期望终端对称KL为$O(\barε+T^{-1/2})$。对于单独更新的反向策略，我们描述了足够的更新衰减以接近残余邻域并保持这些速率。这些结果将平衡目标与采样精度联系起来，澄清了参数化、覆盖和移动组件如何影响GFlowNet训练。
更新时间: 2026-10-09 15:10:02
领域: cs.LG
Fast localized urban microclimate emulator: wind and temperatures in unseen urban morphologies
Urban microclimate, encompassing wind and temperature fields shaped by building geometry, strongly affects building energy use, pedestrian wind comfort, pollutant dispersion, the urban heat island, and public health. Conventional Computational Fluid Dynamics (CFD) is computationally prohibitive for rapid assessments, while many deep learning approaches require extensive training data and struggle with generalization to unseen configurations. In this work, we present the Fast Localized Urban Microclimate Emulator (FLUME), a rapid, data-efficient, and geometry-aware framework based on a Fourier neural operator for predicting three-dimensional wind and temperature fields. FLUME assumes that, under prescribed inflow and thermal forcing, much of the spatial variation of these mean fields is controlled by surrounding building geometry. The framework introduces a Multi-Directional Distance Feature (MDDF), which captures geometry over adaptive, visibility-defined spatial scales. By decomposing simulations of only 23 urban blocks into smaller 3D patches enriched with precomputed MDDF information, FLUME enables learning from limited CFD data. The model achieves mean absolute errors of 0.2 m/s for streamwise velocity and 0.19$^{\circ}$C for temperature on unseen urban configurations. A deep ensemble provides a practical proxy for epistemic uncertainty, and conformal prediction transforms these estimates into calibrated prediction intervals. Ensemble spread is positively related to local prediction error, with greater uncertainty in roof-level shear layers, building wakes, and vertical velocity near the ground. These results indicate FLUME's potential for rapid scenario assessment in urban design and urban climate studies.
Updated: 2026-10-09 15:06:11
标题: 快速本地化城市微气候仿真器：未曾见过的城市形态中的风和温度
摘要: 城市微气候，包括由建筑几何形状塑造的风速和温度场，强烈影响建筑能源利用，行人风舒适度，污染物扩散，城市热岛和公共健康。传统的计算流体动力学（CFD）对于快速评估来说计算量过大，而许多深度学习方法需要大量的训练数据，并且在未见过的配置上泛化困难。在这项工作中，我们提出了快速局部城市微气候模拟器（FLUME），这是一个快速，数据高效且基于傅里叶神经算子的框架，用于预测三维风速和温度场。FLUME假设，在规定的入流和热力驱动下，这些平均场的空间变化很大程度上受到周围建筑几何形状的控制。该框架引入了一个多方向距离特征（MDDF），它在自适应的、可见性定义的空间尺度上捕获几何形状。通过将仅包含23个城市街区的模拟分解为更小的3D块，并丰富预先计算的MDDF信息，FLUME能够从有限的CFD数据中学习。该模型在未见过的城市配置上实现了流向速度0.2 m/s和温度0.19°C的平均绝对误差。深度集合提供了一个实用的表征认知不确定性的代理，而符合性预测将这些估计转化为校准的预测区间。集合的扩散与局部预测误差呈正相关，在屋顶层剪切层、建筑尾流和地面附近垂直速度处的不确定性更大。这些结果表明FLUME在城市设计和城市气候研究中快速场景评估方面具有潜力。
更新时间: 2026-10-09 15:06:11
领域: physics.flu-dyn,cs.LG
Adaptive DHiSparse: Speculative Decoding over a Load-Adaptive KV Hierarchy
Long-context serving runs out of KV memory before the hardware reaches its useful throughput. HiSparse stores history in host memory and fetches only the selected entries, but it uses one cache budget at every load and does not handle draft-model speculative decoding such as DSpark and DFlash2. We present Adaptive DHiSparse, which supports speculative decoding over hierarchical sparse KV and adapts residency to load. It protects the required working set, retains additional history while GPU memory is spare, and recovers the rest through exact fused retrieval. DSpark and DFlash2 keep their proposal and acceptance rules: queries share missing-row retrieval, draft KV uses a window-bounded ring, and verification staging is reused across layer groups. At 100K context and concurrency 128, GLM-5.3 with DFlash2 delivers 793 output tokens/s per decode GPU, 1.45$\times$ HiSparse and 1.27$\times$ non-speculative adaptive serving. Held to the same latency bounds as HiSparse, it sustains up to 4$\times$ the concurrency and 2.3 to 3.7$\times$ the output, depending on the per-token bound. On DeepSeek-V4.1-Flash at 500K context and concurrency 512, Adaptive DHiSparse with DSpark delivers 1.85$\times$ the output of resident serving with DSpark. In a load step, optional residency reduces host fetches from 14.2% to 1.4% of selected rows at low load and releases memory under pressure.
Updated: 2026-10-09 15:02:35
标题: 自适应DHiSparse：适应负载的KV层次结构上的推测解码
摘要: 长上下文服务在硬件达到其有效吞吐量之前就用尽了KV内存。HiSparse将历史存储在主机内存中，并仅获取所选条目，但在每次加载时使用一个缓存预算，并且不处理像DSpark和DFlash2这样的草稿模型的推测解码。我们提出了自适应DHiSparse，它支持在分层稀疏KV上进行推测解码，并根据负载调整常驻性。它保护所需的工作集，在GPU内存空闲时保留额外的历史，并通过精确的融合检索恢复其余部分。DSpark和DFlash2保持其提议和接受规则：查询共享缺失行检索，草稿KV使用窗口有界环，验证分段在层组间重复使用。在100K上下文和并发128的情况下，带有DFlash2的GLM-5.3每个解码GPU提供793个输出标记/秒，是HiSparse的1.45倍和非推测性自适应服务的1.27倍。在与HiSparse相同的延迟限制下，它支持高达4倍的并发性和2.3至3.7倍的输出，取决于每个标记的限制。在DeepSeek-V4.1-Flash上下文和并发512为500K的情况下，具有DSpark的自适应DHiSparse提供了比具有DSpark的驻留服务多1.85倍的输出。在负载步骤中，可选常驻性将主机获取从低负载时的14.2%减少到所选行的1.4%，并在内存受压力时释放内存。
更新时间: 2026-10-09 15:02:35
领域: cs.AI,cs.CL,cs.DC,cs.LG
Sampling-Based Safe Reinforcement Learning
Safe exploration remains a fundamental challenge in reinforcement learning (RL), limiting the deployment of RL agents in the real world. We propose Sampling-Based Safe Reinforcement Learning (SBSRL), a model-based RL algorithm that maintains safety throughout the learning process by enforcing constraints jointly across a finite set of dynamics samples. This formulation approximates an intractable worst-case optimization over uncertain dynamics and enables practical safety guarantees in continuous domains. We further introduce an exploration strategy based on constraining epistemic uncertainty, eliminating the need for explicit exploration bonuses. Under regularity conditions, we derive high-probability guarantees of safety throughout learning and a finite-time sample complexity bound for recovering a near-optimal policy. Empirically, SBSRL achieves safe and efficient exploration both in simulation and in real robotic hardware, and readily extends to practical deep-ensemble implementations that scale to high-dimensional continuous control problems.
Updated: 2026-10-09 15:01:32
标题: 基于采样的安全强化学习
摘要: 安全探索仍然是强化学习（RL）中的一个基本挑战，限制了RL代理在现实世界中的部署。我们提出了一种基于采样的安全强化学习（SBSRL）算法，这是一种基于模型的RL算法，通过在一个有限的动力学样本集合上联合施加约束来保持学习过程中的安全性。这种表达形式近似了一个不可解的关于不确定动力学的最坏情况优化，并且在连续领域中实现了实用的安全保证。我们进一步介绍了一种基于约束认知不确定性的探索策略，消除了对显式探索奖励的需求。在正则条件下，我们推导出了在学习过程中安全性的高概率保证以及恢复接近最优策略的有限时间样本复杂度界限。在实证方面，SBSRL在仿真和真实机器人硬件上都实现了安全而有效的探索，并且可以轻松扩展到能够应对高维连续控制问题的实际深度集成实现。
更新时间: 2026-10-09 15:01:32
领域: cs.LG,cs.AI,cs.RO
ThinkNet: Evolutionary Deep Ensembles for Efficient Motor-Imagery EEG Classification
Practical assistive and rehabilitative brain--computer interfaces require subject-independent motor-imagery EEG (MI-EEG) decoders that generalize to new users under limited target-user data and constrained compute. However, held-out-subject performance can be overstated when test-subject information influences preprocessing, model selection, or ensemble selection. We present \textit{ThinkNet}, a validation-controlled framework that combines train-only normalization, validation-guided evolutionary search, and validation-gated inference to identify compact decoders and inference policies for held-out subjects. We evaluate four-class BCI Competition IV-2a (session T) decoding with nine Leave-One-Subject-Out (LOSO) folds, three seeds, seven fixed decoder entries, and a broader search over ten representative decoder families; the held-out subject is never used for normalization, hyperparameter, architecture, or ensemble-policy selection. In the fixed benchmark, the validation-selected compact decoder achieved 44.35$\pm$15.41\% accuracy with 4.9K parameters, 19 KB FP32 weights, and 0.99 ms batch-1 Orin CUDA inference. Across the broader search, compact models ($\leq$25K parameters) achieved higher mean held-out accuracy than mid-size and large alternatives after selected retraining (40.10\% vs. 35.09\% and 34.78\%). Validation-gated ensembling improved over validation-selected single-model inference, reaching 43.98$\pm$16.25\% in the fixed benchmark and 43.31$\pm$15.88\% for the compact six-family ensemble. A non-deployable oracle analysis revealed a 6.1-point family-selection gap and near-zero validation--test correlation, showing that validation reliability remains a key bottleneck under subject shift. Thus, ThinkNet is a validation-controlled framework for compact MI-EEG model and inference-policy selection, rather than a single-architecture benchmark.
Updated: 2026-10-09 15:00:12
标题: ThinkNet：用于高效运动想象脑电图分类的进化深度集成模型
摘要: 实用的辅助和康复性脑-计算机接口需要独立于受试者的运动想象脑电图（MI-EEG）解码器，这些解码器可以在有限的目标用户数据和受限的计算资源下适用于新用户。然而，当测试受试者信息影响预处理、模型选择或集成选择时，留存受试者的性能可能会被夸大。我们提出了ThinkNet，这是一个经过验证控制的框架，它结合了仅训练时标准化、验证引导的进化搜索和验证门控推断，以识别适用于留存受试者的紧凑解码器和推断策略。我们使用九个留一受试者交叉验证、三个种子、七个固定解码器条目和对十个代表性解码器家族进行更广泛搜索的方法来评估四类BCI竞赛IV-2a（T会话）的解码；留存受试者永远不会用于标准化、超参数、架构或集成策略选择。在固定基准中，验证选择的紧凑解码器在4.9K参数、19 KB FP32权重和0.99毫秒的CUDA推断下实现了44.35±15.41%的准确率。在更广泛的搜索中，紧凑模型（≤25K参数）在经过选择性重训练后实现了比中等和大型替代品更高的平均留存准确率（40.10% vs. 35.09%和34.78%）。验证门控集成优于验证选择的单模型推断，在固定基准中达到了43.98±16.25%的准确率，而在紧凑的六族集成中达到了43.31±15.88%。一个不可部署的预测分析显示了一个6.1分的家族选择差距和几乎为零的验证-测试相关性，表明验证的可靠性仍然是受主体转移的关键瓶颈。因此，ThinkNet是一个用于紧凑MI-EEG模型和推断策略选择的经过验证控制的框架，而不是一个单一架构基准。
更新时间: 2026-10-09 15:00:12
领域: cs.HC,cs.AI,cs.LG,eess.SP
A Baseline Assessment for the Agentic Web Technology Stack
The fast pace in which protocols, frameworks, and technologies are being proposed for LLM-based agents results in a complex scenario for anyone aiming to create state-of-the-art AI solutions. Without an assessment on how different protocols, frameworks, and libraries work together to compose an Agentic Web solution, developers and organizations face different challenges (e.g., skilling, governance, infrastructure) even before becoming part of the Agentic Web ecosystem. In this context, we propose a system for performing Agentic Web evaluations called MASTIF (Multi-Agent System TestIng Framework) and present a baseline assessment of Agentic Web technologies, including 3 LLMs, 3 protocols, and 4 agent implementation frameworks, against 300 Mind2Web benchmark tasks. Results suggest that selecting a task-specific technology stack may result in gains of task completion rate, efficiency score, and social value stability scores; and reductions in verbosity and time to complete tasks. With the presented results, we expect to present a baseline in terms of metrics and results for this complex and fast evolving scenario for those aiming at designing, assessing, and developing multi-agent systems.
Updated: 2026-10-09 14:59:24
标题: 《主动式网络技术堆栈的基准评估》
摘要: 随着针对基于LLM的代理提出了各种协议、框架和技术，创造最先进的人工智能解决方案变得复杂。没有对不同协议、框架和库如何共同组成代理网络解决方案进行评估，开发人员和组织在成为代理网络生态系统的一部分之前就面临不同的挑战（如技能、治理、基础设施等）。在这种情况下，我们提出了一个用于执行代理网络评估的系统，称为MASTIF（多智能体系统测试框架），并针对300个Mind2Web基准任务对代理网络技术进行了基线评估，包括3种LLM、3种协议和4种代理实现框架。结果表明，选择特定任务的技术堆栈可能会提高任务完成率、效率得分和社会价值稳定性得分；并减少冗余度和完成任务所需时间。通过提供的结果，我们希望为那些致力于设计、评估和开发多智能体系统的人提供一个基线，以便在这个复杂且快速发展的场景中提供指标和结果。
更新时间: 2026-10-09 14:59:24
领域: cs.AI,cs.ET,cs.SE
TelcoAgent: A Scalable 5G Multi-KPM Forecasting With 3GPP-Grounded Explainability
Key Performance Measurement (KPM) forecasting is essential for proactive network management of 5G and next-generation telecom networks. However, existing machine learning (ML) approaches face significant limitations in scalability and explainability, restricting their effectiveness in real-world deployments. We propose TelcoAgent, a foundation model-based framework that enables accurate, scalable, and explainable forecasting of multiple KPMs across diverse network cells without the need for site-specific training. Specifically, the framework comprises three key components: (i) an automated three-agent pipeline that constructs a 3rd Generation Partnership Project (3GPP) knowledge graph directly from specification documents, (ii) a scalable, time-series foundation model (TSFM)-based prediction pipeline to deliver accurate, zero-shot forecasting, and finally (iii) a reasoning and explanation pipeline that provides actionable, domain-grounded diagnostics. Evaluated using a 3-month, real-world, city-scale 5G KPM dataset from a U.S.-based network operator, TelcoAgent demonstrates high forecasting accuracy for all 7 considered KPMs per cell across 200 cells, while delivering explainable insights and actionable instructions to address network degradations.
Updated: 2026-10-09 14:55:19
标题: TelcoAgent: 一个可扩展的基于5G的多关键性能指标预测模型，具有3GPP基础的可解释性
摘要: 关键绩效测量（KPM）预测对于主动管理5G和下一代电信网络至关重要。然而，现有的机器学习（ML）方法在可扩展性和可解释性方面存在显着局限，限制了它们在现实部署中的有效性。我们提出了TelcoAgent，这是一个基于基础模型的框架，可以实现准确、可扩展和可解释的多个KPM在不同网络单元之间的预测，而无需进行现场特定的训练。具体而言，该框架包括三个关键组件：（i）一个自动化的三代伙伴项目（3GPP）知识图构建管道，直接从规范文档中构建知识图；（ii）一个可扩展的、基于时间序列基础模型（TSFM）的预测管道，用于提供准确的零射预测；最后（iii）一个推理和解释管道，提供可操作的、领域基础的诊断信息。使用来自美国网络运营商的一个实际、为期3个月的城市规模的5G KPM数据集进行评估，TelcoAgent展示了对于每个单元的7个考虑的KPM的高预测准确性，跨200个单元，同时提供可解释的见解和可操作的说明，以解决网络退化问题。
更新时间: 2026-10-09 14:55:19
领域: cs.AI,cs.LG
Calibration in One Go: Jointly Learning Accurate and Calibrated Neural Networks
Neural networks (NNs), while highly flexible and versatile, are often poorly calibrated. Post-hoc calibration methods adjust the outputs of an already trained model using a separate calibration dataset and additional calibration parameters. This process adds overhead that requires separate optimization of the calibrator and subsequent model selection. In this paper, we show that standard (post-hoc) calibrators can be trained end-to-end with the backbone via a bilevel Stackelberg formulation. This formulation eliminates the need for a separate calibration set or a dedicated calibration loss by jointly optimizing accuracy and calibration using cross-entropy, introducing only one additional hyperparameter: the calibrator learning rate. Although our framework supports calibrators ranging from temperature and matrix scaling to NN layers, our results show that using the final classification layer itself as the calibrator yields accurate, well-calibrated models with no additional parameters. Our approach outperforms SOTA calibrators with minimal added complexity, while also avoiding the multiplicative overhead of separate calibration components in settings such as multi-task learning.
Updated: 2026-10-09 14:54:35
标题: 一次性校准：联合学习准确和校准的神经网络
摘要: 神经网络（NNs）虽然具有高度灵活和多功能性，但通常校准效果不佳。事后校准方法通过使用单独的校准数据集和额外的校准参数调整已经训练好的模型的输出。这个过程增加了额外开销，需要单独优化校准器并进行后续模型选择。在本文中，我们展示了标准（事后）校准器可以通过双层斯塔克尔贝格公式与主干网络一起端对端训练。这种公式通过交叉熵联合优化准确性和校准，消除了对单独校准集或专用校准损失的需求，仅引入了一个额外的超参数：校准器学习率。尽管我们的框架支持校准器的范围从温度和矩阵缩放到神经网络层，但我们的结果表明，将最终分类层本身用作校准器可以产生准确、校准良好的模型，而无需额外参数。我们的方法在最小增加复杂性的情况下优于最先进的校准器，并避免了在多任务学习等设置中单独校准组件的乘法开销。
更新时间: 2026-10-09 14:54:35
领域: cs.LG,cs.AI,stat.ML
Variance-Reduced Policy Gradient is Nearly Minimax-Optimal
Standard Policy Gradient (PG) algorithms suffer from high estimation variance because they rely on Monte Carlo updates. While off-policy variance reduction mitigates this problem, establishing optimal sample complexity currently requires either restrictive assumptions, such as uniformly bounded importance weights, or impractical second-order operations. In this work, we introduce Loopless Variance-Reduced PG (LVR-PG), a unified theoretical framework that abstracts a class of first-order estimators that integrate recursive updates with defensive importance sampling, including known estimators such as PAGE-PG and STORM-PG. We prove that this algorithmic class achieves $\tilde{\mathcal{O}}(ε^{-2})$ last-iterate global convergence under weak gradient domination for general parametric policies. We establish this guarantee using purely first-order updates, with minimal assumptions and without gradient preconditioners. By extending state-of-the-art lower bounds to continuous parametric policies, we establish an $Ω(ε^{-2})$ sample complexity limit, confirming our framework is minimax-optimal up to logarithmic factors and horizon terms.
Updated: 2026-10-09 14:54:26
标题: 方差减少的策略梯度几乎达到极小极值点
摘要: 标准策略梯度（PG）算法受到高估计方差的困扰，因为它们依赖于蒙特卡洛更新。虽然离线策略的方差减少可以缓解这个问题，但建立最优样本复杂度目前要求限制性假设，比如均匀有界的重要性权重，或者不切实际的二阶操作。在这项工作中，我们介绍了无循环方差减少PG（LVR-PG），这是一个统一的理论框架，抽象出一类第一阶估计器，它们将递归更新与防御性重要性抽样结合起来，包括已知的估计器，如PAGE-PG和STORM-PG。我们证明了这个算法类在一般参数化策略的弱梯度支配下实现了$\tilde{\mathcal{O}}(ε^{-2})$最优迭代全局收敛。我们使用纯粹的第一阶更新来建立这个保证，没有很多假设，也没有梯度预处理器。通过将最新的下界扩展到连续参数化策略，我们建立了一个$Ω(ε^{-2})$的样本复杂度限制，证实我们的框架在对数因子和时间步长上是极小最优的。
更新时间: 2026-10-09 14:54:26
领域: cs.LG,cs.AI
Transformer-based In-Context Score Operator Learning for Diffusion Models
Diffusion models typically learn the generative dynamics of a single target distribution or a family of distributions indexed by conditioning variables fixed during training. This limits their ability to adapt to a previously unseen distribution from only a small set of examples. We introduce an in-context solution operator learning framework for diffusion models, in which each target distribution is treated as a task drawn from a meta-distribution and is specified at inference time by an empirical context set. Our method learns an in-context score operator that maps the context samples, a noisy query point, and the diffusion time to the task-dependent score, enabling reverse-diffusion sampling on unseen tasks without retraining or parameter updates. A key observation is that the task-dependent score admits a Gaussian-kernel conditional-mean representation. Its empirical Nadaraya--Watson approximation can be represented exactly by a model in our constructed transformer class, providing a principled foundation for the proposed query-dependent attention architecture. The operator is trained through amortized denoising score matching across task distributions. We establish generalization-error bounds that quantify the effects of the context size, the numbers of training tasks and within-task samples, model complexity, and intrinsic and ambient dimensions. Numerically, we evaluate the method on a range of scientific and visual datasets, including Navier-Stokes, ERA5 climate data, QuickDraw sketches, and Food-101. The results show that the proposed method performs well on relatively simple task families, while its performance degrades as task complexity increases, thereby illustrating both the promise and the limitations of in-context learning for diffusion models.
Updated: 2026-10-09 14:53:49
标题: 基于Transformer的上下文得分操作符学习用于扩散模型
摘要: 扩散模型通常学习单个目标分布的生成动态，或者在训练过程中固定的条件变量索引的一组分布。这限制了它们适应仅通过一小组示例看不见的分布的能力。我们引入了一个上下文解决方案操作学习框架，用于扩散模型，在这个框架中，每个目标分布被视为从一个元分布中抽取的任务，并且在推断时由一个经验上下文集指定。我们的方法学习了一个上下文分数操作器，它将上下文样本、一个嘈杂的查询点和扩散时间映射到任务相关的分数，从而实现了对未见任务的逆扩散采样，而无需重新训练或参数更新。一个关键观察是，任务相关的分数可以通过高斯核条件均值表示。它的经验Nadaraya-Watson逼近可以通过我们构建的变压器类模型精确表示，为所提出的查询相关注意力架构提供了原则性基础。操作员通过跨任务分布的摊销去噪分数匹配进行训练。我们建立了量化上下文大小、训练任务数量和任务内样本数量、模型复杂度以及内在和环境维度影响的泛化误差边界。在一系列科学和视觉数据集上对该方法进行了数值评估，包括Navier-Stokes、ERA5气候数据、QuickDraw素描和Food-101。结果表明，所提出的方法在相对简单的任务系列上表现良好，但随着任务复杂性的增加，其性能下降，从而展示了上下文学习对于扩散模型的潜力和局限性。
更新时间: 2026-10-09 14:53:49
领域: cs.LG
VQIL: Visible Quantum Interference Learning for Perturbation Response Ranking on Graphs
Learning how local perturbations affect a network is important for identifying which components drive or absorb a system's response. We formulate this as seed-conditioned response ranking: given a perturbation at one seed node, ranking all other nodes by response strength. This is challenging when the model is given only the interaction topology, while the mechanism governing how perturbations propagate must be learned from data. We present VQIL (Visible Quantum Interference Learning), a trainable Szegedy quantum walk for seed-conditioned ranking. Edge weights, link phases, and per-node phase rotations are optimized end-to-end by backpropagation while the walk operator remains unitary throughout training. Node scores are obtained from a geometrically discounted coherent sum of quantum-walk amplitudes, enabling quantum interference before the final squared-norm readout. Because each learned parameter is attached to a named edge or node, the trained model provides directly inspectable hypotheses about the roles of individual graph components in the propagation mechanism. On AC power-flow perturbations over six transmission grids, with electrical parameters withheld from the model, VQIL improves Spearman rank correlation from 0.147 for the untrained walk to 0.831 and outperforms an identically trained random-walk model. On gene perturbation over five protein-protein interaction networks with Perturb-seq supervision, VQIL improves rank correlation from 0.171 to 0.620, outperforming the strongest learned classical and neural baselines. These results show that trainable quantum-walk interference provides a scalable and interpretable inductive bias for learning seed-conditioned response rankings on graphs.
Updated: 2026-10-09 14:53:35
标题: VQIL：图上可见的量子干涉学习用于扰动响应排名
摘要: 学习如何局部扰动影响网络对于确定哪些组件驱动或吸收系统响应至关重要。我们将这个问题形式化为种子条件的响应排名：给定一个种子节点的扰动，通过响应强度对所有其他节点进行排名。当模型仅给定交互拓扑时，这是具有挑战性的，而控制扰动传播的机制必须从数据中学习。我们提出了VQIL（可见量子干涉学习），一种可训练的Szegedy量子行走用于种子条件排名。边权重、链接相位和每个节点的相位旋转通过反向传播进行端到端优化，而行走算子在整个训练过程中保持酉。节点得分是通过量子行走振幅的几何折扣和谐求和获得的，从而实现了在最终的平方范数读数之前的量子干涉。由于每个学习参数都与命名的边或节点相关联，训练模型直接提供了关于个别图组件在传播机制中的作用的可检验假设。在六个输电网上进行交流电力流扰动，电气参数被模型隐瞒，VQIL将斯皮尔曼等级相关性从未训练的行走的0.147提高到0.831，并优于一个经过完全相同训练的随机行走模型。在五个蛋白质相互作用网络上进行基因扰动，并使用Perturb-seq监督，VQIL将排名相关性从0.171提高到0.620，优于最强的学习经典和神经基线。这些结果表明，可训练的量子行走干涉为学习图上种子条件响应排名提供了可扩展和可解释的归纳偏见。
更新时间: 2026-10-09 14:53:35
领域: cs.LG
X-AVDD: Cross-Attentive Audio-Visual Dataset Distillation
Audio-visual dataset distillation aims to synthesize a small dataset from a large paired audio-visual dataset, while nearly preserving the training utility for neural networks. Existing approaches such as DM and AVDD face two critical limitations: 1) they are memory-intensive, making them difficult to scale to complex architectures and higher samples-per-class (SPC) settings, and 2) they often fail to preserve strong cross-modal semantic correspondence. To tackle these issues, we propose X-AVDD, a cross-attention-based distillation method that explicitly captures inter-modal relationships during synthesis. By decoupling cross-modal alignment from joint optimization, X-AVDD is substantially more memory-efficient, reducing end-to-end synthesis cost by $\sim 8\times$ GPU-hours, while producing highly aligned synthetic data. Empirically, X-AVDD significantly outperforms other audio-visual distillation methods. On VGGSound, an audio-visual dataset of short YouTube clips, X-AVDD improves top-1 test accuracy from $8.2\%\rightarrow24.0\%$ at SPC=$10$ and $9.8\%\rightarrow26.5\%$ at SPC=$20$. Furthermore, X-AVDD improves cross-architecture generalization (e.g., $25.5\%\rightarrow35.3\%$ for a ViT target on AVE dataset at SPC $=50$) and successfully scales to complex architectures such as AudioCLIP, achieving $62.3\%$ accuracy using only $5.65\%$ of the full dataset, against the full dataset accuracy of $72.4\%$ while completing in $9\times$ lower training time compared to the full dataset.
Updated: 2026-10-09 14:51:26
标题: X-AVDD: 跨注意力音频-视觉数据集提炼
摘要: 音频-视觉数据集精炼旨在从大型配对音频-视觉数据集中合成一个小型数据集，同时几乎保留神经网络的训练效用。现有方法如DM和AVDD面临两个关键限制：1）它们占用内存多，使其难以扩展到复杂架构和更高的每类样本数（SPC）设置，并且2）它们经常无法保持强大的跨模态语义对应。为了解决这些问题，我们提出了X-AVDD，这是一种基于交叉注意力的精炼方法，它在合成过程中明确捕捉了跨模态关系。通过将跨模态对齐与联合优化解耦，X-AVDD在内存效率上显著提高，将端到端合成成本降低了约8倍的GPU小时，同时产生高度对齐的合成数据。经验上，X-AVDD在其他音频-视觉精炼方法上显著优于其他方法。在VGGSound上，这是一个由短YouTube剪辑组成的音频-视觉数据集，X-AVDD将SPC=10时的top-1测试准确率从8.2%提高到24.0%，将SPC=20时的准确率从9.8%提高到26.5%。此外，X-AVDD提高了跨架构泛化能力（例如，在SPC=50时，ViT目标在AVE数据集上的准确率从25.5%提高到35.3%），并成功扩展到复杂架构，如AudioCLIP，在仅使用全数据集的5.65%时实现了62.3%的准确率，而全数据集的准确率为72.4%，同时比全数据集的训练时间缩短了9倍。
更新时间: 2026-10-09 14:51:26
领域: cs.CV,cs.LG,cs.SD
Differentiable Hybrid Modelling for Learning and Optimising Chemical Transport Processes from Experimental Data
Transport models underpin the design and optimisation of chemical processes, yet their constitutive laws are typically hand-picked, their experimental initial conditions are assumed exact, and their computational cost restricts optimisation to a handful of process variables. Here we introduce a differentiable hybrid modelling framework that embeds neural networks within a GPU-accelerated population balance equation solver to discover constitutive laws and assimilate initial conditions directly from experimental crystallisation data. Using thirteen real experiments with multi-dimensional size and shape measurements, we show that treating measured initial conditions as exact yields a physically inconsistent rate law despite improved predictive accuracy, whereas assimilating them recovers a physically consistent, non-linear law and reduces trajectory error sevenfold relative to conventional modelling. End-to-end differentiability further enables gradient-based optimisation of high-dimensional operating profiles, identifying a 300-variable temperature profile that reduces crystal aspect ratio by 25% at constant mass, facilitating superior product attributes. Our framework augments, rather than replaces, mechanistic modelling, offering a general route to learning transport models from data and optimising processes with them.
Updated: 2026-10-09 14:43:00
标题: 可微混合建模：从实验数据学习和优化化学传输过程
摘要: 运输模型为化学过程的设计和优化提供了基础，然而它们的构成法则通常是手工选择的，实验初始条件被假定为精确，计算成本限制了优化到少数过程变量。在这里，我们介绍了一种可微分的混合建模框架，将神经网络嵌入到GPU加速的粒度平衡方程求解器中，以直接从实验结晶数据中发现构成法则和同化初始条件。利用十三个具有多维尺寸和形状测量的真实实验，我们表明，将测量的初始条件视为精确会导致一个在预测精度上有所改善但物理上不一致的速率定律，而同化这些条件会恢复一个物理上一致的非线性定律，并且相对于传统建模减少轨迹误差七倍。端到端的可微性进一步实现了基于梯度的高维操作剖面的优化，识别出一个能够在恒定质量下将晶体长宽比降低25%的300个变量的温度剖面，促进了更优异的产品属性。我们的框架增强了，而不是取代了，机械建模，提供了一种从数据中学习传输模型并利用它们优化过程的通用途径。
更新时间: 2026-10-09 14:43:00
领域: cs.CE,cs.LG,math.NA
On the Encoding of Calendar Features in Covariate-Informed Time Series Foundation Models
Time series foundation models (TSFMs) increasingly offer covariate support, with calendar information being the most common type of categorical covariate in practice, as it is available for every point in time. However, how these features are encoded is largely treated as an implementation detail rather than as an object of study, despite encoding being one of the few design choices left for pretrained models at inference time. With the present study, we demonstrate that the way information is propagated into the model substantially affects forecast accuracy of TSFMs. Therefore, we systematically compare seven calendar encodings, including three conventional target-agnostic representations (ordinal, one-hot, and cyclic) and four target encodings, across four state-of-the-art TSFMs on nine datasets from multiple domains in a zero-shot rolling evaluation. The nine datasets are influenced by social activities, such as weekends and holidays, making calendar information particularly relevant for forecasting. We find that encoding choice affects forecast accuracy across all models, with MASE differences of up to 9.9 % on non-holidays and 83 % on holidays within a single model. Effectiveness and utilization vary substantially across models. Improvements are concentrated around holidays, where calendar information provides signals unavailable from target history, while some model encoding combinations also improve non-holiday forecasts. A SHAP-based attribution analysis locates the contribution of the covariates relative to the autoregressive component and explains why some model-encoding combinations degrade performance. Our findings suggest that current benchmarks for covariate-informed TSFMs can distort model comparisons by imposing fixed encoding strategies.
Updated: 2026-10-09 14:41:10
标题: 关于在协变量信息时间序列基础模型中的日历特征编码
摘要: 时间序列基础模型（TSFMs）越来越多地提供协变量支持，其中日历信息是实践中最常见的类别协变量类型，因为它适用于每个时间点。然而，这些特征是如何编码的往往被视为实现细节，而不是研究对象，尽管编码是在推断时预训练模型的少数设计选择之一。通过本研究，我们证明了信息传播到模型中的方式显著影响了TSFMs的预测准确性。因此，我们系统地比较了七种日历编码，包括三种传统的目标不可知表示（序数、独热和循环）和四种目标编码，在四种最先进的TSFMs上对来自多个领域的九个数据集进行了零样本滚动评估。这九个数据集受社会活动的影响，如周末和假期，使得日历信息对于预测尤为重要。我们发现，编码选择影响了所有模型的预测准确性，一个模型内假期和非假期之间的MASE差异可达9.9％和83％。有效性和利用率在模型之间差异明显。改进主要集中在假期周围，其中日历信息提供了来自目标历史不可用的信号，而一些模型编码组合也改善了非假期的预测。基于SHAP的归因分析确定了协变量相对于自回归分量的贡献，并解释了为什么一些模型编码组合会降低性能。我们的研究结果表明，当前用于协变量信息TSFMs的基准可能通过强加固定编码策略来扭曲模型比较。
更新时间: 2026-10-09 14:41:10
领域: cs.LG
Not All Dimensions Are Equal: Outlier-Aware Attention Linearization with Grouped Quadratic Kernels
Linearizing softmax attention offers a route to efficient Transformer inference, but preserving pretrained attention remains challenging. We identify an important source of approximation error overlooked by standard linearization methods: a small number of outlier dimensions contribute disproportionately to query-key dot products and account for a large fraction of the approximation error. Motivated by this heterogeneity, we propose Outlier-Aware Attention Linearization (OAL), a dimension-aware method for constructing quadratic attention approximations. Starting from a second-order Taylor approximation, we introduce a Hadamard-product formulation that makes individual query-key contributions explicit. This formulation relaxes globally shared coefficients to support dimension-specific linear terms and cross-dimensional quadratic interactions while preserving separable evaluation. To retain this flexibility with fewer parameters, OAL groups dimensions with similar contribution scales to share coefficients, while allowing distinct treatment of the few outliers. The resulting positive quadratic kernel is calibrated against pretrained attention outputs and admits linear-time evaluation with respect to sequence length for fixed head dimensions. On Qwen2.5-0.5B, OAL reduces frozen-attention output relative MSE by approximately 8.9% compared with DiJiang. Replacing middle-layer attention modules and jointly fine-tuning the kernel coefficients with LoRA further lowers WikiText-103 perplexity from 31.71 to 27.50 compared with a learned global quadratic, while improving accuracy on PIQA, HellaSwag, and ARC-E. These results demonstrate the value of outlier-aware dimension grouping for both attention approximation and pretrained-model conversion. The code is available at https://github.com/wuqianyue09/OAL
Updated: 2026-10-09 14:40:19
标题: 并不是所有的维度都是相等的：具有分组二次核的异常值感知注意力线性化
摘要: 线性化softmax注意力提供了一条有效的Transformer推理路径，但保留预训练的注意力仍然具有挑战性。我们确定了标准线性化方法忽视的一个重要的近似误差来源：少数异常维度对查询-键点积起到不成比例的贡献，并占据了近似误差的很大一部分。在这种异质性的推动下，我们提出了Outlier-Aware Attention Linearization（OAL），这是一种基于维度的方法，用于构建二次注意力近似。从二阶泰勒近似开始，我们引入了一个Hadamard乘积公式，使得单独的查询-键贡献显式。这种公式放宽了全局共享系数，以支持维度特定的线性项和跨维度的二次交互，同时保持可分离的评估。为了在参数更少的情况下保持这种灵活性，OAL将具有类似贡献规模的维度分组以共享系数，并允许对少数异常值进行不同处理。得到的正二次核针对预训练的注意力输出进行了校准，并对于固定头维度的序列长度进行了线性时间评估。在Qwen2.5-0.5B上，与DiJiang相比，OAL将冻结注意力输出的相对MSE降低了约8.9％。用LoRA替换中间层注意力模块，并联合微调核系数，与学习的全局二次相比，将WikiText-103的困惑度从31.71降低到27.50，同时提高了PIQA、HellaSwag和ARC-E的准确性。这些结果展示了对于注意力近似和预训练模型转换，异常值感知维度分组的价值。代码可在https://github.com/wuqianyue09/OAL 上找到。
更新时间: 2026-10-09 14:40:19
领域: stat.ML,cs.AI,cs.LG
Inject a Turn, Not a Token: Learning Where to Redirect LLM Reasoning for Self-Correction
Large language models struggle to reliably identify and correct errors in their own reasoning. Unlike prior training-based approaches that fine-tune an LLM to generate revisions or feedback, we learn where to intervene in failed rollouts and determine how to redirect them. To choose where to intervene, we train a 130K-parameter MLP head on a frozen encoder using only final-answer correctness. For redirection, we find that under a fixed generation budget, the tested inline injections fail to reliably redirect reasoning, whereas presenting the same instruction in a new user turn changes the model's course. The resulting correction framework, UPTURN (User-Prompt TURN injection at a learned position), requires no LLM fine-tuning: it retains a selected prefix, injects an instruction as a new user turn, and regenerates the suffix. Across four reasoners and six math benchmarks, UPTURN improves average accuracy in every setting over a 32-rollout baseline. It outperforms self-correction and reflection-token injection baselines under matched correction budgets, transfers across reasoners and to GPQA-Diamond without retraining, and complements GRPO and self-correction methods.
Updated: 2026-10-09 14:37:51
标题: 注入一个转折，而不是一个标记：学习在哪里重定向LLM推理以进行自我校正
摘要: 大型语言模型在可靠地识别和纠正自身推理错误方面存在困难。与之前基于训练的方法不同，该方法通过微调LLM来生成修订或反馈，我们学习在失败的推理中介入的位置，并确定如何重新引导它们。为了选择介入的位置，我们在一个冻结的编码器上训练了一个130K参数的MLP头，只使用最终答案的正确性。在重新定向方面，我们发现在固定的生成预算下，测试的内联注入无法可靠地重新引导推理，而将相同指令呈现在新用户回合中会改变模型的走向。由此产生的校正框架UPTURN（在学习位置注入用户提示TURN），无需LLM微调：它保留所选前缀，将指令注入为新用户回合，并重新生成后缀。在四个推理者和六个数学基准测试中，UPTURN在每个设置中均提高了平均准确性，超过了32次推出基线。在匹配的校正预算下，它优于自我校正和反思令牌注入基线，在不重新训练的情况下跨推理者和到GPQA-Diamond，并且与GRPO和自我校正方法相辅相成。
更新时间: 2026-10-09 14:37:51
领域: cs.CL,cs.AI
A Step Harder, Not a Step Too Far: The Benefits and Limits of Curriculum Difficulty Progression in GRPO
Curriculum learning typically organizes training samples from easy to hard. In GRPO, medium-difficulty samples are attractive for online selection because they are more likely to yield informative group-relative learning signals, and prioritizing them can improve training efficiency. However, it remains unclear whether persistently selecting such samples also leads to the strongest final model performance, or whether training can benefit from progressing toward harder samples. Meanwhile, sample difficulty changes as the policy evolves, so fixed difficulty estimates can become misaligned with the current policy. We therefore re-estimate sample difficulty under the current policy and, inspired by the easy-to-hard principle of curriculum learning, design and evaluate a staged curriculum that progressively shifts the active difficulty band toward harder samples. Experiments on GSM8K show that persistent medium-difficulty selection does not achieve the strongest observed performance. Across three random seeds, progressing to a moderately harder difficulty band improves test accuracy to $0.6591 \pm 0.0077$, compared with $0.6449 \pm 0.0024$ for the fixed-medium curriculum. However, further progression to a harder difficulty band reduces accuracy to $0.6405 \pm 0.0053$. These results show that staged difficulty progression can outperform persistent medium-difficulty selection, but its benefit is non-monotonic. Effective GRPO curricula can therefore benefit from progressing beyond medium difficulty, while avoiding excessive progression toward overly difficult samples.}
Updated: 2026-10-09 14:37:32
标题: 一步更难，而非步步太远：GRPO课程难度逐步增加的益处和局限性
摘要: 课程学习通常将训练样本从易到难进行组织。在GRPO中，中等难度的样本对于在线选择是有吸引力的，因为它们更有可能产生信息丰富的相对组学习信号，并优先考虑它们可以提高训练效率。然而，目前尚不清楚是否持续选择这些样本也会导致最强的最终模型性能，或者训练是否可以从向更难的样本发展中受益。同时，随着策略的演变，样本的难度也会发生变化，因此固定难度估计可能会与当前策略不一致。因此，我们在当前策略下重新估计样本难度，并受到课程学习易到难原则的启发，设计和评估了一个分阶段的课程，逐渐将主动难度范围向更难的样本转移。在GSM8K上的实验表明，持续选择中等难度并不能达到最强的性能。在三个随机种子上，逐渐转向一个适度更难的难度范围将测试准确性提高至$0.6591 \pm 0.0077$，而固定中等课程为$0.6449 \pm 0.0024$。然而，进一步进展到更难的难度范围会降低准确性至$0.6405 \pm 0.0053$。这些结果表明，分阶段难度进展可以胜过持续中等难度选择，但其益处是非单调的。有效的GRPO课程因此可以受益于超越中等难度，同时避免过度进展至过于困难的样本。
更新时间: 2026-10-09 14:37:32
领域: cs.AI
A Multi-Agent LLM System for Automated Tuning of Specialized Neural Network Models
Many machine learning approaches are two-level: an expert selects a class of methods, within this class the choice of key components is critical and usually manual. This paper addresses the automatic tuning of these components and proposes a framework transferable across models of different nature. The multi-agent large language model (LLM) framework has three roles---analyst, coder, and recorder~---with an experiment log and a sandboxed execution environment. Unlike systems that optimize the entire ML script or search architectures in a discrete space, the framework operates over an open but semantically narrow space: the agent generates code only for the specified components, leaving the overall network structure unchanged. The approach is validated in two directions: tuning a classification spiking neural network with correlation learning, and solving high-order differential equations with Physics-Informed Neural Networks (PINN). Across five tasks---spiking classification on MNIST and Digits, and boundary-value problems for the generalized nonlinear Schrödinger equation (GNLSE) and Kuramoto--Sivashinsky (KS) equations---the agent outperforms the baselines: classification accuracy is 13 percentage points (MNIST) and 6 percentage points (Digits) higher; the error against the analytical single-soliton solution is reduced approximately 2491-fold; the invariant-preservation error drops approximately 325-fold in the selected GNLSE run with a Gaussian perturbation and by more than 100-fold on average over repeated initializations; and the mass drift drops 14-fold for the KS problem with a Gaussian perturbation. The framework thus tunes key components on tasks of different nature without changing its internal structure.
Updated: 2026-10-09 14:34:22
标题: 一个用于自动调节专门神经网络模型的多智能体LLM系统
摘要: 许多机器学习方法都是两级的：专家选择一类方法，在这个类中选择关键组件是至关重要的，通常是手动的。本文解决了这些组件的自动调整，并提出了一个跨不同性质模型可转移的框架。多智能体大型语言模型（LLM）框架有三个角色---分析员、编码员和记录员---具有一个实验日志和一个隔离执行环境。与优化整个ML脚本或在离散空间中搜索体系结构的系统不同，该框架在一个开放但语义狭窄的空间内运作：代理只为指定的组件生成代码，而不改变整体网络结构。该方法在两个方向上得到验证：调整具有相关学习的分类脉冲神经网络，并使用基于物理的神经网络（PINN）解决高阶微分方程。在五个任务中---在MNIST和数字上进行脉冲分类，以及为广义非线性薛定谔方程（GNLSE）和Kuramoto--Sivashinsky（KS）方程解决边值问题---代理超越了基准线：分类准确率比MNIST高13个百分点，比Digits高6个百分点；与分析单孤子解的误差约减少2491倍；在选择的带高斯扰动的GNLSE运行中，不变性保持误差约降低325倍，平均在重复初始化中降低超过100倍；对于带高斯扰动的KS问题，质量漂移降低了14倍。因此，该框架在不改变其内部结构的情况下调整了不同性质任务的关键组件。
更新时间: 2026-10-09 14:34:22
领域: cs.AI,cs.CL,cs.LG,cs.MA,cs.NE
Pretraining and Test-Time Training Budgets Matter for ARC: A Controlled Study of VARC and TRM
The Abstraction and Reasoning Corpus (ARC) is increasingly used to evaluate whether neural networks can infer task-specific transformations from few demonstrations. Recent ARC models differ in architecture, but also in broader training and evaluation protocols, including pretraining data, data augmentation, test-time setup, and optimization budgets, making conclusions about architecture benefits debatable. In this work, we study how the performance of the vision-inspired VARC and recursion-based TRM models is affected by pretraining and test-time training (TTT) budgets under a common inductive protocol on ARC-AGI-1. We cross five pretraining budgets with four TTT budgets for each test task. Each budget specifies a task-variant count and total sample presentations, defining the controlled training budgets. We observe that increasing TTT sample presentations improves the standard ARC pass@2 metric in all settings for both models. Increasing the TTT task-variant count consistently helps only TRM, whereas VARC often performs better with fewer variants. VARC also remains competitive without pretraining, while TRM benefits much more from it. These results show that pretraining and TTT budgets affect ARC evaluation and motivate reporting at comparable budgets.
Updated: 2026-10-09 14:33:30
标题: 预训练和测试时间训练预算对ARC有影响：VARC和TRM的控制研究
摘要: The Abstraction and Reasoning Corpus (ARC)越来越被用来评估神经网络是否能够从少量演示中推断特定任务的转换。最近的ARC模型在架构上有所不同，同时在更广泛的训练和评估协议中也存在差异，包括预训练数据，数据增强，测试时设置和优化预算，这使得关于架构优势的结论具有争议性。在这项工作中，我们研究了在ARC-AGI-1上采用常见的归纳协议时，受到预训练和测试时训练（TTT）预算影响的基于视觉的VARC和基于递归的TRM模型的性能。我们将五个预训练预算与每个测试任务的四个TTT预算相结合。每个预算都指定了任务变体计数和总样本呈现次数，定义了受控训练预算。我们观察到增加TTT样本呈现次数可以改善两种模型在所有设置下的标准ARC pass@2指标。增加TTT任务变体计数只有对TRM有一致的帮助，而VARC通常在较少变体的情况下表现更好。VARC在没有预训练的情况下仍然具有竞争力，而TRM则更多受益于预训练。这些结果表明，预训练和TTT预算影响ARC评估，并促使在可比较的预算下进行报告。
更新时间: 2026-10-09 14:33:30
领域: cs.LG,cs.AI
Deep End-to-End Survival Analysis with Temporal Consistency
In this study, we present a novel Survival Analysis algorithm designed to efficiently handle large-scale longitudinal data. Our approach draws inspiration from Reinforcement Learning principles, particularly the Deep Q-Network paradigm, extending Temporal Learning concepts to Survival Regression. A central argument in our method relies on temporal consistency, a recursive consistency condition between survival predictions made at consecutive states. Our framework introduces a delayed target network that decouples the generation of temporally consistent training targets from model updates, providing a more stable training signal while enabling mini-batch optimization and end-to-end training. Additionally, it supports arbitrarily complex architectures, enabling the modeling of intricate temporal dependencies. Through numerous experiments we provide empirical evidence demonstrating our framework's ability to exploit temporal consistency across datasets of varying sizes. Moreover, our algorithm outperforms benchmarks on datasets with long sequences, demonstrating its ability to capture long-term patterns. Finally, ablation studies show that the degree of target delay has a substantial effect on predictive performance and training variability. Across two distinct survival model families, delayed target updates consistently improve over the same-timescale baseline.
Updated: 2026-10-09 14:30:22
标题: 深度端到端生存分析与时间一致性
摘要: 在这项研究中，我们提出了一种新颖的生存分析算法，旨在高效处理大规模纵向数据。我们的方法受到强化学习原则的启发，特别是深度Q网络范例，将时间学习概念扩展到生存回归。我们方法中的一个核心论点是时间一致性，即在连续状态下进行的生存预测之间的递归一致性条件。我们的框架引入了一个延迟目标网络，将生成时间一致的训练目标与模型更新分开，提供更稳定的训练信号，同时实现小批量优化和端到端训练。此外，它支持任意复杂的架构，能够建模复杂的时间依赖关系。通过大量实验，我们提供了实证证据，证明我们的框架能够利用不同规模数据集间的时间一致性。此外，我们的算法在具有长序列的数据集上优于基准，展示了其捕捉长期模式的能力。最后，消融研究表明目标延迟程度对预测性能和训练变异性有显著影响。在两个不同的生存模型家族中，延迟目标更新始终优于相同时间尺度基线。
更新时间: 2026-10-09 14:30:22
领域: cs.LG
Scalable Sparse Autoregressive Variational Inference for Non-Gaussian Spatial Fields
Latent field models are essential for spatial statistics, but the posterior distribution of the latent field, even with a Gaussian process (GP) prior, is analytically intractable whenever the likelihood is non-Gaussian. Existing approximations are either too slow or too restrictive for large non-Gaussian datasets. We present a scalable variational inference framework that mirrors the sparse nearest-neighbor autoregressive structure of the prior in the variational family, uses normalizing-flow conditionals for expressiveness, and amortizes all variational parameters via a single shared neural conditioner. Doubly stochastic optimization with reduced ancestor sets achieves polylogarithmic complexity per iteration.
Updated: 2026-10-09 14:30:12
标题: 可扩展的稀疏自回归变分推断用于非高斯空间场
摘要: 潜在场模型对于空间统计是至关重要的，但是即使使用高斯过程（GP）先验，潜在场的后验分布在似然函数为非高斯分布时也是难以分析的。现有的近似方法要么速度太慢，要么对大型非高斯数据集太过严格。我们提出了一个可扩展的变分推断框架，该框架在变分族中反映了先验的稀疏最近邻自回归结构，使用归一化流条件来增加表达能力，并通过单个共享的神经条件器对所有变分参数进行摊销。通过减少祖先集合的双随机优化实现了每次迭代的对数复杂度。
更新时间: 2026-10-09 14:30:12
领域: stat.ML,cs.LG,stat.CO,stat.ME
TabReliBench: Evaluating Tabular Foundation Models Beyond Predictive Accuracy
Tabular foundation models (TFMs) are increasingly popular for their strong predictive performance. Many also provide posterior predictive distributions (PPDs), whose reliability remains less studied. We assess reliability through point prediction, distributional quality, marginal and subgroup calibration, and interval efficiency. We then introduce TabReliBench, a large classification and regression benchmark for these comparisons and dataset-level analyses. We find that TFMs' rankings change between TabArena and our benchmark and across reliability metrics. Regression TFMs usually achieve near-nominal marginal coverage and retain narrow intervals, but this favorable marginal behavior does not imply valid conditional coverage. Dataset-level analyses show that classification difficulty often accompanies poor calibration, whereas regression predictive performance is not a reliable indicator of calibration. Ensembling improves predictive performance but does not consistently improve calibration of PPDs; TFMs also benefit little from conformal post-processing, which does not reliably resolve subgroup gaps. Reliable TFM selection therefore requires multidimensional evaluation beyond aggregate rankings. Code is available at https://github.com/Tianqi-Zhao/TabReliBench.
Updated: 2026-10-09 14:29:55
标题: TabReliBench：评估表格基础模型的预测准确性之外的性能
摘要: 表格基础模型（TFMs）因其强大的预测性能而越来越受欢迎。许多TFMs还提供后验预测分布（PPDs），其可靠性仍未得到充分研究。我们通过点预测、分布质量、边际和子组校准以及区间效率来评估可靠性。然后我们介绍TabReliBench，这是一个用于这些比较和数据集级分析的大型分类和回归基准。我们发现TFMs在TabArena和我们的基准以及跨可靠性指标之间的排名会发生变化。回归TFMs通常实现接近名义边际覆盖率，并保持狭窄的区间，但这种有利的边际行为并不意味着有效的条件覆盖率。数据集级分析显示，分类困难通常伴随着较差的校准，而回归预测性能并不是校准的可靠指标。集成改进了预测性能，但并不一致地改善PPDs的校准；TFMs也很少从符合后处理中受益，这并不能可靠地解决子组差距。因此，可靠的TFM选择需要多维度评估，超越聚合排名。代码可在https://github.com/Tianqi-Zhao/TabReliBench获得。
更新时间: 2026-10-09 14:29:55
领域: cs.LG
A Critical Look at Power Sampling when Test Time Sharpening LLMs
Reinforcement learning with verifiable rewards (RLVR) has substantially improved LLM reasoning, yet growing evidence suggests it mainly performs distribution sharpening: concentrating probability mass on reasoning paths the base model already produces. Recent work therefore seeks RLVR's gains without needing reward signals, by sampling from the power distribution $π_α\propto π_{\text{base}}^α$ via Markov chain Monte Carlo or sequential Monte Carlo. We show that these samplers fail to converge to $π_α$ under realistic compute budgets, and that their gains are largely recreated via simple Best-of-$N$ with appropriate grading . This suggests they succeed by selecting high-probability sequences, not by approximating $π_α$, raising a natural question: is the power distribution itself a worthwhile target? We test this hypothesis directly by fine-tuning the base model to match $π_α$, sidestepping sampler convergence issues entirely. Despite approximating the target far more faithfully than existing samplers, this strategy yields gains that are noisy and specific to certain models and sampling tasks. Together, our results suggest that the value of power sampling lies not in faithfully targeting $π_α$, but in the simpler goal of extracting sequences the base model already deems likely.
Updated: 2026-10-09 14:29:28
标题: 一个关于测试时间缩短LLM时权力抽样的批判性观察
摘要: 具有可验证奖励的强化学习（RLVR）显著改进了LLM推理，然而越来越多的证据表明它主要执行分布锐化：将概率质量集中在基础模型已经产生的推理路径上。因此，最近的工作试图在不需要奖励信号的情况下实现RLVR的收益，通过通过马尔可夫链蒙特卡洛或顺序蒙特卡洛从幂分布$π_α\propto π_{\text{base}}^α$中采样。我们发现这些采样器在现实计算预算下无法收敛到$π_α，并且它们的收益主要通过适当分级的简单Best-of-$N$重现。这表明它们通过选择高概率序列而成功，而不是通过逼近$π_α，提出一个自然的问题：幂分布本身是否是一个值得追求的目标？我们通过微调基础模型以匹配$π_α直接测试这个假设，完全避开了采样器的收敛问题。尽管比现有采样器更忠实地逼近目标，这种策略产生的收益是嘈杂的，并且特定于某些模型和采样任务。综合起来，我们的结果表明，幂采样的价值不在于忠实地瞄准$π_α，而在于更简单的目标，即提取基础模型已经认为可能的序列。
更新时间: 2026-10-09 14:29:28
领域: stat.ML,cs.LG
ToFu: Exploiting and Fuzzing the LLM-Orchestrator-Tool Trust Boundary
Large language models (LLMs) are increasingly deployed as agents that interact with external software through tools, with an orchestration layer translating model outputs into structured tool invocations. This architecture creates an LLM-orchestrator-tool trust boundary where tool semantics may go unenforced and LLM-generated arguments may be incorrectly treated as trusted. We present ToFu, a tool-aware framework that automatically derives semantic security checks for LLM tools from their existing metadata and a CWE-guided sanitizer corpus. ToFu specializes these checks to individual tools and automatically generates fuzzing harnesses and semantic oracles using runtime introspection and outcome sanitization. We evaluated ToFu across 273 MCP servers containing 3405 tools and identified verified bugs corresponding to CWE weakness classes in 343 tools across 118 of the servers. Our findings show that real-world tools frequently fail to enforce their documented input semantics, making it unsafe to treat LLM-generated arguments as trusted.
Updated: 2026-10-09 14:27:32
标题: ToFu: 利用和模糊测试LLM-Orchestrator-Tool的信任边界
摘要: 大型语言模型（LLMs）越来越多地被部署为通过工具与外部软件交互的代理，其中一个编排层将模型输出转换为结构化工具调用。这种架构创建了一个LLM-编排器-工具信任边界，工具语义可能未得到强制执行，LLM生成的参数可能被错误地视为可信任。我们提出了ToFu，这是一个工具感知框架，它可以自动从现有元数据和基于CWE的消毒器语料库中推导出LLM工具的语义安全检查。ToFu将这些检查专门应用于各个工具，并利用运行时内省和结果消毒自动生成模糊测试工具和语义神谕。我们在包含3405个工具的273个MCP服务器上评估了ToFu，并在118个服务器中识别出与CWE弱点类相对应的343个工具中的已验证错误。我们的研究结果表明，现实世界中的工具经常未能强制执行其记录的输入语义，因此将LLM生成的参数视为可信是不安全的。
更新时间: 2026-10-09 14:27:32
领域: cs.CR
Electron-Density-Informed Molecular Property Prediction Using Surrogate Functionals
Machine learning methods are increasingly used to predict chemical properties directly from the molecular structure to bypass the often prohibitive cost of quantum mechanical calculations. Recent work has shown that including the ground-state electron density as input to the machine learning model can improve accuracy in this setting. This work employs the recent surrogate functional framework originally developed for orbital-free density functional theory and extends it to predicting molecular properties. We show that passing the ground-state density as input to the model, while simultaneously enforcing it to be the minimum of the surrogate functional, surpasses state-of-the-art performance for six out of seven extensive properties of the QM9 dataset. Additionally, the approach shows improved generalization to larger molecules taken from the QM40 dataset. On all extensive properties of both datasets, our approach outperforms models that are informed by the ground-state density but do not enforce the surrogate functional constraints.
Updated: 2026-10-09 14:27:05
标题: 基于电子密度信息的分子性质预测使用代理函数
摘要: 机器学习方法越来越被用于直接从分子结构预测化学性质，以避免量子力学计算通常昂贵的成本。最近的研究表明，将基态电子密度作为机器学习模型的输入可以提高在这种情况下的准确性。本研究采用了最近为无轨道密度泛函理论开发的替代功能框架，并将其扩展到预测分子性质。我们展示了将基态密度作为模型输入，并同时强制它成为替代功能的最小值，可以超越QM9数据集中七个广泛性质中的六个的最新性能。此外，这种方法在从QM40数据集中获取的较大分子上显示出改进的泛化能力。在这两个数据集的所有广泛性质中，我们的方法优于那些受基态密度启发但不强制替代功能约束的模型。
更新时间: 2026-10-09 14:27:05
领域: physics.chem-ph,cs.LG
User Trajectory Prediction Unifying Global and Local Temporal Information
Trajectory prediction is essential for formulating proactive strategies that anticipate user mobility and support advance preparation. Therefore, how to reduce the forecasting error in user trajectory prediction within an acceptable inference time arises as an interesting issue. However, trajectory data contains both global and local temporal information, complicating the extraction of the complete temporal pattern. Moreover, user behavior occurs over different time scales, increasing the difficulty of capturing behavioral patterns. To address these challenges, a trajectory prediction model based on multilayer perceptron (MLP), multi-scale convolutional neural network (MSCNN), and cross-attention (CA) is proposed. Specifically, MLP is used to extract the global temporal information of each feature. In parallel, MSCNN is employed to extract the local temporal information by modeling interactions among features within a local temporal range. Convolutional kernels with different sizes are used in MSCNN to capture temporal information at multiple resolutions, enhancing the model's adaptability to different behavioral patterns. Finally, CA is applied to fuse the global and local temporal information. Experimental results show that our model reduces mean squared error (MSE) by 5.04% and mean absolute error (MAE) by 4.35% compared with ModernTCN in 12-step prediction, while maintaining similar inference time.
Updated: 2026-10-09 14:26:25
标题: 用户轨迹预测：统一全球和本地时间信息
摘要: 轨迹预测对于制定积极的策略以预测用户移动并支持提前准备至关重要。因此，在可接受的推断时间内如何减少用户轨迹预测中的预测误差成为一个有趣的问题。然而，轨迹数据包含全局和局部时间信息，使得完整时间模式的提取变得复杂。此外，用户行为发生在不同的时间尺度上，增加了捕捉行为模式的难度。为了解决这些挑战，提出了一种基于多层感知器（MLP）、多尺度卷积神经网络（MSCNN）和交叉注意力（CA）的轨迹预测模型。具体而言，MLP用于提取每个特征的全局时间信息。同时，MSCNN用于提取局部时间信息，通过对局部时间范围内特征之间的交互建模。MSCNN中使用不同尺寸的卷积核来捕获多个分辨率下的时间信息，增强模型对不同行为模式的适应性。最后，CA被应用于融合全局和局部时间信息。实验结果显示，与ModernTCN在12步预测中相比，我们的模型将均方误差（MSE）降低了5.04％，均方绝对误差（MAE）降低了4.35％，同时保持类似的推断时间。
更新时间: 2026-10-09 14:26:25
领域: cs.LG
SelectArena: A Modern Benchmark for Feature Selection on Real-world Tabular Data
Selecting a subset of features (columns) from a tabular dataset is a ubiquitous challenge in real-world applications. Industry experts seek to minimize model training costs, climate scientists require an interpretable subset of factors to understand extreme weather events, and clinicians need to choose among diagnostic tests due to financial and time constraints. When looking at the scientific literature to determine which feature selection method to use, practitioners across domains face an unrepresentative landscape of benchmarking. Despite extensive research, existing feature selection benchmarks provide limited guidance because they rely on outdated models, ad hoc evaluation protocols, or artificially generated data that do not reflect real-world problems. To close the gap between scientific benchmarking and real-world applications, we introduce SelectArena, a modern benchmark for feature selection. We compare 15 widely used feature selection methods on 71 manually curated real-world tabular datasets, covering classification and regression problems, using standardized evaluation protocols and state-of-the-art tabular (foundation) models. Our results reveal clear trade-offs across predictive performance, runtime, stability, and validity. Tree-based methods, CART and RFImportance, along with F-Test are Pareto-optimal when taking into account predictive performance and runtime. F-Test, information-theory-, and distance-based methods lead on stability and validity, with F-Test emerging as a balanced choice across our metrics. Search-based methods are slow, unstable and achieve low performance. SelectArena turns these findings into practical guidance and, as an extensible platform, supports future feature selection research.
Updated: 2026-10-09 14:21:47
标题: SelectArena：一个现代的基准，用于在真实世界的表格数据上进行特征选择
摘要: 从一个表格数据集中选择一部分特征（列）是现实世界应用中普遍存在的挑战。行业专家寻求最小化模型训练成本，气候科学家需要一个可解释的因素子集来理解极端天气事件，临床医生需要在财务和时间限制下选择诊断测试。在查看科学文献以确定要使用哪种特征选择方法时，跨领域的从业者面临着一个不具代表性的基准测试场景。尽管进行了大量研究，现有的特征选择基准测试提供的指导有限，因为它们依赖于过时的模型、临时评估协议或人为生成的数据，这些数据不反映现实世界问题。为了弥合科学基准测试与现实世界应用之间的差距，我们介绍了SelectArena，一个用于特征选择的现代基准测试。我们在71个手动筛选的真实世界表格数据集上比较了15种广泛使用的特征选择方法，涵盖分类和回归问题，使用标准化评估协议和最先进的表格（基础）模型。我们的结果揭示了在预测性能、运行时间、稳定性和有效性之间明显的权衡。基于树的方法，包括CART和RFImportance，以及F-Test在考虑预测性能和运行时间时是帕累托最优的。F-Test、信息理论和基于距离的方法在稳定性和有效性方面领先，F-Test成为我们的指标中的平衡选择。基于搜索的方法速度慢、不稳定且性能低。SelectArena将这些发现转化为实际指导，并作为一个可扩展的平台，支持未来的特征选择研究。
更新时间: 2026-10-09 14:21:47
领域: cs.LG,cs.AI
OTel: Open Telco AI Datasets, Benchmarks, and Models
We present Open Telco (OTel), an open telecom AI resource that releases derived telecom datasets for retrieval, reranking, instruction tuning, and safety/abstention, together with 30 full-parameter post-trained baselines spanning 10 embedding models, 3 rerankers, and 17 language models. The community has already engaged substantially with the resource: as of May 3, 2026, the released models have been downloaded over 16 million times and the project has received 157+ pieces of media coverage worldwide. Building on prior open telecom datasets and benchmarks, OTel provides documented telecom data sources, held-out evaluation partitions, trained embedding models, rerankers, context-grounded LLMs, and safety/abstention data in one unified resource. Each baseline starts from an open-weight model and is post-trained on OTel-derived data using an open training recipe, then evaluated on held-out OTel evaluation partitions. OTel post-training improves performance across all three model families: embedding retrieval reaches 93.1% NDCG@10, reranking reaches 0.947 MRR@10, and language-model correctness reaches 87.8%. We release OTel as a reproducible starting point and invite the community to expand the data, improve embedding and reranking models, and build stronger context-grounded telecom LLMs.
Updated: 2026-10-09 14:20:47
标题: OTel：开放的电信AI数据集、基准和模型
摘要: 我们提出了Open Telco（OTel），这是一个开放的电信人工智能资源，释放了衍生的电信数据集，用于检索、重新排名、指导调整以及安全/弃权，同时提供了30个完全参数的后训练基线，涵盖了10个嵌入模型、3个重新排名器和17个语言模型。社区已经充分利用了这一资源：截至2026年5月3日，释放的模型已被下载超过1600万次，该项目已在全球范围内获得了157多篇媒体报道。在之前的开放电信数据集和基准的基础上，OTel提供了有记录的电信数据源、保留评估分区、训练好的嵌入模型、重新排名器、基于上下文的LLM以及安全/弃权数据，统一在一个资源中。每个基线从一个开放权重模型开始，在OTel衍生数据上进行后训练，使用开放的训练配方，然后在保留的OTel评估分区上进行评估。OTel后训练改善了所有三个模型系列的性能：嵌入检索达到了93.1%的NDCG@10，重新排名达到了0.947的MRR@10，语言模型的正确率达到了87.8%。我们将OTel发布为一个可重现的起点，并邀请社区扩展数据、改进嵌入和重新排名模型，并构建更强大的基于上下文的电信LLM。
更新时间: 2026-10-09 14:20:47
领域: cs.AI
Disentangling Total-Variance and Signal-to-Noise-Ratio Improves Diffusion Models
Diffusion models have been widely adopted as a powerful generative framework across numerous scientific domains, yet their inherently iterative sampling process remains a significant computational bottleneck. While this is typically mitigated by reducing the number of reverse diffusion steps, sample quality becomes highly sensitive to the chosen noise schedule, i.e. the manner in which noise is introduced and the signal is reduced at each step. Although prior work has improved on the original variance-preserving and variance-exploding schedules, these approaches passively adjust the total variance, without direct independent control over it. In this work, we propose a novel total-variance/signal-to-noise-ratio disentangled (TV/SNR) framework, where TV and SNR can be controlled independently. Our approach reveals that schedules with exponentially exploding TV can be consistently improved by adopting a constant TV schedule while preserving the underlying SNR schedule. Furthermore, we demonstrate that generalizing the SNR schedule of optimal transport flow matching significantly enhances generation efficiency. Crucially, these gains remain consistent across various reverse diffusion solvers and reduce the search space to two independent and interpretable parameters, enabling fast and robust sampling for complex scientific applications, such as molecular structure generation.
Updated: 2026-10-09 14:19:51
标题: 解开总方差和信噪比，改善扩散模型
摘要: 扩散模型已被广泛应用作为强大的生成框架，涵盖了众多科学领域，然而其固有的迭代抽样过程仍然是一个重要的计算瓶颈。尽管通常可以通过减少逆向扩散步骤的数量来缓解这一问题，但样本质量对选择的噪声时间表敏感，即噪声如何引入以及信号在每一步如何减少。尽管先前的工作改进了原始的保持方差和爆炸方差时间表，但这些方法 passively 调整了总体方差，而没有直接独立控制它。在这项工作中，我们提出了一个新颖的总方差/信噪比分离（TV/SNR）框架，其中 TV 和 SNR 可以独立控制。我们的方法表明，通过采用恒定的 TV 时间表，同时保持基础的 SNR 时间表，可以持续改进具有指数爆炸 TV 的时间表。此外，我们证明了将最优传输流匹配的 SNR 时间表进行泛化可以显著增强生成效率。关键是，这些收益在各种逆向扩散求解器中保持一致，并将搜索空间缩减为两个独立且可解释的参数，为复杂科学应用（如分子结构生成）提供快速和稳健的抽样。
更新时间: 2026-10-09 14:19:51
领域: cs.LG,stat.ML
Learning Size Independent Constraint Models from Input Data and Positive and Negative Example Solutions
In this paper we describe a Constraint Acquisition Tool that generates size independent constraint programs from input data and positive and negative example solutions of multiple problem instances. The system finds size specific models based on typical constraint patterns over global and binary constraints, then generalizes the models to a size independent model that can be run on unseen input data. The system produces MiniZinc code as well as a textual description of the model. We evaluate the tool on the example problems of the PTHG21 Challenge and show the feasibility of the approach.
Updated: 2026-10-09 14:18:49
标题: 从输入数据和正负示例解中学习大小无关的约束模型
摘要: 在这篇论文中，我们描述了一个约束获取工具，它可以根据输入数据和多个问题实例的正面和负面示例解生成与大小无关的约束程序。该系统基于全局和二进制约束的典型约束模式找到特定大小的模型，然后将这些模型概括为一个可以在未知输入数据上运行的大小无关模型。该系统生成MiniZinc代码以及模型的文本描述。我们在PTHG21挑战的示例问题上评估了该工具，并展示了该方法的可行性。
更新时间: 2026-10-09 14:18:49
领域: cs.AI
Model-agnostic Knowledge Augmented Machine Learning for Weakly Annotated Bioimaging Data
Machine learning models are increasingly used for automated feature extraction from weakly annotated bioimaging data, such as microscopy and cytomorphology images. Due to the sole dependency of automated feature extraction, existing methods yield inconsistent classification performance across weakly annotated datasets. Here, we propose a model-agnostic knowledge augmentation method that enables ML models to exploit domain knowledge beyond what automated feature extraction can capture. The proposed approach improves classification performance across seven ML architectures. When integrated with masked autoencoder-based self-supervised learning, our method achieves the strongest overall performance, improving precision by 2.9% over the next-best method on three in-house microscopy datasets and by an average of 6.8% across three public cytomorphology benchmarks. We perform nucleus-level classification based on mislocalization of TAR DNA-binding protein 43 (TDP-43) from the nucleus to the cytoplasm, a phenotype linked to Amyotrophic Lateral Sclerosis (ALS), using three independent microscopy datasets of motor neurons derived from human induced pluripotent stem cells. Accurate quantification of this phenotype at scale remains challenging and may contribute to inconsistencies across studies. Application of the best-performing model to these microscopy datasets reveals biological interpretations related to ALS, including dose-dependent effects of the proteasomal stressor, MG132, and the oxidative stressor, sodium arsenite (SA), on TDP-43 mislocalization, variable pathology in motor neurons from sporadic and familial ALS donors, and a lack of protection of the pharmacological compounds, edaravone and niclosamide, against SA-induced TDP-43 mislocalization. Together, our pipeline enables in-vitro cellular image analysis to advance understanding of ALS-associated pathology and potential treatments.
Updated: 2026-10-09 14:17:48
标题: 不考虑模型的知识增强机器学习用于弱标注的生物成像数据
摘要: 机器学习模型越来越多地被用于从弱标注的生物成像数据中自动提取特征，比如显微镜和细胞形态学图像。由于自动特征提取的唯一依赖，现有方法在弱标注数据集上产生了不一致的分类性能。在这里，我们提出了一种模型无关的知识增强方法，使机器学习模型能够利用领域知识，超越自动特征提取所能捕捉到的内容。所提出的方法改进了七种机器学习架构的分类性能。当与基于掩蔽自动编码器的自监督学习集成时，我们的方法在三个内部显微镜数据集上实现了最强的整体性能，将精度提高了2.9%，并在三个公共细胞形态学基准测试中平均提高了6.8%。我们基于细胞核内TAR DNA结合蛋白43 (TDP-43)的错位进行细胞核级分类，这与肌萎缩性侧索硬化（ALS）相关，使用了来源于人诱导多能干细胞的运动神经元的三个独立显微镜数据集。在规模上准确量化这种表型仍然具有挑战性，并可能导致研究之间的不一致性。将最佳模型应用于这些显微镜数据集，揭示了与ALS相关的生物解释，包括蛋白酶体应激剂MG132和氧化应激剂亚砷酸钠（SA）对TDP-43错位的剂量依赖效应，ALS患者运动神经元中可变的病理变化，以及药理化合物埃达罗韦和尼克罗苯对SA诱导的TDP-43错位的保护缺失。总之，我们的流水线使体外细胞图像分析能够推动对ALS相关病理和潜在治疗方法的理解。
更新时间: 2026-10-09 14:17:48
领域: eess.IV,cs.AI,cs.CV
Physics-guided surrogate learning enables zero-shot control of turbulent wings
Turbulent boundary layers over aerodynamic surfaces are a major source of aircraft drag, yet their control remains challenging due to multiscale dynamics and spatial variability, particularly under adverse pressure gradients. Reinforcement learning has outperformed state-of-the-art strategies in canonical flows, but its application to realistic geometries is limited by computational cost and transferability. Here we show that these limitations can be overcome by exploiting local structures of wall-bounded turbulence. Policies are trained in turbulent channel flows matched to wing boundary-layer statistics and deployed directly onto a NACA4412 wing section at $Re_c=2\times10^5$ without additional training on the wing, constituting zero-shot deployment. The transferred policies reduce total drag by 11.1\%, while state-of-the-art opposition control reduces it by 8.9\%. With the actuation cost taken into account, the corresponding net-energy savings are 10.9\% and 8.8\%, respectively. Surrogate training provides an estimated improvement of four orders of magnitude in exploration efficiency over direct on-wing training, enabling scalable flow control.
Updated: 2026-10-09 14:16:19
标题: 物理引导的替代学习实现湍流翅膀的零控制
摘要: 飞机表面上的湍流边界层是飞机阻力的主要来源，但由于多尺度动态和空间变化，特别是在不利的压力梯度下，它们的控制仍然具有挑战性。在经典流动中，强化学习已经超越了最先进的策略，但其在实际几何形状上的应用受到计算成本和可转移性的限制。在这里，我们展示了通过利用壁面湍流的局部结构可以克服这些限制。在与机翼边界层统计数据匹配的湍流通道流中训练策略，并直接部署到具有$Re_c=2\times10^5$的NACA4412机翼剖面上，而无需在机翼上进行额外训练，构成零射击部署。转移的策略将总阻力减少了11.1％，而最先进的对抗控制将其减少了8.9％。考虑到执行成本，相应的净能源节约分别为10.9％和8.8％。代理训练估计探索效率比直接在机翼上进行训练提高了四个数量级，从而实现可扩展的流控制。
更新时间: 2026-10-09 14:16:19
领域: physics.flu-dyn,cs.AI
PANDA: Prior-guided Attentional Dual-path Architecture
Hypergraph Neural Networks (HNNs), designed to model higher-order relations beyond pairwise interactions, typically rely either on fixed structural weights or on normalised attention. Fixed weights can propagate misleading information in heterophilic settings, while normalisation makes attention coefficients competitive, in the sense that, in a message-passing framework, increasing the weight of one sender must reduce the weight of others in the same aggregation set. Our main contribution is the introduction of PANDA (Prior-guided AttentioNal Dual-path Architecture), a plug-in for two-stage message passing HNNs applicable to both undirected and directed settings. The main path interpolates between learned receiver-normalised attention and structural incidence priors, while a non-competitive (not normalised) auxiliary path lets each sender contribute its own transformed information. We also show that, with fixed structural coefficients, PANDA recovers a certain complex-valued convolution operator adopted in previous works. Across four HNNs spanning spatial and spectral methods in both directed and undirected settings, plugging PANDA in yields an average improvement of 5.18 percentage points across eight real-world benchmark datasets. The code is available at https://github.com/jainris/panda.
Updated: 2026-10-09 14:15:14
标题: 熊猫：先验引导的关注双通道架构
摘要: 超图神经网络（HNNs）旨在模拟高阶关系，超出了成对交互作用，通常依赖于固定的结构权重或标准化注意力。固定权重在异构设置中可能传播误导信息，而标准化使注意力系数具有竞争性，即在消息传递框架中，增加一个发送者的权重必须减少同一聚合集中其他发送者的权重。我们的主要贡献是引入PANDA（Prior-guided AttentioNal Dual-path Architecture），这是一个适用于无向和有向设置的两阶段消息传递HNNs的插件。主路径在学习的接收者标准化注意力和结构关联先验之间插值，而非竞争性（未标准化）辅助路径让每个发送者贡献其自己变换后的信息。我们还表明，在固定结构系数的情况下，PANDA恢复了先前工作中采用的某种复数值卷积算子。在涵盖了空间和频谱方法的四个HNNs中，无论是在有向还是无向设置中，插入PANDA都使八个真实世界基准数据集的平均改善率提高了5.18个百分点。代码可在https://github.com/jainris/panda上找到。
更新时间: 2026-10-09 14:15:14
领域: cs.LG
Humanize Olympiad Agents: A Multi-Session Workflow for Olympiad Problem Solving with Solve, Review, Wiki, Verify
Large language model (LLM) agents are rapidly improving on long-horizon scientific and olympiad tasks. Recent systems have produced a Lean-formalized resolution of the Navier--Stokes problem and autonomous solutions to four open questions in the Erdős Conjectures database. Yet neither system releases both its underlying model and complete inference stack. More broadly, a single agent session remains brittle on difficult problems: early errors can propagate, missing domain knowledge can stall progress, and apparent success can mask a misinterpretation or weakening of the original task. We introduce HOA (Humanize Olympiads Agent), an open, flow-level framework that coordinates multiple agent sessions through three complementary feedback channels. First, an independent reviewer evaluates each candidate and provides actionable feedback. Second, a structured wiki supplies task-specific knowledge that may be missing or unreliable in model pretraining. Third, a verifier checks generated Lean proofs against trusted challenge statements. With GPT-5.6 Sol, HOA achieves perfect scores on the IMO, IPhO, and IOI 2026, near-full marks on the IChO 2026 problem set, and solves all 672 problems (100%) on PutnamBench. On the Lean-Eval leaderboard, HOA ranks first, solving 251 of 304 problems (82.6%). We further evaluate HOA with open models. Using Kimi-K3, HOA achieves full scores on the IMO, IPhO, and IOI, and a gold-level score of 97.1% on the IChO. These results closely match the closed-source GPT-5.6 Sol + /Goal baseline on the IMO and IPhO, while exceeding it on the IChO and IOI, demonstrating that systematic feedback can help bridge substantial differences in base-model capability.
Updated: 2026-10-09 14:14:14
标题: 人性化奥林匹克代理：一个用于奥林匹克问题解决的多会话工作流程，包括解决、审查、维基和验证
摘要: 大型语言模型（LLM）代理在长期科学和奥林匹克任务上迅速提高。最近的系统已经产生了Navier-Stokes问题的Lean形式化解决方案，以及Erdős猜想数据库中四个开放问题的自主解决方案。然而，没有一个系统同时发布其基础模型和完整的推理堆栈。更广泛地说，单个代理会话在困难问题上仍然脆弱：早期错误可能会传播，缺少领域知识可能会阻碍进展，明显的成功可能会掩盖对原始任务的错误解释或削弱。我们介绍了HOA（人化奥林匹克代理），一个开放的、流水线级别的框架，通过三种互补的反馈渠道协调多个代理会话。首先，一个独立的审阅员评估每个候选人并提供可操作的反馈。其次，一个结构化的维基提供了可能在模型预训练中缺失或不可靠的特定任务知识。第三，一个验证器检查生成的Lean证明与受信任的挑战性陈述相符。使用GPT-5.6 Sol，HOA在IMO、IPhO和IOI 2026上获得满分，在IChO 2026问题集上接近满分，并在PutnamBench上解决了所有672个问题（100%）。在Lean-Eval排行榜上，HOA排名第一，解决了304个问题中的251个（82.6%）。我们进一步使用开放模型评估了HOA。使用Kimi-K3，HOA在IMO、IPhO和IOI上获得了满分，并在IChO上获得了97.1%的金牌分数。这些结果与闭源GPT-5.6 Sol + /Goal基线在IMO和IPhO上密切匹配，同时在IChO和IOI上超越它，证明系统化的反馈可以帮助弥合基础模型能力上的重大差异。
更新时间: 2026-10-09 14:14:14
领域: cs.AI
Private and Distributionally Robust Federated Multiclass Classification
Fault diagnosis in asset-intensive industries requires multiclass models trained on noisy, sensitive data collected across geographically distributed client sites. Federated learning enables collaborative training without centralizing data, but most approaches do not provide formal privacy guarantees for individual records stored at client sites and can degrade under label noise. To address these limitations, we develop a differentially private, distributionally robust framework for federated multiclass classification that provides sample-level privacy for records stored at each client. We build on a federated Wasserstein distributionally robust multiclass objective with the Crammer-Singer loss, solved using an Alternating Direction Method of Multipliers (ADMM)-based algorithm. To protect the released client updates, our method injects calibrated Gaussian noise into the proximal linearized client updates of both the local model and the local distributionally robust variable before aggregation at the central server, and we quantify the cumulative privacy loss using the moments accountant. We provide per-round sample-level $(ε,δ)$-differential privacy guarantees and an overall composed privacy guarantee across all training rounds, together with convergence results and an explicit privacy-utility tradeoff. Experiments on simulated industrial fault data and UCI benchmark datasets show that our method outperforms differentially private federated empirical risk minimization and stochastic gradient descent baselines in most tested settings, with the largest gains under strong privacy-induced noise.
Updated: 2026-10-09 14:13:47
标题: 私有和分布鲁棒的联邦多类分类
摘要: 资产密集型行业中的故障诊断需要在地理分布的客户站点收集到的嘈杂、敏感数据上进行训练的多类模型。联邦学习实现了协作训练而无需集中数据，但大多数方法并不为存储在客户站点的个体记录提供正式的隐私保证，并且在标签噪声下可能下降。为了解决这些限制，我们开发了一个差分隐私、分布鲁棒的联邦多类分类框架，为存储在每个客户端的记录提供样本级隐私。我们基于一个联邦Wasserstein分布鲁棒多类目标函数，使用基于交替方向乘法器（ADMM）的算法解决Crammer-Singer损失。为了保护发布的客户端更新，我们的方法在局部模型和局部分布鲁棒变量的近端线性化客户端更新中注入校准的高斯噪声，然后在中央服务器聚合，并使用时刻会计师量化累积隐私损失。我们提供每轮样本级$(ε,δ)$-差分隐私保证以及跨所有训练轮的整体组合隐私保证，以及收敛结果和明确的隐私-效用权衡。在模拟工业故障数据和UCI基准数据集上的实验表明，在大多数测试设置中，我们的方法在强隐私引起噪声的情况下优于差分私有联邦经验风险最小化和随机梯度下降基线，获得最大收益。
更新时间: 2026-10-09 14:13:47
领域: cs.LG,math.OC
H2Table: Hierarchical Hypergraph-Enhanced Large Language Models for Complex Table Reasoning
Tables are ubiquitous across diverse domains, yet reasoning over them remains a significant challenge for modern large language models (LLMs). Current approaches typically linearize tables into sequences, inherently overlooking their intrinsic two-dimensional and hierarchical structure. To address this, we propose H2Table (Hierarchical Hypergraph-Enhanced Table Reasoning), a novel framework that represents complex tables as hierarchical nested hypergraphs. To process this representation, we design a tailored hypergraph encoder to facilitate message passing between hyperedges (headers) and nodes (cells), thereby perceiving the semantic entailment relationships between them within complex tables. Furthermore, we introduce a set of learnable query vectors acting as a lightweight bridge to extract representative structural embeddings from the encoder into the LLM. Experimental results demonstrate that our approach effectively handles complex table question answering tasks with hierarchical nested headers. Notably, on the HiTab dataset, H2Table achieves an average improvement of 22.88% over state-of-the-art baselines on highly complex tables with a nesting depth of four. Our code is available at: https://github.com/lila120/h2table.
Updated: 2026-10-09 14:10:37
标题: H2Table：用于复杂表格推理的分层超图增强大型语言模型
摘要: 表格在各个领域中普遍存在，然而对其进行推理仍然是现代大型语言模型（LLMs）面临的重大挑战。目前的方法通常将表格线性化为序列，忽视了它们固有的二维和层次结构。为了解决这个问题，我们提出了H2Table（Hierarchical Hypergraph-Enhanced Table Reasoning），这是一个将复杂表格表示为分层嵌套超图的新框架。为了处理这种表示，我们设计了一个定制的超图编码器，以促进超边（标题）和节点（单元格）之间的消息传递，从而感知复杂表格中它们之间的语义蕴涵关系。此外，我们引入了一组可学习的查询向量，作为从编码器中提取代表性结构嵌入到LLM中的轻量级桥梁。实验结果表明，我们的方法有效处理具有层次嵌套标题的复杂表格问题回答任务。值得注意的是，在HiTab数据集上，H2Table在具有四层嵌套深度的高度复杂表格上平均提高了22.88%以上的水平。我们的代码可在以下链接找到：https://github.com/lila120/h2table。
更新时间: 2026-10-09 14:10:37
领域: cs.AI
JitterVLA: A Replay Benchmark for Motion Quality in VLA-Driven Human-Robot Manipulation
Vision-language-action (VLA) policies issue sparse targets whose arrival at the robot controller is delayed and irregular, so a reasonable predicted action can still produce abrupt motion near people. Remedies are evaluated on different streams, delays and metrics. JitterVLA replays identical commands and arrival traces through alternative controllers, isolating the controller's effect. We apply it to synthetic reaches, hand trajectories from 20 people, three open VLA pipelines (838 episodes) and a physical UR3. Interpolation lowered modeled p99 jerk for every source, by 83% (615 to 106 m/s^3) on human motion with 200 ms delay, at a 2.0 cm tracking cost, and the effect replicated on 598 new episodes; jerk-limited generation lowered jerk further at a larger cost. Entropy-discontinuity associations had opposite, replicated signs in two pipelines, and entropy gating helped in one but not the other. On the UR3, interpolation lowered accelerometer-derived jerk by 3.0-14.0% in all six exploratory pairs, while a position-derived metric did not resolve the contrast. JitterVLA supports screening before human-centred studies; code and data will be released.
Updated: 2026-10-09 14:09:55
标题: JitterVLA：一种用于VLA驱动的人机协作中运动质量的重播基准
摘要: 视觉-语言-动作（VLA）策略发布稀疏目标，其到达机器人控制器的时间延迟和不规则，因此合理的预测动作仍可能在人附近产生突然运动。对不同流、延迟和度量标准进行了治疗评估。JitterVLA通过替代控制器重播相同的命令和到达迹象，隔离控制器的影响。我们将其应用于合成到达、20人的手部轨迹、三个开放的VLA管道（838个剧集）和一个物理UR3。在具有200毫秒延迟的人体运动中，插值将每个来源的建模p99杰克降低了83%（从615到106 m/s^3），以2.0厘米的跟踪成本，效果在598个新剧集上复制；受杰克限制的生成进一步降低了杰克，但成本更高。熵-不连续性关联在两个管道中具有相反的、可复制的符号，熵门控在一个管道中有帮助但在另一个管道中没有。在UR3上，插值将加速度计导出的杰克降低了3.0-14.0%的六个探索性对，而位置导出的度量没有解决这种对比。JitterVLA支持在人类中心研究之前进行筛选；代码和数据将发布。
更新时间: 2026-10-09 14:09:55
领域: cs.RO,cs.HC,cs.LG
SWE-DeceptionBench: Benchmarking Deceptive Behaviors of LLMs Across the Software Development Lifecycle
Large Language Models (LLMs) are increasingly used throughout the software development lifecycle. This growing reliance on LLMs raises concerns about their trustworthiness, with deception posing a particularly important risk. Deceptive behavior takes the form of misleading responses or assistance in fabricating, concealing, or distorting information. To study this behavior, we introduce SWE-DeceptionBench, a benchmark for evaluating LLM deception in software engineering. The benchmark covers six lifecycle stages and two deception triggers: Active Deception, where models recognize infeasible or conflicting requirements but still present them as achievable, and Passive Deception, where models help fulfill explicit user requests to mislead others. We further introduce Neutral, Reward, and Pressure conditions to study the effect of external inducements, resulting in 720 test instances in Chinese and English. Evaluating seven mainstream LLMs, we find substantial deception even under the Neutral condition, with aggregate rates of 49.88% for Active Deception and 64.29% for Passive Deception. Deception rates vary across lifecycle stages, and external inducements affect Active and Passive Deception to different extents. We also evaluate lightweight defenses, finding that guard models miss many deceptive requests, and Self-Reminder reduces but does not eliminate deceptive responses. These findings highlight the need for more effective deception detection and mitigation methods to support trustworthy LLM-based software engineering.
Updated: 2026-10-09 14:09:09
标题: SWE-DeceptionBench：LLM在软件开发生命周期中欺骗行为的基准测试
摘要: 大型语言模型（LLMs）在整个软件开发生命周期中越来越被广泛使用。对LLMs的增长依赖引起了人们对其可信度的担忧，其中欺骗行为尤为重要。欺骗行为采取了误导性回应或协助伪造、隐瞒或扭曲信息的形式。为了研究这种行为，我们引入了SWE-DeceptionBench，一个用于评估软件工程中LLM欺骗的基准。该基准涵盖了六个生命周期阶段和两种欺骗触发器：主动欺骗，即模型识别不可行或冲突的要求，但仍将它们呈现为可实现的；被动欺骗，即模型帮助实现明确用户请求以误导他人。我们进一步引入了中立、奖励和压力条件来研究外部诱因的影响，导致中文和英文共计720个测试实例。评估了七种主流LLMs，我们发现即使在中立条件下，欺骗行为仍然相当严重，主动欺骗的总体比率为49.88%，被动欺骗的总体比率为64.29%。欺骗比率在生命周期阶段之间变化，并且外部诱因对主动和被动欺骗产生不同程度的影响。我们还评估了轻量级的防御措施，发现守卫模型错过了许多欺骗性请求，而自我提醒减少但并未消除欺骗性回应。这些发现凸显了需要更有效的欺骗检测和缓解方法来支持可信赖的基于LLM的软件工程。
更新时间: 2026-10-09 14:09:09
领域: cs.SE,cs.AI
Kernel-based External Validity of Experimental Studies
The external validity of a study is the extent to which its findings generalize to unseen experimental conditions, such as datasets. We formalize the results of a study together with its research question, and define external validity as the probability that two independent and same-sized replications of the study produce results whose maximum mean discrepancy (MMD) is below a given tolerance. This approach, centered on the choice of a space of results and of an appropriate reproducing kernel for the MMD, is flexible enough to be deployed on multiple different types of results and research questions. Additionally, we establish an asymptotic power-law relation between the size of a study and the quantiles of the MMD, which we use to estimate the desired size of an externally valid study from a small-scale preliminary study. We deploy our framework to rankings with ties and discuss kernels suited to different research questions. Experiments on existing case studies and on a simulated newly designed study illustrate the applicability of the framework. We release our code as part of the genexpy Python module.
Updated: 2026-10-09 14:08:09
标题: 基于核的实验研究的外部有效性
摘要: 这项研究的外部有效性是指其研究结果在未见实验条件（如数据集）中的泛化程度。我们将研究结果与其研究问题形式化，并将外部有效性定义为两次独立且同样规模的研究复制产生的结果的最大均值差异（MMD）低于给定容差的概率。这种方法以选择结果空间和适当的MMD再生核为中心，足够灵活，可在多种不同类型的结果和研究问题上部署。此外，我们建立了研究规模与MMD分位数之间的渐近幂律关系，用于从小规模初步研究中估算外部有效研究的所需规模。我们将框架应用于带有并列排名的研究，并讨论适用于不同研究问题的核。对现有案例研究和模拟的新设计研究进行实验证明了框架的适用性。我们将我们的代码发布为genexpy Python模块的一部分。
更新时间: 2026-10-09 14:08:09
领域: stat.ME,cs.LG,stat.ML
APEX-Harm: Evaluating Agent Safety in Shared Professional Workspaces
Professional agents increasingly work through command-line interfaces and multiple tools over shared workspaces containing files, messages, and business applications. These environments create safety risks that are difficult to capture with attacks placed at a single, known read location. We introduce APEX-Harm, a benchmark for evaluating prompt and script injection against long-horizon agents performing finance, consulting, and legal tasks in large workspaces. Each task contains 115 to 328 files, together with email, calendar, chat, document, spreadsheet, PDF, and code-execution tools exposed through MCP servers. The benchmark includes six attack categories covering static and dynamic injections, task-specific script replacement, and attacker-provided generic scripts, and evaluates attack success, exposure, and task performance. We evaluate 12 frontier models on 180 task instances per model. Opus 5 and Sonnet 5 have the lowest overall attack success rates (8% and 20%), while task-specific script replacement and attacker-provided generic script attacks remain substantially more effective, averaging 81-84% and 73% attack success, respectively. Gemini 3.8 Flash achieves the highest mean task score (0.877) while still exhibiting a 33% attack success rate. These results show that completing the requested task is not sufficient evidence of safe behavior: an agent can produce a correct deliverable while also performing an unauthorized action in the workspace.
Updated: 2026-10-09 14:05:47
标题: "APEX-Harm：评估共享职业工作空间中的代理安全性"
摘要: 专业代理商越来越多地通过命令行界面和多个工具在包含文件、消息和业务应用程序的共享工作空间中工作。这些环境会产生难以捕捉的安全风险，因为攻击者可以在已知读取位置放置攻击。我们引入了 APEX-Harm，这是一个用于评估长期视野代理商在大型工作空间中执行金融、咨询和法律任务时进行提示和脚本注入的基准。每个任务包含115到328个文件，以及通过MCP服务器公开的电子邮件、日历、聊天、文档、电子表格、PDF和代码执行工具。该基准包括六个攻击类别，涵盖静态和动态注入、任务特定脚本替换以及攻击者提供的通用脚本，并评估攻击成功率、曝光率和任务性能。我们对每个模型的180个任务实例进行评估。Opus 5和Sonnet 5具有最低的总体攻击成功率（8%和20%），而任务特定脚本替换和攻击者提供的通用脚本攻击仍然明显更有效，平均攻击成功率为81-84%和73%。Gemini 3.8 Flash实现了最高的平均任务得分（0.877），同时仍然表现出33%的攻击成功率。这些结果表明，完成被要求的任务并不足以证明安全行为：一个代理商可以产生正确的交付物，同时还在工作空间中执行未经授权的操作。
更新时间: 2026-10-09 14:05:47
领域: cs.CR,cs.CL
The One-Word Census: Answer-Choice Conformity Across 105 Language Models
When a language model must choose one answer from a large space of equally valid options, which answer does it choose, and how often is it the answer every other model chooses? Asked to "pick a word," 105 language models from more than twenty labs chose serendipity 46% of the time. We measure this convergence, and each model's share in it, with 96 single-turn prompts that each name a category with many valid one-word answers ("Name a tree."), asked eight times per model and scored by exact match, with no embeddings and no judge. A model's answer-choice surprisal is the average -log2 probability of its answers under the pooled answers of all other models. In 28 of 96 categories a single answer takes at least 80% of all answers. The concentration does not depend on small or persona-tuned models: the 87 major-lab models are at least as concentrated as the full field. Lightly post-trained and persona-tuned models are the most divergent; heavily post-trained assistants from the major labs are the most conformist. Models that avoid the modal answer mostly land on the same runner-up. Within the major providers' lineages, release order shows no panel-wide trend once model tier is controlled; GPT, Gemini, Grok and Qwen become more conformist across releases, and Claude's generation-5 releases reverse. On open checkpoints of three post-training pipelines, supervised fine-tuning is the largest step toward the field's answers. Against human category-production norms, the field is more concentrated than people in 18 of 20 shared categories. All prompts, transcripts, and code are public.
Updated: 2026-10-09 14:01:46
标题: 一词普查：105种语言模型中的答案选择一致性
摘要: 当一个语言模型必须从一个大量同样有效的选项中选择一个答案时，它会选择哪个答案，而且它是每个其他模型都选择的答案多少次？要求“选择一个词”，来自二十多个实验室的105个语言模型有46%的时间选择了serendipity。我们使用96个单回合提示来测量这种收敛，以及每个模型在其中的份额，每个提示都命名一个具有许多有效单词答案的类别（“命名一棵树”），每个模型每次回答8次，并且通过精确匹配进行评分，没有嵌入和评判。一个模型的答案选择惊讶是其答案在所有其他模型的汇总答案下的平均-log2概率。在96个类别中，有28个类别的一个答案占据了至少80%的所有答案。这种集中并不取决于小型或个性化的模型：87个主要实验室的模型至少与整个领域一样集中。轻度后训练和个性化调整的模型是最不一致的；重度后训练的助手来自主要实验室，是最顺从的。那些避免模态答案的模型大多会落在同一个替补答案上。在主要提供商的系谱中，一旦控制了模型层次，发布顺序在整个面板上没有趋势；GPT、Gemini、Grok和Qwen在发布过程中变得更加顺从，而Claude的第五代发布则相反。在三个后训练管道的公开检查点上，受监督的精细调整是朝着领域答案的最大一步。与人类类别生成规范相比，在20个共享类别中的18个类别中，领域比人更加集中。所有提示、转录和代码都是公开的。
更新时间: 2026-10-09 14:01:46
领域: cs.CL,cs.AI,cs.CY
Some remarks on the connection between generative diffusion and optimal transport
We discuss three simple connections between generative diffusion and optimal transport. First, the joint law of the data and of their Ornstein-Uhlenbeck noising is, at every time, the solution of an entropic optimal-transport problem, at a temperature set by the noise level. Backward diffusion can consequently be read as an annealing which gradually decreases the entropic regularization. Second, we study the speciation transition--how the modes of the data distribution emerge during generation--and compare how it happens in diffusion and flow matching with respect to optimal transport. Using Gaussian mixtures as high-dimensional example, we show that the dynamical instability through which diffusion and flow matching separate the modes becomes, for optimal transport, an exponentially thin boundary layer at the beginning of generation, which is rounded by entropic regularization. Third, considering a generic forward dynamics with a general confining potential, we show that the backward process driven by the score also satisfies a drift-based variational principle whose minimal cost is the entropy difference between initial and final distributions.
Updated: 2026-10-09 13:58:26
标题: 一些关于生成扩散和最优输运之间关联的评论
摘要: 我们讨论了生成扩散和最优输运之间的三种简单联系。首先，数据及其Ornstein-Uhlenbeck噪声的联合法律在每个时间点都是一个熵最优输运问题的解，温度由噪声水平设定。因此，向后扩散可以被看作是逐渐减少熵正则化的一个退火过程。其次，我们研究了物种分化转变——数据分布的模态是如何在生成过程中出现的，并比较了扩散和流匹配与最优输运的关系。以高维高斯混合模型为例，我们展示了扩散和流匹配通过动力学不稳定性分离模态的方式，在最优输运中变成了在生成开始时指数级薄的边界层，被熵正则化圆滑。第三，考虑到具有一般约束势能的通用前向动力学，我们展示了由得分驱动的向后过程也满足基于漂移的变分原理，其最小成本是初始和最终分布之间的熵差异。
更新时间: 2026-10-09 13:58:26
领域: stat.ML,cond-mat.dis-nn,cs.LG
Fast Rates for Nonstationary Weighted Risk Minimization
Weighted empirical risk minimization is a common approach to prediction under distribution drift. This article studies its out-of-sample prediction error under nonstationarity. We provide a general decomposition of the excess risk into a learning term and an error term associated with distribution drift, and prove oracle inequalities for the learning error under mixing conditions. The learning bound holds uniformly over arbitrary weight classes and accounts for the effective sample size induced by the weight vector, the complexity of the weight and hypothesis classes, and potential data dependence. We also analyze the drift error in the square loss setting and provide explicit bounds for structural-break and smoothly drifting processes. We then combine the learning and drift analysis in (auto-) regression problems with linear models, basis approximations, and neural networks. In all cases, the optimal weights are characterized by a bias--variance trade-off between learning and drift, while minimax-optimal rates (up to logarithmic factors) are recovered when specialized to unweighted and stationary settings.
Updated: 2026-10-09 13:57:29
标题: 非平稳加权风险最小化的快速率
摘要: 加权经验风险最小化是预测分布漂移下的常见方法。本文研究了非稳态条件下的样本外预测误差。我们提供了超额风险的一般分解，将其分为学习项和与分布漂移相关的误差项，并在混合条件下证明了学习误差的奥拉克不等式。学习界限在任意权重类别上均成立，并考虑了权重向量引发的有效样本大小、权重和假设类的复杂性，以及潜在的数据依赖性。我们还分析了在平方损失设置下的漂移误差，并为结构性断点和平稳漂移过程提供了明确界限。然后我们将学习和漂移分析结合到线性模型、基础逼近和神经网络中的（自动）回归问题中。在所有情况下，最佳权重由学习和漂移之间的偏差-方差权衡特征化，当专门应用于无权重和稳态设置时，能够恢复到极小极限速率（最多对数因子）。
更新时间: 2026-10-09 13:57:29
领域: stat.ML,cs.LG,math.ST
EvoDrive: Pareto Evolution for Safety-Critical Autonomous Driving via Self-Improving LLM Agents
Generating safety-critical scenarios is essential for validating and improving autonomous driving systems, yet it inherently requires maximizing adversariality to expose failures while preserving realism. Existing methods usually manage this trade-off with handcrafted heuristics, confining generation to known priors and overlooking underexplored patterns. While recent open-ended agentic evolution can push this limit, unconstrained general agents lack strict simulator grounding and tend to collapse the multi-objective tension into single-scalar maximization. Here we present EvoDrive, the first automated, LLM-based agentic evolution framework for multi-objective scenario generation. EvoDrive employs a simulator-grounded actor-critic architecture where a memory-driven actor iteratively proposes improvements to the generators and critics filter out implausible candidates, and a self-evolving world evaluator routes promising proposals to optimize simulation budgets. EvoDrive further maintains a Pareto archive of evaluated candidates to preserve diverse attack-realism trade-offs and guide future evolution via simulation feedback. Benchmark results on MetaDrive and CARLA show that EvoDrive not only significantly expands the Pareto frontier across various generators, but also produces valuable scenarios for policy training.
Updated: 2026-10-09 13:56:29
标题: EvoDrive：通过自我改进的LLM代理进行安全关键的自动驾驶帕累托进化
摘要: 生成安全关键场景对于验证和改进自动驾驶系统至关重要，然而这需要最大化对抗性以暴露故障同时保持现实感。现有方法通常通过手工启发式管理这种权衡，将生成限制在已知先验条件内，忽略了未经探索的模式。虽然最近的开放式代理进化可以推动这一限制，但无约束的通用代理缺乏严格的模拟器基础，并倾向于将多目标紧张状态转化为单标量最大化。在这里，我们提出了EvoDrive，第一个基于LLM的自动代理进化框架，用于多目标场景生成。EvoDrive采用了一个基于模拟器的演员-评论家架构，其中一个记忆驱动的演员迭代地提出改进生成器的建议，评论家过滤掉不切实际的候选者，一个自我进化的世界评估器将有前途的提案路由到优化模拟预算中。EvoDrive进一步维护了一个评估候选者的帕累托归档，以保留多样化的攻击-现实主义权衡，并通过模拟反馈指导未来的进化。在MetaDrive和CARLA上的基准结果显示，EvoDrive不仅显著扩展了各种生成器的帕累托边界，还为策略训练产生了有价值的场景。
更新时间: 2026-10-09 13:56:29
领域: cs.AI
Vocal Intensity Conversion with Continuous SPL Control via Adversarial Training
Vocal intensity conversion renders a recorded utterance at a different vocal intensity, rather than simply scaling its amplitude, while preserving its content and speaker identity. Existing systems control intensity through a reference utterance or a few discrete categories. Instead, we control the conversion with any target sound pressure level (SPL) on a continuous dB scale. A lightweight converter, trained as a conditional cycle-consistent GAN without parallel data, transforms the features of a frozen speech encoder, from which a pretrained vocoder resynthesizes the speech. We compare three representations, log-mel spectrograms, WavLM, and SpeechTokenizer, and introduce an objective protocol that measures the achieved intensity change against the target one on a calibrated corpus. With WavLM, the converter achieves a measured 83% of the target change in intensity and also shifts the fundamental frequency. A perceptual evaluation confirms that perceived vocal intensity increases with the target level on amplitude-normalized stimuli.
Updated: 2026-10-09 13:55:07
标题: 用对抗训练实现连续声压级控制的声音强度转换
摘要: 语音强度转换使记录的话语产生不同的语音强度，而不仅仅是缩放其振幅，同时保留其内容和说话者身份。现有系统通过参考话语或几个离散类别来控制强度。相反，我们使用连续的分贝刻度控制转换过程中的目标声压级（SPL）。一个轻量级的转换器，经过训练的条件循环一致的GAN，没有并行数据，转换了冻结语音编码器的特征，从中预训练的声码器重新合成了语音。我们比较了三种表示形式，对数-梅尔频谱图、WavLM和SpeechTokenizer，并介绍了一个客观的协议，该协议测量了在校准语料库上实现的强度变化与目标强度变化之间的差异。使用WavLM，转换器在强度变化的测量中达到了目标的83%，同时也改变了基频。感知评估证实，当以振幅标准化的刺激物上的目标水平增加时，感知到的语音强度也会增加。
更新时间: 2026-10-09 13:55:07
领域: eess.AS,cs.LG,cs.SD
CorrFlow: Exact and Permutation-Equivariant Flow Matching on the Correlation Elliptope
Correlation matrices combine nonlinear positive-definiteness constraints with a simultaneous row-column permutation symmetry. Their generation therefore involves two distinct problems. First, every output must remain on the correlation elliptope, irrespective of how its variables are ordered. Second, the generator must learn the domain-specific distribution occupied by empirical data inside that valid space. Existing financial image generators mix these problems: they learn ordered matrix entries and repair invalid draws afterward. We introduce CorrFlow, which separates them. A generalized Fisher transform maps full-rank correlation matrices bijectively to unconstrained coordinates, guaranteeing valid outputs without projection. A graph flow-matching network learns the empirical distribution in those coordinates and is permutation equivariant: reordering the variables only reorders the output. We compare a conditional CorrGAN reimplementation, image diffusion, and CorrFlow on $80\times80$ Taiwan equity matrices using five seeds per neural method. CorrFlow produces 100% valid draws without repair and obtains feature sliced-Wasserstein distance $.712\pm.141$, versus $4.339\pm.252$ for CorrGAN and $4.610\pm.191$ for image diffusion after their invalid outputs are projected back to the correlation space. It also improves this distance over historical lookup, while lookup retains higher local coverage. These results support the combined construction under the evaluated training budget; they do not establish superiority over fully tuned image models or isolate the contribution of each component.
Updated: 2026-10-09 13:53:24
标题: CorrFlow：在相关椭球体上精确且置换等变的流匹配
摘要: 相关矩阵将非线性正定性约束与同时的行列置换对称性结合在一起。它们的生成因此涉及两个不同的问题。首先，无论变量如何排序，每个输出都必须保持在相关椭球上。其次，生成器必须学习占据有效空间内的经验数据的特定领域分布。现有的金融图像生成器混合了这些问题：它们学习有序的矩阵条目，并在之后修复无效的抽样。我们引入了CorrFlow，它将它们分开。广义的Fisher变换将全秩相关矩阵双射地映射到无约束坐标，保证没有投影的有效输出。图流匹配网络在这些坐标中学习经验分布，并且具有置换等变性：重新排序变量只重新排序输出。我们使用每种神经方法的五个种子，在$80\times80$台湾股票矩阵上比较了有条件的CorrGAN重新实现、图像扩散和CorrFlow。CorrFlow产生100%有效的抽样，无需修复，并获得特征切片Wasserstein距离为$.712\pm.141$，而CorrGAN和图像扩散在将其无效输出投影回相关空间后的距离分别为$4.339\pm.252$和$4.610\pm.191$。它还改进了这个距离超过历史查找，而查找保留了更高的局部覆盖率。这些结果支持在评估的训练预算下的组合建设；它们不能证明优于完全调整的图像模型或分离每个组件的贡献。
更新时间: 2026-10-09 13:53:24
领域: stat.ML,cs.LG,q-fin.RM
A small cognitive core explains human behavior at scale
Psychology has proposed many principles of cognition. Each, however, was developed in isolation, leaving their combined explanatory power unknown. We assembled a small set of established principles into one model, which we call COntextual Retrieval and Encoding (CORE). CORE consists of a hierarchy of episodic memories into which prediction errors are written and whose encoding, retrieval, and forgetting depend on context. Fitted jointly to 7,036,442 choices from 137 experiments, the model predicts human choices better than black-box models and reproduces 75% of reported behavioral effects when simulated. Moreover, its latent variables show classic signatures of Bayesian learning and event segmentation. Taken together, these findings suggest that a small cognitive core accounts for much of human behavior. They also open a path toward making psychology more than the sum of its parts, by building computational models on the entire experimental record.
Updated: 2026-10-09 13:51:05
标题: 一个小的认知核心解释了人类行为的规模
摘要: 心理学提出了许多认知原则。然而，每个原则都是孤立发展的，它们的联合解释力量是未知的。我们将一小组已建立的原则组合成一个模型，我们称之为COntextual Retrieval and Encoding (CORE)。CORE包括一个由情节性记忆构成的层次结构，其中写入了预测错误，其编码、检索和遗忘取决于上下文。该模型联合拟合了来自137个实验的7,036,442次选择，比黑匣子模型更好地预测了人类选择，并在模拟时复制了75%的报告行为效应。此外，其潜在变量显示出贝叶斯学习和事件分割的经典特征。总的来说，这些发现表明，一个小的认知核心解释了人类行为的大部分。它们还为使心理学不仅仅是其部分之和的路径打开了一条道路，通过在整个实验记录上构建计算模型。
更新时间: 2026-10-09 13:51:05
领域: cs.LG
Urban Power Grid Topology and Hierarchy Identification from Open Data
Understanding the complex topology and hierarchy of urban power grid is crucial for energy prognosis, power flow management, and system resilience analysis. However, detailed grid information remains largely proprietary. This information, nevertheless, is crucial especially for analyzing the last-mile distribution networks connecting individual buildings. This paper addresses this challenge by developing an open-data-driven framework to identify urban power grid topology, from high-voltage transmission down to individual building connections. We integrate public infrastructure data (power-lines, substations, transformers, poles) to map the high and medium-voltage grid skeleton. We then leverage geospatial machine learning on OpenStreetMap building data to cluster loads of building energy demand. Upon the clustering, we develop the physical topology of the final distribution lines linking the clustered buildings and low-voltage distribution grids. Our derived grid topology incorporates physical and geographical constraints like voltage level and road networks, thus leading to reasonable topology and hierarchy of grid networks. We apply the developed framework to the districts of Alna in Oslo, Norway, and Waiheke Island in New Zealand, and we reconstruct the complete grid topology that connects all the buildings, power lines, and electricity infrastructure assets. With the developed framework, we provide a critical tool that facilitates power system analysis, e.g., power flow optimization, cascading failure simulation, and grid resilience against the penetration of distributed renewable generation. This work also opens the opportunity to build energy digital twins that fuse power assets and grid topology, geographic constraints, and fine-grained power demand profiles.
Updated: 2026-10-09 13:49:37
标题: 城市电网拓扑结构和层级识别基于开放数据
摘要: 理解城市电网的复杂拓扑结构和层次结构对于能源预测、电力流管理和系统弹性分析至关重要。然而，详细的电网信息仍然大多是专有的。然而，这些信息对于分析连接各个建筑物的最后一英里配电网络尤为重要。本文通过开放数据驱动框架来解决这一挑战，以识别城市电网拓扑结构，从高压输电到个别建筑物的连接。我们整合公共基础设施数据（输电线路、变电站、变压器、杆塔）来绘制高中压电网骨架。然后，我们利用开放街道地图建筑数据上的地理空间机器学习来对建筑物能源需求进行聚类。在聚类之后，我们开发了连接聚类建筑物和低压配电网的最终配电线路的物理拓扑结构。我们得到的电网拓扑结构包括电压等级和道路网络等物理和地理约束条件，从而导致合理的电网拓扑结构和层次结构。我们将开发的框架应用于挪威奥斯陆的Alna地区和新西兰怀赫克岛，并重建了连接所有建筑物、电力线路和电力基础设施资产的完整电网拓扑结构。通过开发的框架，我们提供了一个关键工具，促进了电力系统分析，例如电力流优化、级联故障模拟以及电力系统对分布式可再生能源渗透的弹性。这项工作还为构建融合电力资产和电网拓扑结构、地理约束和精细化电力需求概况的能源数字孪生提供了机会。
更新时间: 2026-10-09 13:49:37
领域: cs.LG
AudioTETRIS: Token Elimination via Transitional Importance Signals for Efficient Audio LLMs
Large Audio Language Models (LALMs) achieve strong performance on speech-text tasks, but standard audio encoders produce hundreds of tokens for even short audio clips. Because self-attention scales quadratically in sequence length, this inflates inference-time compute substantially. Current training-free token compression methods prune or merge tokens based solely on coarse attention or feature similarity. However, treating audio tokens as generic, isotropic features smooths over spectral transitions, destroying fine-grained phonetic transitions essential for decoding speech. In contrast, speech signals exhibit highly structured temporal patterns where information density peaks at acoustic transitions. We propose AudioTETRIS, a training-free, two-stage framework that uses the audio encoder's representation geometry as an acoustic prior for token importance before applying attention-based compression. Instead of discarding encoder structure after the encoding stage, AudioTETRIS propagates this prior directly into the LLM decoder to compute a per-token fused score. This score drives progressive top-$k$ token selection across decoder layers without requiring additional training or auxiliary models. Evaluated on Qwen2-Audio-7B, Phi-4-Multimodal, and Audio Flamingo 3, AudioTETRIS retains near-uncompressed performance on several speech tasks using only 25% of audio tokens, matches or surpasses full-token accuracy at moderate compression rates, and consistently outperforms existing training-free baselines.
Updated: 2026-10-09 13:49:22
标题: AudioTETRIS: 通过过渡重要性信号进行令牌消除，以提高音频LLMs的效率
摘要: 大型音频语言模型（LALMs）在语音文本任务上表现出色，但标准音频编码器即使对于短音频剪辑也会生成数百个标记。由于自注意力在序列长度上呈二次扩展，这会大大增加推理时间的计算量。当前的无训练标记压缩方法仅基于粗糙的注意力或特征相似性对标记进行修剪或合并。然而，将音频标记视为通用，各向同性特征会平滑频谱转换，破坏对解码语音至关重要的细粒度语音转换。相比之下，语音信号展现出高度结构化的时间模式，信息密度在声学转换处达到峰值。我们提出了AudioTETRIS，一个无训练的两阶段框架，它利用音频编码器的表示几何结构作为标记重要性的声学先验，然后应用基于注意力的压缩。AudioTETRIS在编码阶段后不会丢弃编码器结构，而是直接将这个先验传播到LLM解码器中，计算每个标记的融合分数。这个分数驱动解码器层之间的逐渐top-k标记选择，而无需额外的训练或辅助模型。在Qwen2-Audio-7B、Phi-4-Multimodal和Audio Flamingo 3上进行评估，AudioTETRIS仅使用音频标记的25%，在几个语音任务上保持接近未压缩性能，在适度的压缩率下匹配或超越全标记准确性，并始终优于现有的无训练基线。
更新时间: 2026-10-09 13:49:22
领域: cs.SD,cs.AI,eess.AS
EchoTab: Learning Measurement-Grounded Representations for Echocardiographic Diagnosis through Tabular Reconstruction
Diagnostic information in multi-view medical imaging studies, like echocardiography, is distributed across views, each capturing different aspects of cardiac structure and function. In clinical practice, these views are first interpreted through quantitative measurements, which then support specific diagnoses. However, existing echo representation learning methods do not explicitly model measurements as an intermediate representation between views and diagnosis. As a result, individual measurements are not explicitly represented in a form that downstream diagnostic tasks can selectively use, hindering accurate diagnosis. Inspired by this hierarchical clinical workflow, we introduce EchoTab, a measurement-grounded representation learning framework that represents multi-view echocardiography studies using measurement-specific tokens as an intermediate representation between echo views and diagnosis. Each measurement token selectively attends to the views relevant to its estimation and is trained through tabular reconstruction to predict the corresponding measurement value. The resulting tokens form a factorized echo representation, which encode clinically meaningful structural and functional properties and can be selectively used for downstream diagnosis. Unlike the commonly used contrastive approaches, EchoTab does not require negative samples or large batch sizes. Experiments on two datasets demonstrate that EchoTab learns effective measurement-grounded representations that improve diagnosis classification in multi-view echocardiography and its learned attention aligns with established measurement-view relationships used by cardiologists. Our code is accessible at https://github.com/michelleespranita/EchoTab.
Updated: 2026-10-09 13:49:04
标题: EchoTab：通过表格重建学习心脏超声诊断的基于测量的表示
摘要: 在多视角医学影像研究中，如超声心动图，诊断信息分布在各个视角上，每个视角捕捉心脏结构和功能的不同方面。在临床实践中，这些视角首先通过定量测量进行解读，然后支持特定的诊断。然而，现有的超声表示学习方法并没有明确将测量作为视角和诊断之间的中间表示来建模。因此，个体测量并没有以一种下游诊断任务可以选择使用的形式明确表示，从而阻碍了准确的诊断。 受这种层次化临床工作流程的启发，我们介绍了EchoTab，一个基于测量的表示学习框架，使用测量特定的标记作为超声心动图研究的中间表示，位于视图和诊断之间。每个测量标记选择性地关注与其估计相关的视图，并通过表格重建进行训练，以预测相应的测量值。结果产生的标记形成了一个因子化的超声表示，编码了临床意义的结构和功能属性，并可以被选择性地用于下游诊断。与常用的对比方法不同，EchoTab不需要负样本或大批量大小。在两个数据集上的实验证明，EchoTab学习到了有效的基于测量的表示，提高了多视角超声心动图的诊断分类，并且其学习到的注意力与心脏病专家使用的已建立的测量-视图关系相吻合。我们的代码可在https://github.com/michelleespranita/EchoTab 上获取。
更新时间: 2026-10-09 13:49:04
领域: cs.LG
A Systematic Study of Security Enforcement at Android Inter-Component Boundaries
Android's inter-process communication (IPC) mechanisms (Activities, Services, BroadcastReceivers, ContentProviders, and URI scheme handlers) form the primary interface between applications and between applications and the system. We present a systematic study of security enforcement comprising 120 evaluations (99 inter-component interface evaluations across five interface categories, 10 platform security-control evaluations, and 11 dedicated positive controls) in over 30 Google and AOSP packages on Android 17 (API 37). Using an expected-versus-observed framework under a zero-permission attacker model, we evaluate whether each boundary's enforcement mechanism functions as designed. We find 97 enforcement gaps (80.8%) and 23 enforcement holds (19.2%). From these observations, we derive a 14-category enforcement gap taxonomy organized across manifest, code, platform, and architectural layers. Within our purposively selected corpus, observed enforcement outcomes differ substantially across interface categories: evaluated Activities and BroadcastReceivers exhibit a 100% gap rate, while ContentProviders, which benefit from framework-provided defenses such as setStrict(), achieve a 27.6% holds rate. We find that all seven evaluated platform compatibility changes with security intent are either disabled or gated at a targetSdkVersion that none of the applications in our evaluated corpus targets. We document 18 custom URI scheme registrations handling sensitive data (authentication tokens, payment credentials, medical information) despite the known unverifiability of custom schemes (RFC 8252 Section 8.1). Our study includes 23 positive controls, calibration cases demonstrating that the evaluation procedure distinguishes functioning enforcement from observed enforcement failures.
Updated: 2026-10-09 13:47:34
标题: 一个对Android组件间边界安全执行的系统性研究
摘要: Android的进程间通信（IPC）机制（Activities、Services、BroadcastReceivers、ContentProviders和URI方案处理程序）形成了应用程序之间以及应用程序与系统之间的主要接口。我们对安全执行进行了系统研究，包括120次评估（跨五个接口类别的99次组件间接口评估、10次平台安全控制评估和11次专门的正向控制评估），涉及30多个Android 17（API 37）中的Google和AOSP软件包。在零权限攻击者模型下使用预期与观察到的框架，我们评估每个边界的执行机制是否按设计工作。我们发现了97个执行漏洞（80.8%）和23个执行保持（19.2%）。基于这些观察结果，我们推导出了一个在清单、代码、平台和架构层面上组织的14个类别的执行漏洞分类。在我们特意选择的语料库中，观察到的执行结果在接口类别之间有显著差异：评估的Activities和BroadcastReceivers表现出100%的漏洞率，而ContentProviders，受益于诸如setStrict()等框架提供的防御措施，实现了27.6%的保持率。我们发现所有七个受评估的平台兼容性变更与安全意图的变更都被禁用或限制在我们评估的语料库中的任何应用程序都没有目标SdkVersion。我们记录了18种处理敏感数据（身份验证令牌、支付凭据、医疗信息）的自定义URI方案注册，尽管已知自定义方案的不可验证性（RFC 8252第8.1节）。我们的研究包括23个正向控制，校准案例展示了评估程序区分工作执行和观察到的执行失败。
更新时间: 2026-10-09 13:47:34
领域: cs.CR
Surrogate Graph Partitioning for Spatial Prediction
Spatial prediction refers to the estimation of unobserved values from spatially distributed observations. Although recent advances have improved the capacity to model diverse observation types, adoption in practice remains limited in industries that demand interpretability. To mitigate this gap, surrogate models that explain black-box predictors provide a promising path toward interpretable decision making. In this study, we propose a graph partitioning problem to construct spatial segments that minimize the sum of within-segment variances of individual predictions. The assignment of data points to segments can be formulated as a mixed-integer quadratic programming problem. While this formulation potentially enables the identification of exact segments, its computational complexity becomes prohibitive as the number of data points increases. Motivated by this challenge, we develop an approximation scheme that leverages the structural properties of graph partitioning. Experimental results demonstrate the computational efficiency of this approximation in identifying spatial segments.
Updated: 2026-10-09 13:46:05
标题: 空间预测的代理图分区
摘要: 空间预测是指从空间分布的观测值中估计未观测值。尽管最近的进展提高了模拟多种观测类型的能力，但在需要可解释性的行业中的实际应用仍受限制。为了弥合这一差距，解释黑盒预测器的替代模型为可解释的决策制定提供了一个有前途的途径。在本研究中，我们提出了一个图分区问题，用于构建最小化个别预测内部方差总和的空间分段。将数据点分配给分段可以被制定为一个混合整数二次规划问题。虽然这种制定可能使得确切分段的确定成为可能，但随着数据点数量的增加，计算复杂性变得禁止。受到这一挑战的激励，我们开发了一种利用图分区的结构特性的近似方案。实验结果表明，这种近似方法在识别空间分段方面具有计算效率。
更新时间: 2026-10-09 13:46:05
领域: stat.ML,cs.LG
MENTOR: Learning from Failure with On-Policy Pedagogical Intervention
Reinforcement Learning from Verifiable Rewards (RLVR) is a key driver for Large Reasoning Models, but it is vulnerable to "zero-reward collapse", a training failure where a sudden increase in task difficulty causes the learning signal to vanish. We argue that the conventional approach of discarding these failed trajectories is wasteful, as they contain valuable diagnostic information about the model's cognitive gaps, a bright side that can be leveraged for recovery. Existing solutions, however, often regress to static, "off-policy-style" interventions like pre-defined hint libraries or perform post-hoc refinement, failing to provide the adaptive, forward-looking guidance needed to escape the collapse. To address this, we propose a paradigm shift towards on-policy pedagogical intervention and introduce MENTOR, a multi-agent framework that acts as a live mentor. MENTOR diagnoses the root cause of failure in real-time and dynamically generates targeted, instructional prompts to guide the learning agent's next attempt. As a plug-and-play module, MENTOR integrates seamlessly into existing policy optimization pipelines like GRPO without modifying the training infrastructure. Extensive experiments on math and code generation show that MENTOR effectively mitigates zero-reward collapse, outperforming state-of-the-art baselines. Notably, it achieves massive average relative gains of up to 78.3% on Llama-3.2-3B-Instruct over its base model.
Updated: 2026-10-09 13:43:49
标题: 导师：通过政策干预学习失败
摘要: 来自可验证奖励的强化学习（RLVR）是大型推理模型的关键驱动因素，但容易受到“零奖励崩溃”的影响，即训练失败，突然增加任务难度导致学习信号消失。我们认为，传统方法将这些失败的轨迹丢弃是浪费的，因为它们包含有关模型认知缺陷的有价值的诊断信息，这是可以利用来恢复的好的一面。然而，现有解决方案往往退化为静态的“离线策略风格”干预，如预定义的提示库或进行事后精细化调整，无法提供适应性的、前瞻性的指导以逃脱崩溃。为了解决这个问题，我们提出了朝向基于政策的教学干预的范式转变，并介绍了MENTOR，一个充当实时导师的多智能体框架。MENTOR实时诊断失败的根本原因，并动态生成有针对性的指导性提示，以引导学习代理的下一次尝试。作为即插即用模块，MENTOR可以无缝集成到现有的政策优化管道中，如GRPO，而无需修改训练基础设施。在数学和代码生成方面进行的大量实验表明，MENTOR有效地缓解了零奖励崩溃，表现优于最先进的基线。值得注意的是，在Llama-3.2-3B-Instruct上，它实现了高达78.3%的巨大平均相对增益，超过了其基本模型。
更新时间: 2026-10-09 13:43:49
领域: cs.AI
JEV-as-a-Judge: Accept When Confident, Escalate When Unsure
LLM-as-a-judge scales evaluation, but reasoning judges are slow and costly. We study JEV-as-a-Judge: evaluation with JEV, a decision-only judge that returns label probabilities instead of text, and whose confidence decides whether to accept its verdict or escalate to a reasoning judge. Against sixteen generative and reward-model judges, with blinded human adjudication, JEV comes within three points of GPT-6 wherever a verdict can be read off the text, at 0.36% of its fee and a 0.15-second median latency, and falls behind where the verdict must be derived, as in math, code, and logic. Its confidence marks this boundary. With a threshold frozen in advance, accepting confident verdicts and escalating the rest is 0.9 points more accurate than GPT-6 on 1,610 held-out pairs at 41% of its fee, and in a pre-specified live test on two new workloads the cascade matches GPT-6's accuracy exactly. Confidence routing weakens on style-adversarial pairs and reference-free prose; we close with a simple recipe for validating thresholds locally.
Updated: 2026-10-09 13:41:59
标题: JEV作为评判者：当有信心时接受，当不确定时升级
摘要: LLM作为评判标准进行评估，但是理性的评判者速度慢且成本高昂。我们研究了JEV作为评判者：使用JEV进行评估，JEV是一个仅返回标签概率而不是文本的决策评判者，其信心决定是否接受其裁决或升级至理性评判者。在对十六个生成模型和奖励模型评判者进行研究时，通过盲审人员裁决，JEV在可以从文本中读取裁决的地方与GPT-6相差不到三个点，费用仅为其0.36%，中位延迟为0.15秒，在需要推导裁决的地方，如数学、代码和逻辑等领域，则落后于GPT-6。其信心标志着这个界限。通过预先设定的阈值，接受有信心的裁决并升级其余的裁决，在1610个留存对中比GPT-6准确率高0.9个点，费用为其41%，并在两个新工作负载的预先指定的实时测试中，级联模型的准确率完全匹配GPT-6的准确率。在风格对抗对和无参考的散文中，信心路由变得较弱；我们最后提供了一个简单的本地阈值验证配方。
更新时间: 2026-10-09 13:41:59
领域: cs.AI
Lower Bounds for Randomized Linear-Oracle Optimization over Strongly Convex Sets
Recent results [arXiv:2602.04378, arXiv:2602.22608] have established worst-case objective-error lower bounds of order $1/T^2$ after $T$ linear minimization oracle (LMO) calls, matching the upper bound [arXiv:1406.1305] in its dependence on $T$ and thereby resolving a long-standing open question. Upon close inspection, however, both lower bounds are somewhat brittle. The first [arXiv:2602.04378] applies only to the standard Frank--Wolfe algorithm, and, as we show, its prescribed slow exact-line-search trajectories can amplify sufficiently small initialization errors exponentially, corresponding to roughly one additional bit of initialization accuracy per iteration. The second [arXiv:2602.22608] obtains lower bounds for any deterministic LMO-based first-order method via a resisting oracle that adversarially adapts to the queries of the algorithm. In both cases it is conceivable that randomization may circumvent these lower bounds; in particular, neither establishes a lower bound for unrestricted randomized first-order LMO methods. We close this gap by proving an $Ω(L\operatorname{diam}(S)^2/T^2)$ distributional lower bound for randomized first-order linear minimization oracle methods over strongly convex sets. While our construction reuses the family of [arXiv:2602.22608], our proof characterizes the conditional distribution of the hidden permutation after adaptive queries and bounds the expected decrease of a combinatorial statistic controlling optimizer variance. We further show that the same error threshold holds with probability at least $3/4$ and yields a stopping-time tail bound.
Updated: 2026-10-09 13:40:50
标题: 随机线性-Oracle优化在强凸集合上的下界
摘要: 最近的结果[arXiv:2602.04378, arXiv:2602.22608]已经建立了在进行了$T$次线性最小化预测器（LMO）调用之后，目标误差的最坏情况下限为$1/T^2$阶的上下界[arXiv:1406.1305]，与其依赖于$T$的上界匹配，从而解决了一个长期存在的未解问题。然而，经过仔细检查，这两个下界都有些脆弱。第一个[arXiv:2602.04378]仅适用于标准的Frank-Wolfe算法，并且正如我们所展示的，其规定的缓慢精确线搜索轨迹可以指数地放大足够小的初始化误差，相当于每次迭代增加大约一个额外的初始化精度比特。第二个[arXiv:2602.22608]通过一个对抗性适应于算法查询的抵抗预测器获得了任何确定性LMO-based一阶方法的下界。在这两种情况下，随机化可能会规避这些下界；特别地，两者都没有为无限制的随机化一阶LMO方法建立下界。我们通过证明对于在强凸集合上的随机化一阶线性最小化预测器方法的$Ω(L\operatorname{diam}(S)^2/T^2)$分布下界来填补这一差距。虽然我们的构造重用了[arXiv:2602.22608]中的家族，但我们的证明表征了在自适应查询后隐藏排列的条件分布，并限制了控制优化器方差的组合统计量的期望减少。我们进一步展示了以至少$3/4$的概率获得相同的误差阈值，并提供了一个停止时间尾部界。
更新时间: 2026-10-09 13:40:50
领域: math.OC,cs.LG
Probabilistic Forecasting of Electric Vehicle Charging in a Time-varying Hierarchical Infrastructure
Uncertainty quantification for electric vehicle (EV) charging demand is essential to turn the increasing pressure of EV adoption on power systems into a source of flexibility. We study a large-scale dataset of public charging stations in Scotland, where demand exhibits a hierarchical structure, organized across multiple aggregation levels, from individual charging points to regional and national scales. The charging infrastructure is continuously evolving, with new stations deployed and existing ones occasionally removed, which calls for forecasting and uncertainty quantification methods tailored to this evolving structure. To address these challenges, we combine hierarchical forecasting with recent advances in conformal prediction with missing data. The conformal procedures considered produce valid prediction intervals at every node of the hierarchy and the algorithm CP-MNR-Nested successfully accounts for the time-varying hierarchical structure of the data.
Updated: 2026-10-09 13:39:08
标题: 电动汽车充电的概率预测在时变的分层基础设施中
摘要: 电动汽车（EV）充电需求的不确定性量化对于将EV采用对电力系统产生的增加压力转化为灵活性来源至关重要。我们研究了苏格兰公共充电站的大规模数据集，其中需求呈现出分层结构，跨越多个聚合层次，从个别充电点到区域和全国规模。充电基础设施不断发展，新站点部署和现有站点偶尔被移除，这需要针对这种不断发展的结构定制的预测和不确定性量化方法。为解决这些挑战，我们将分层预测与最新的缺失数据的符合预测方法相结合。所考虑的符合程序在层次结构的每个节点都产生有效的预测区间，而算法CP-MNR-Nested成功地考虑了数据的时变分层结构。
更新时间: 2026-10-09 13:39:08
领域: stat.AP,cs.LG,stat.ML
CASCADE: An Agentic Regulatory Network Framework for Patient-Data-Validated Downstream Perturbation Prediction
CASCADE is an agentic framework that predicts downstream transcriptional effects of gene perturbation from precomputed ARACNe regulatory networks, exposed via MCP. Prior work validates such tools by checking whether predicted genes are known cancer genes (membership); we instead test whether the predicted direction of change matches reality, using focal-gene copy-number amplification as a dosage-based proxy for the inverse of knockdown against real TCGA patient tumor data. For MYC, CASCADE's predicted knockdown targets show strong concordance with real amplified-vs-non-amplified tumor expression across three cancer types (BRCA: 90.0%, COAD: 72.0%, STAD: 85.7%; all p<=0.0013), well above permutation baselines, surviving a PAM50 subtype control and replicating in an independent cohort (METABRIC, 87.2%). Compared against curated MSigDB gene-set baselines via Fisher's exact test, CASCADE's accuracy is not shown to exceed existing public knowledge of MYC- or E2F-driven biology, though its gene-specific direction-calling clearly outperforms a naive uniform guess. Extending to fifteen additional genes, validation proves gene-specific rather than universal: proliferation-machinery regulators mostly replicate, while lineage-identity transcription factors and one cyclin-D paralog (CCND2) consistently fail, a pattern we discuss as a hedged, post-hoc hypothesis. We separately benchmark whether an LLM-based agent correctly grounds natural-language requests into CASCADE's real MCP tool calls. Across 35 queries, a documented local model reaches 71.4% exact match (85.7% for a larger model); schema and gene-alias failures are resolved by scale or server-side correction, but both models confidently default to the wrong perturbation type on ambiguous queries, a failure a targeted fix could not resolve because its trigger condition never occurs.
Updated: 2026-10-09 13:38:23
标题: CASCADE：一种基于患者数据验证的下游扰动预测的代理调控网络框架
摘要: CASCADE是一个代理框架，可以从预先计算的ARACNe调控网络中预测基因扰动的下游转录效应，通过MCP公开。先前的工作通过检查预测基因是否为已知癌症基因（成员资格）来验证这类工具；相反，我们测试预测的变化方向是否与现实相符，使用基因拷贝数增加作为基于剂量的反向对抗真实TCGA患者肿瘤数据的代理。对于MYC基因，CASCADE预测的敲除靶标与三种癌症类型（BRCA：90.0％，COAD：72.0％，STAD：85.7％）的真实放大与非放大肿瘤表达显示出强烈的一致性（所有p值<=0.0013），远高于排列基线，在PAM50亚型控制下生存并在一个独立队列（METABRIC，87.2％）中复制。通过Fisher's exact test与经过精心筛选的MSigDB基线比较，CASCADE的准确性未显示超过现有公开MYC或E2F驱动生物学知识，尽管其基因特异性方向判断明显优于朴素的统一猜测。将验证扩展到其他十五个基因，验证证明基因特异性而非通用性：大多数增殖机制调节因子大多复制，而谱系标识转录因子和一个cyclin-D同源体（CCND2）一直失败，这是我们讨论的一种犹豫不决的事后假设模式。我们单独评估了LLM-based代理是否正确将自然语言请求转化为CASCADE的真实MCP工具调用。在35个查询中，一个记录的局部模型达到71.4％的精确匹配（对于一个更大的模型为85.7％）；模式和基因别名失败通过规模或服务器端校正解决，但两个模型在模糊查询中都自信地默认为错误的扰动类型，一个有针对性的修复无法解决，因为其触发条件从未发生。
更新时间: 2026-10-09 13:38:23
领域: cs.AI
RARA: Robust Policy Fine-Tuning through Retrieval-Anchored Representation Alignment for Robotic Manipulation
Pre-trained robot policies, such as Vision-Language-Action (VLA) models, provide a strong foundation for generalist robotic manipulation, but adapting them to new tasks with only a few demonstrations remains challenging. Directly fine-tuning a pre-trained policy often achieves strong in-distribution performance, yet it can distort the representations learned during pre-training and weaken the model's generalization ability under out-of-distribution shifts. We propose Retrieval-Anchored Representation Alignment (RARA), a two-stage adaptation method, to address this issue. First, RARA retrieves representation features from pretraining with similar task intent and motion, and aligns the policy's representations of the target demonstrations with the pre-trained representations of the retrieved demonstrations. Second, RARA fine-tunes the policy on the target demonstrations while limiting how much its representations of the retrieved demonstrations change during fine-tuning. By reducing the representation mismatch before action learning and preserving pre-trained knowledge during adaptation, RARA limits unnecessary representation drift during fine-tuning. Experiments with Diffusion Policy and GR00T on 13 tasks in RoboTwin, RoboCasa, and on a real robot show that RARA improves the average success of fine-tuning by 9.5 points in-distribution and 13.6 points out-of-distribution, and outperforms all compared baselines. Project page: https://shua-kang.github.io/RARA_robot/
Updated: 2026-10-09 13:31:20
标题: RARA：通过检索锚定的表示对机器人操作进行稳健的策略微调
摘要: 预训练的机器人策略，例如Vision-Language-Action（VLA）模型，为通用机器人操作提供了坚实的基础，但在仅有少数示范的情况下，将它们适应到新任务仍然具有挑战性。直接对预训练策略进行微调通常可以实现强大的分布内性能，但可能会扭曲在预训练期间学到的表示，并在分布外转移时削弱模型的泛化能力。我们提出了一种两阶段适应方法——检索锚定表示对齐（RARA），以解决这个问题。首先，RARA从具有类似任务意图和运动的预训练中检索表示特征，并将目标示范的策略表示与检索示范的预训练表示对齐。其次，RARA在目标示范上微调策略，同时限制其在微调期间对检索示范的表示进行多大程度的改变。通过在动作学习之前减少表示不匹配并在适应过程中保留预训练知识，RARA限制了微调期间不必要的表示漂移。在RoboTwin、RoboCasa和真实机器人上进行的13项任务上，使用Diffusion Policy和GR00T进行的实验表明，RARA将微调的平均成功率分布内提高了9.5个点，分布外提高了13.6个点，并优于所有比较的基线。项目页面：https://shua-kang.github.io/RARA_robot/
更新时间: 2026-10-09 13:31:20
领域: cs.RO,cs.CV,cs.LG
Bolzano: From Expert-Guided Proof Search to Automated Open-Problem Solving
Large language models are increasingly contributing to mathematical research, where progress often depends on efficient proof search, incremental improvements and careful verification. We describe Bolzano, a multi-agent open-source system that uses parallel prover agents with a verifier agent and maintains a human-readable research state. Initial manual use on expert-selected problems yielded 8 results whose proofs were checked by domain experts. Motivated by these case studies, we ran Bolzano without problem-specific human guidance on about 3,800 open problems extracted from four sets of papers, solving about 200 open problems. One experiment used papers accepted to STOC 2026, a top conference in theoretical computer science. There, we answered four questions raised in the papers, as confirmed by their authors.
Updated: 2026-10-09 13:27:33
标题: 波尔扎诺：从专家引导的证明搜索到自动化的开放问题解决
摘要: 大型语言模型越来越多地为数学研究做出贡献，其中进展通常取决于高效的证明搜索、增量改进和仔细验证。我们描述了Bolzano，这是一个使用并行证明代理和验证代理的多代理开源系统，并维护人类可读的研究状态。在专家选择的问题上进行的初步手动使用产生了8个结果，这些结果的证明由领域专家检查。受这些案例研究的启发，我们在大约3800个来自四套论文中提取的开放问题上运行了Bolzano，解决了大约200个开放问题。其中一个实验使用了被STOC 2026接受的论文，这是理论计算机科学顶级会议。在那里，我们回答了论文中提出的四个问题，并得到了作者的确认。
更新时间: 2026-10-09 13:27:33
领域: cs.AI,cs.CL
Preference-Conditioned Multi-Objective Reinforcement Learning for Tactical Decision Making in Heavy-Duty Vehicles
A central difficulty in applying reinforcement learning to real-world applications such as autonomous driving is that conventional scalar reward formulations obtained by aggregating multiple often competing objectives may obscure the structure of their trade-offs. These challenges are particularly relevant to decision-making for heavy-duty vehicles, which requires balancing safety, efficiency, and operational costs under demanding commercial requirements. We present a preference-conditioned multi-objective reinforcement learning framework that explicitly represents these objectives and utilizes Generalized Policy Improvement and Proximal Policy Optimization to learn policies capturing different trade-offs between them. The proposed method approximates the Pareto front across three conflicting objectives: safety, quantified in terms of collisions and successful completion; energy efficiency, quantified using energy cost; and time efficiency, quantified using driver cost. We evaluate the framework in a scalable highway traffic simulation environment tailored for tactical decision making in heavy-duty trucks. It yields smooth and interpretable Pareto fronts, and evaluations across traffic densities and against an analytical baseline show that the learned policies capture meaningful trade-offs. The preference-conditioned policy enables transitions between different driving behaviors without retraining, demonstrating the potential of multi-objective reinforcement learning for adaptive decision making in autonomous heavy-duty vehicles.
Updated: 2026-10-09 13:22:43
标题: 偏好条件下的重型车辆战术决策多目标强化学习
摘要: 将强化学习应用于现实世界应用（如自动驾驶）的一个关键困难在于，通过汇总多个往往相互竞争的目标获得的传统标量奖励公式可能会掩盖它们之间的权衡结构。这些挑战特别适用于重型车辆的决策制定，因为在苛刻的商业要求下需要在安全性、效率和运营成本之间取得平衡。我们提出了一种偏好条件化的多目标强化学习框架，明确表示这些目标，并利用广义政策改进和近端政策优化来学习捕捉它们之间不同权衡的策略。所提出的方法在三个冲突目标之间逼近帕累托前沿：安全性，以碰撞和成功完成来量化；能源效率，以能量成本来量化；时间效率，以驾驶成本来量化。我们在专为重型卡车战术决策制定而定制的可扩展公路交通仿真环境中评估了该框架。它产生了平滑且可解释的帕累托前沿，对于不同交通密度的评估以及与分析基线的对比表明，学习到的策略捕捉了有意义的权衡。偏好条件化策略使得在不重新训练的情况下可以在不同的驾驶行为之间进行过渡，展示了多目标强化学习在自主重型车辆中适应性决策制定方面的潜力。
更新时间: 2026-10-09 13:22:43
领域: cs.LG,cs.AI,eess.SY
Constructing Pareto Frontiers of Optimal Reasoning Termination Methods for Overthinking in Reasoning Models
Reasoning models can benefit from test-time scaling, but continued reasoning may become ineffective or harmful, a phenomenon termed overthinking. Evaluations against full thinking compare accuracy and token usage but lack references for overthinking severity and strategies' theoretical mitigation potential. To characterize overthinking at the model-task level, we first introduce Step-wise Truncation Probing (STP), repeatedly generating answers from multiple retained prefixes along each fixed sampled reasoning trajectory. Using each trajectory's earliest measured position attaining maximum empirical accuracy as a reference, we define Overthinking Ratio and Accuracy Degradation as the fraction of thinking tokens generated afterward and the accuracy decrease from this reference to full thinking, respectively. Secondly, we then formulate dynamic routing, fixed budgets, and dynamic early exit within a unified framework to quantify the theoretical mitigation potential of different termination strategies. Based on this unified framework, we use the complete STP measurements to construct a token-accuracy Pareto frontier for each strategy under a common normalized thinking-token budget. We further develop OTBench, a reusable benchmark built from the STP process, which produced 70.5B reasoning tokens and the corresponding probing measurements across six models and four benchmark groups.
Updated: 2026-10-09 13:22:37
标题: 构建理论模型中过度思考的最佳推理终止方法的帕累托前沿
摘要: 推理模型可以从测试时间缩放中受益，但持续的推理可能会变得无效或有害，这种现象被称为过度思考。针对完整思考的评估比较准确性和令牌使用，但缺乏关于过度思考严重程度和策略理论缓解潜力的参考。为了表征模型-任务级别的过度思考，我们首先引入Step-wise Truncation Probing（STP），沿着每个固定采样的推理轨迹重复生成来自多个保留前缀的答案。利用每条轨迹达到最大实证准确性的最早测量位置作为参考，我们定义过度思考比例和准确性下降，分别为生成之后的思考令牌比例和从该参考到完整思考的准确性下降。其次，我们进一步制定动态路由、固定预算和动态早期退出，统一框架内量化不同终止策略的理论缓解潜力。基于这一统一框架，我们利用完整的STP测量来为每种策略在共同标准化思考令牌预算下构建令牌-准确性帕累托前沿。我们进一步开发了OTBench，这是一个由STP过程构建的可重用基准，生成了70.5B推理令牌和相应的探测测量，涵盖了六个模型和四个基准组。
更新时间: 2026-10-09 13:22:37
领域: cs.AI
When Should the Teacher Move? Temporal Coupling and Stability in Self On-Policy Distillation
Self on-policy distillation trains a student policy against a teacher derived from its own parameter history, yet the teacher's update schedule -- which governs the \emph{temporal coupling} between teacher and student -- has not been systematically studied as a stability variable. Through a controlled schedule sweep on Qwen3-8B, we establish that \emph{isolation periods}, defined as complete teacher freezing between updates, are the key structural property enabling stable learning, not teacher age. To characterize these underlying training dynamics, we introduce a diagnostic framework of temporal KL structure, refresh shock, and length-tail risk. This framework further uncovers \emph{state-oblivious collapse}: optimal short-horizon fixed schedules catastrophically fail under long-horizon training because a clock-driven refresh can copy a transiently drifting student into the teacher in a single, irreversible step. This failure mode is invisible under short-horizon evaluation and mechanistically distinct from EMA's chronic contamination. To address this, we propose \emph{Consolidation-Gated Teacher Refresh} (CGTR), which preserves isolation periods while gating each refresh on joint evidence of reward improvement and length-tail safety, ensuring every teacher movement responds to genuine student consolidation rather than a clock signal. With a single shared parameter set and no per-dataset retuning, CGTR achieves \textbf{zero collapse} and the best final score on all four tasks (Chemistry, Biology, Physics, ToolUse), self-regulating its refresh frequency to each task's learning dynamics.
Updated: 2026-10-09 13:22:21
标题: 老师何时该移动？自我政策蒸馏中的时间耦合和稳定性
摘要: 自我策略蒸馏训练一个学生策略对抗来自自身参数历史的教师，然而教师的更新计划 -- 控制教师和学生之间的时间耦合的更新计划 -- 尚未被系统地研究作为稳定变量。通过对Qwen3-8B进行受控计划调整，我们确定了完全冻结教师更新之间的“隔离期”是实现稳定学习的关键结构属性，而不是教师的年龄。为了表征这些潜在的训练动态，我们引入了一个诊断框架，包括时间KL结构、刷新冲击和长度尾风险。该框架进一步揭示了“状态无视崩溃”：在长期训练中，最佳短期固定计划会因为时钟驱动的刷新在一个单一、不可逆的步骤中将短暂漂移的学生复制到教师中而导致灾难性失败。这种失败模式在短期评估中是不可见的，机械上与EMA的慢性污染有着明显差异。为了解决这个问题，我们提出了“巩固门控教师刷新”（CGTR），它在保留隔离期的同时，在奖励改善和长度尾部安全性的共同证据上对每次刷新进行门控，确保每次教师移动都是对真实学生巩固的回应，而不是时钟信号。通过一个共享参数集和没有每个数据集的重新调整，CGTR实现了零崩溃并在所有四个任务（化学、生物、物理、工具使用）上获得了最佳最终得分，自我调节其刷新频率以适应每个任务的学习动态。
更新时间: 2026-10-09 13:22:21
领域: cs.LG,cs.AI
Preference Elicitation for Policy Optimization and Application to Aligning Heart Transplantation with Human Values
Preference elicitation is essential for aligning AI systems with human values. Prior approaches (e.g., for organ allocation) often ask stakeholders to compare the decisions of an algorithm (e.g., patient A vs. patient B). Such a decision-level approach conflates the means with the ends. Instead, we elicit preferences directly over allocation outcomes to learn a utility function for policy optimization. We construct a novel preference elicitation algorithm for linear utilities that outperforms prior techniques in practice. Our algorithm has two phases. The first phase learns cutting planes through pairwise comparisons to rapidly shrink the space of possible attribute weights and warm-starts the second phase by eliminating dominated regions. The second phase then provably converges to the user's utility function. We apply our technique to heart transplant allocation, where a policy must balance competing objectives such as post-transplant outcomes, waitlist mortality, geographic ease, and equity. Using our algorithm, we conduct a user study to learn and aggregate a human-value-aligned utility function. We evaluate the linear representation against LLM-based methods supplied with users' natural language explanations, finding that none of these methods is more accurate than the learned linear utilities. We use the resulting utility function to optimize heart transplant policies that are significantly better aligned with human values. Compared to the hindsight optimum, the status quo policy achieves a competitive ratio of just 0.66, while our method is near-optimal with a competitive ratio of 0.95.
Updated: 2026-10-09 13:18:06
标题: 偏好引导政策优化及其在将心脏移植与人类价值观对齐中的应用
摘要: 偏好引导对于调整人工智能系统与人类价值观至关重要。先前的方法（如器官分配）通常要求利益相关者比较算法的决策（例如，患者A与患者B）。这种决策级别方法混淆了手段和目的。相反，我们直接引导对分配结果的偏好，以学习政策优化的效用函数。我们构建了一种针对线性效用的新型偏好引导算法，在实践中优于先前的技术。我们的算法有两个阶段。第一阶段通过配对比较学习切割面，从而快速缩小可能属性权重空间，并通过消除支配区域来启动第二阶段。第二阶段可以证明收敛到用户的效用函数。我们将我们的技术应用于心脏移植分配，其中政策必须平衡竞争目标，如术后结果、等待名单死亡率、地理便利性和公平性。利用我们的算法，我们进行了用户研究，学习和整合符合人类价值观的效用函数。我们评估了线性表示与用户自然语言解释提供的LLM方法之间的方法，发现没有任何这些方法比学习的线性效用更准确。我们使用得到的效用函数优化了与人类价值观显著更加一致的心脏移植政策。与事后最优相比，现行政策的竞争比率仅为0.66，而我们的方法接近最优，竞争比率为0.95。
更新时间: 2026-10-09 13:18:06
领域: cs.AI,cs.LG
FACT: Fidelity-Aware Construction of Articulated Twins
Visually plausible articulated assets may still fail during contact interactions or exhibit inaccurate motion. We present FACT (Fidelity-Aware Construction of Articulated Twins), an agentic framework that progressively constructs articulated twins to improve geometry, contact, and dynamic fidelity. The agent drives an evidence--diagnosis--revision loop on a shared editable representation, selecting measurements and model edits using quantitative feedback, while numerical tools execute and validate the updates. It reconstructs editable articulated geometry from images through feature planning, targeted measurements, and diagnostic refinement. On this reference, it repairs collision proxies through task-aware local repartitioning before fidelity-constrained compression. Finally, it constructs response models from passive-response videos, using simulation residuals to guide model revision and constrained physical parameter fitting. Experiments show that FACT improves geometric reconstruction over baselines, enables more reliable interaction with simpler collision proxies, and better reproduces held-out physical responses than direct parameter inference.
Updated: 2026-10-09 12:58:36
标题: 事实：关于关节双胞胎的忠实构建
摘要: 可视化逼真的关节资产在接触交互过程中仍可能失败或展现不准确的运动。我们提出了FACT（Fidelity-Aware Construction of Articulated Twins），这是一个代理框架，逐步构建关节双体以提高几何、接触和动态保真度。该代理在共享的可编辑表示上驱动证据-诊断-修订循环，利用定量反馈选择测量和模型编辑，同时数值工具执行和验证更新。它通过特征规划、定向测量和诊断细化从图像重建可编辑的关节几何。在此基础上，它在受限制的保真度压缩之前通过任务感知的局部重新分配修复碰撞代理。最后，它从被动响应视频中构建响应模型，使用模拟残差指导模型修订和受限的物理参数拟合。实验表明，FACT相较于基线改善了几何重建，使得与更简单的碰撞代理进行更可靠的交互，并比直接参数推断更好地复制了保留的物理响应。
更新时间: 2026-10-09 12:58:36
领域: cs.RO,cs.AI
Improving Certified Robustness via Adversarial Distillation
Certified training aims to produce models whose predictions can be formally verified against adversarial perturbations, typically by optimising upper bounds on the worst-case loss over an allowed perturbation set. For neural networks, certified training methods based purely on tight relaxation bounds produce networks that are amenable to certification, but sacrifice standard accuracy. Conversely, adversarial training often yields stronger empirical robustness and standard accuracy, but the resulting models are generally difficult to certify with neural network verifiers. Recently, the literature has shown that better standard-certified accuracy trade-offs can be achieved by combining adversarial training objectives with loose over-approximations based on Interval Bound Propagation (IBP), effectively interpolating between lower and upper bounds of the worst-case loss. Building on this, we introduce AD-CERT, a certified training objective that combines adversarial distillation with an IBP upper bound. We show that distilling adversarial information over the logit space from an empirically robust teacher provides an effective lower bound surrogate for certified training, with AD-CERT achieving state-of-the-art certified performance on several robustness benchmarks for ReLU networks. Furthermore, in a unified setup, distilling adversarial information at the logit-level is shown to improve certified accuracy over a robust feature-space distillation objective by up to 5.40 percentage points. Our code has been made available at https://github.com/matteo-melis/ad-cert.
Updated: 2026-10-09 12:57:54
标题: 通过对抗蒸馏提高认证鲁棒性
摘要: 认证培训旨在产生模型，其预测可以针对对抗性扰动进行正式验证，通常是通过优化允许扰动集上的最坏情况损失的上界。对于神经网络，基于严格松弛上界的认证训练方法产生的网络易于认证，但牺牲了标准精度。相反，对抗训练通常会产生更强的经验鲁棒性和标准精度，但由此产生的模型通常难以通过神经网络验证器进行认证。最近的文献表明，通过将对抗训练目标与基于区间界传播（IBP）的松弛上界相结合，可以实现更好的标准认证精度权衡，有效地在最坏情况损失的下界和上界之间进行插值。在此基础上，我们介绍了AD-CERT，这是一个将对抗蒸馏与IBP上界相结合的认证训练目标。我们表明，从具有经验鲁棒性的教师中在逻辑空间上蒸馏对抗信息为认证训练提供了有效的下界替代品，AD-CERT在几个ReLU网络的鲁棒性基准上实现了最先进的认证性能。此外，在统一设置中，通过在逻辑级别上蒸馏对抗信息，显示可以将认证精度提高高达5.40个百分点，超过了在鲁棒特征空间蒸馏目标上的认证精度。我们的代码已经在https://github.com/matteo-melis/ad-cert 上公开。
更新时间: 2026-10-09 12:57:54
领域: cs.LG,cs.AI
Defence against the dark arts of distillation: Unsupervised discovery of subliminal learning
When a language model is distilled from a teacher that shares its base parameters, the student can inherit behavioural traits of the teacher---including misalignment---even from data that is semantically unrelated to the trait (Cloud et al., 2026). All existing defences require knowing the trait. We remove this requirement and prove that a subliminally transferred trait can be recovered from the parameter updates alone, without supervision, by exploiting a single asymmetry: the trait is expressed consistently across diverse datasets generated by the teacher, whereas task-specific and shared capabilities vary. The resulting estimator returns a weight-space direction that can be amplified to discover and interpret the trait or projected out to suppress it. Applied to the subliminal-learning benchmarks of (Cloud et al., 2026), our method recovers the transmitted animal-preference traits without any trait labels and suppresses their expression with almost no impact to task performance.
Updated: 2026-10-09 12:55:50
标题: 对抗蒸馏的黑暗艺术：无监督发现潜意识学习
摘要: 当一个语言模型被从共享基础参数的教师中提炼出来时，学生可以继承教师的行为特征，甚至在与该特征语义无关的数据中也可以继承这些行为特征（Cloud等，2026年）。所有现有的防御手段都需要了解这个特征。我们去除了这一要求，并证明一个潜意识传递的特征可以仅通过参数更新来恢复，无需监督，通过利用一个单一的不对称性：特征在教师生成的多样数据集中始终表现一致，而任务特定和共享能力会有所不同。结果估算器返回一个权重空间方向，可以放大以发现和解释特征，或者投影出来以抑制它。应用于（Cloud等，2026年）的潜意识学习基准测试中，我们的方法恢复了传递的动物偏好特征，而没有任何特征标签，并且几乎没有对任务性能产生影响。
更新时间: 2026-10-09 12:55:50
领域: stat.ML,cs.AI,cs.LG
MAGMA-Bench: Evaluating Long-Horizon Embodied Agents under Dynamic Tasks and Execution Failures
Long-horizon embodied agents must maintain task-relevant information, coordinate heterogeneous robots, and adapt their decisions as execution unfolds. Existing benchmarks investigate dynamic task execution and embodied multi-agent coordination, but provide limited evaluation of their intersection. We introduce MAGMA-Bench, an architecture-agnostic benchmark for persistent task reasoning in dynamic multi-robot environments. MAGMA-Bench evaluates agents across increasing task horizons and controlled online perturbations, including execution failures, task interruptions, and mission updates, with scenarios requiring coordination across robots with heterogeneous capabilities. It uses a structured skill-level interface to separate task-level reasoning from perception and low-level control. We evaluate locally deployable open-weight planners, multi-agent state-tracking strategies, and frontier models. Even the strongest evaluated models fail on a substantial fraction of clean episodes, and their success rates decrease further under online perturbations. Supervised adaptation greatly improves a small model, but performance remains substantially lower on longer and multi-robot missions. We release the benchmark, 261,646 task-level training examples, and an architecture-agnostic expert-trace generation pipeline.
Updated: 2026-10-09 12:54:11
标题: MAGMA-Bench：在动态任务和执行失败下评估长期视角的具身化智能体
摘要: 长期视角的具体代理必须保持与任务相关的信息，协调异构机器人，并根据执行过程中的情况调整决策。现有的基准测试研究动态任务执行和具体多代理协调，但对它们的交集进行了有限的评估。我们引入了MAGMA-Bench，这是一个面向动态多机器人环境中持久任务推理的架构无关基准测试。MAGMA-Bench通过逐渐增加的任务视野和受控的在线干扰来评估代理，包括执行失败、任务中断和任务更新，这些场景需要协调具有异构能力的机器人之间的协作。它使用结构化的技能级接口将任务级推理与感知和低级控制分离开来。我们评估了可在本地部署的开放权重规划器、多代理状态跟踪策略和前沿模型。即使是最强大的评估模型在大部分情况下也会失败，而在在线干扰下，它们的成功率进一步下降。监督式适应可以极大地改善一个小模型，但在更长和多机器人任务上性能仍然明显较低。我们发布了这个基准测试，261,646个任务级训练示例，以及一个架构无关的专家跟踪生成管道。
更新时间: 2026-10-09 12:54:11
领域: cs.AI,cs.RO
InterviewPlayground: A Validated Simulation Environment for Evaluating AI Interviewers
Increasingly, AI interviewers are being developed to elicit open-ended responses in applications like market research, public polling, preference elicitation, and social science research. However, evaluating AI interviewers is challenging because they function in extended, multi-turn interactions where they must adapt to participant behaviors. To address this need, we develop InterviewPlayground, a simulation environment for evaluating AI interviewers using simulated study participants whose behaviors are grounded in social theory. Simulated studies in InterviewPlayground produce an InterviewReportCard, which assesses the performance of AI interviewers using a suite of validated measures. To test whether our simulation-based evaluations predict performance with human participants, we conduct 15 real qualitative studies with five AI interviewers, three interview topics, and 450 human participants and compare them to simulated studies in InterviewPlayground. We find that AI interviewer performance in InterviewPlayground predicts performance in human studies with an average Pearson correlation of 0.86 across 12 measures, and the simulated interactions from InterviewPlayground reproduce key findings from behavioral analysis of AI interviewers in the human studies. Together, these findings support the validity of InterviewPlayground in assessing AI interviewer performance and examining potential failure modes. Our work contributes a simulation environment for AI interviewers supported with empirical validation, and more broadly, a roadmap for future work to develop validated, simulation-based evaluations of conversational AI systems.
Updated: 2026-10-09 12:49:05
标题: InterviewPlayground：用于评估AI面试官的经过验证的模拟环境
摘要: 越来越多的人工智能面试官正在开发，用于在市场研究、民意调查、偏好引导和社会科学研究等应用中引发开放式回应。然而，评估人工智能面试官是具有挑战性的，因为它们在延长的、多回合的互动中运作，在这种互动中它们必须适应参与者的行为。为了满足这种需求，我们开发了InterviewPlayground，这是一个模拟环境，用于评估人工智能面试官，使用模拟研究参与者，他们的行为基于社会理论。在InterviewPlayground中的模拟研究产生一个InterviewReportCard，评估人工智能面试官的表现，使用一套经过验证的指标。为了测试我们基于模拟的评估是否预测与人类参与者的表现，我们进行了15项真实的定性研究，涉及五个人工智能面试官、三个面试主题和450名人类参与者，并将它们与InterviewPlayground中的模拟研究进行比较。我们发现，在InterviewPlayground中的人工智能面试官表现预测了在人类研究中的表现，12项指标的平均皮尔逊相关系数为0.86，而InterviewPlayground中的模拟互动再现了人类研究中对人工智能面试官的行为分析的关键发现。总的来说，这些发现支持了InterviewPlayground在评估人工智能面试官表现和检查潜在故障模式方面的有效性。我们的工作为人工智能面试官提供了一个模拟环境，支持经验验证，并更广泛地为未来开发经过验证的、基于模拟的评估对话式人工智能系统提供了一个路线图。
更新时间: 2026-10-09 12:49:05
领域: cs.AI,cs.CL
Agent Step Value: Auditing Evaluation-Channel Sensitivity in Black-Box Agent Traces
Reusing evaluator-derived step rewards across evaluation channels raises a measurement question: does the assigned direction remain stable? We define Agent Step Value (ASV) as a channel-indexed before/after target-margin gain and identify the state-by-channel interaction from a matched 2x2 design. Across frozen PubMed question-answering transitions, cross-channel gain signs disagree on 48.3% of comparisons, compared with 0.3% for same-channel retries. Transition interactions are nearly balanced in sign (median +0.014; 10% trimmed mean -0.033), but asymmetric negative magnitudes produce a positive direct mean and a negative generated-view mean under the empirical target mixture. Equal-target weighting preserves a negative mean interaction but not this strict mean reversal. Matched retrieval designs localize the negative component to the generated-view coordinate and retain its direction across three sign checks that vary readout and evaluator stack. In a matched 512-token source-contract follow-up, source-only generation restores the positive mean direction on artifact-bearing retrievals and removes parser-detected support claims from artifact-free before-state views. These results support an explanation in which generated support inflates the pre-retrieval baseline. The resulting audit can localize channel sensitivity and test an intervention before step rewards are reused.
Updated: 2026-10-09 12:43:19
标题: Agent Step Value: 审计黑盒代理跟踪中的评估通道敏感性
摘要: 在评估通道之间重复使用评估者衍生的步骤奖励引发了一个测量问题：分配的方向是否保持稳定？我们将Agent Step Value (ASV)定义为一个通道索引的之前/之后目标边际收益，并从一个匹配的2x2设计中确定状态-通道交互作用。在冻结的PubMed问答转换中，跨通道增益符号在48.3%的比较中存在分歧，而同一通道的重试仅为0.3%。转换交互作用在符号上几乎是平衡的（中位数+0.014; 10%修剪平均值-0.033），但不对称的负幅度在实证目标混合下产生了一个正的直接平均值和一个负的生成视图平均值。等目标加权保留了一个负的平均交互作用，但不是这种严格的平均逆转。匹配的检索设计将负组件定位到生成视图坐标，并在变化读取和评估者堆栈的三个符号检查中保持其方向。在匹配的512令牌源合同跟进中，仅源产生将在带有工件的检索中恢复正的平均方向，并将解析器检测到的支持声明从无工件的之前状态视图中移除。这些结果支持了一个解释，即生成的支持膨胀了检索前基线。由此产生的审计可以定位通道敏感性，并在重复使用步骤奖励之前测试干预。
更新时间: 2026-10-09 12:43:19
领域: cs.AI
Beyond Domain-Routed Supervision: Cross-Teacher Guarding for Multi-Teacher On-Policy Distillation
Multi-teacher on-policy distillation (MOPD) extends on-policy distillation (OPD) by routing domain-specific data to specialized teachers, integrating their capabilities through dense token-level supervision on student-generated trajectories. However, domain routing only distinguishes the supervision sources for different domains; updates from all domains still act on shared parameters. When teachers differ substantially in response style and domain expertise, supervision that is effective in isolation may still lead to persistent gradient conflicts and capability degradation during joint optimization. To address this challenge, we propose Cross-Teacher Guarding for MOPD (Road MOPD), which introduces non-routed teachers as auxiliary cross-domain critics on the same student trajectories and coordinates multi-teacher supervision through token-level credit assignment and gradient updates. Specifically, cross-teacher-guided credit assignment uses teacher agreement to strengthen token-level corrections jointly supported by the teachers and attenuate conflicting signals; preventive multi-teacher gradient projection uses auxiliary gradients from the same trajectories to constrain potentially conflicting updates and adaptively controls projection strength to retain target-domain learning signals. In experiments, Road MOPD mitigates degradation in the disadvantaged domain, gains over 5 percentage points under high conflict, and improves overall performance in low-conflict and three-domain settings.
Updated: 2026-10-09 12:41:51
标题: 超越领域路由监督：多教师在线政策蒸馏的跨教师保护
摘要: 多教师在线政策蒸馏（MOPD）通过将领域特定数据路由到专门的教师，通过对学生生成的轨迹进行密集的标记级监督来整合它们的能力，扩展了在线政策蒸馏（OPD）。然而，领域路由仅区分不同领域的监督来源；所有领域的更新仍然作用于共享参数。当教师在响应风格和领域专业知识方面存在明显差异时，单独有效的监督可能仍会导致联合优化过程中持续的梯度冲突和能力下降。为了解决这一挑战，我们提出了适用于MOPD的跨教师守护（Road MOPD），在相同的学生轨迹上引入非路由教师作为辅助跨领域评论者，并通过标记级信用分配和梯度更新协调多教师监督。具体而言，跨教师引导的信用分配利用教师的一致性来加强由教师共同支持的标记级校正，并减弱冲突信号；预防性的多教师梯度投影使用相同轨迹的辅助梯度来约束潜在的冲突更新，并自适应地控制投影强度以保留目标领域的学习信号。在实验中，Road MOPD减轻了劣势领域的退化，在高冲突下提高了5个百分点以上，并在低冲突和三个领域设置中改善了整体表现。
更新时间: 2026-10-09 12:41:51
领域: cs.AI
Causal Effect Identifiability in Causal Abstractions: A Strict Hierarchy of Criteria
Identifying causal effects from observational data typically requires a fully specified causal graph, an assumption that is often unrealistic in complex or high-dimensional settings. Causal abstractions address this limitation by retaining only partial causal information, thereby defining equivalence classes of compatible causal graphs. In this paper, we study causal effect identifiability directly at the level of these graph collections, rather than at the level of individual causal diagrams. We formalize several notions of identifiability in this setting and organize them into a hierarchy that clarifies their relationships and differences. In particular, we establish that a causal effect may be identified over a collection of graphs without admitting a single causal derivation that holds uniformly across all compatible graphs. These results help reveal gaps and clarify ambiguities in the existing literature.
Updated: 2026-10-09 12:41:46
标题: 因果效应在因果抽象中的可识别性：严格的标准层次结构
摘要: 从观察数据中识别因果效应通常需要一个完全指定的因果图，这种假设在复杂或高维环境中往往是不现实的。因果抽象通过保留部分因果信息来解决这个限制，从而定义了兼容因果图的等价类。在本文中，我们直接研究这些图集合的因果效应可辨识性，而不是在个别因果图的级别上。我们在这个环境中形式化了几种可辨识性概念，并将它们组织到一个明确其关系和差异的层次结构中。特别是，我们建立了一个因果效应可能在一个图集合上被识别，而不承认一个在所有兼容的图中都统一适用的单一因果推导。这些结果有助于揭示现有文献中的差距并澄清歧义。
更新时间: 2026-10-09 12:41:46
领域: cs.AI
Physics-inspired Generative AI models via real hardware-based noisy quantum diffusion
Quantum Diffusion Models (QDMs) are an emerging paradigm in Generative AI that aims to use quantum properties to improve the performances of their classical counterparts. However, existing algorithms are not easily scalable due to the limitations of near-term quantum devices. Following our previous work on QDMs, here we propose and implement two physics-inspired protocols. In the first, we use the formalism of quantum stochastic walks, showing that a specific interplay of quantum and classical dynamics in the forward process produces statistically more robust models generating sets of MNIST images with lower Fréchet Inception Distance (FID) than using totally classical dynamics. In the second approach, we realize an algorithm to generate images by exploiting the intrinsic noise of real IBM quantum hardware with only four qubits. Our work could be a starting point to pave the way for new scenarios for large-scale algorithms in quantum Generative AI, where quantum noise is neither mitigated nor corrected, but instead exploited as a useful resource.
Updated: 2026-10-09 12:41:30
标题: 通过真实硬件噪声量子扩散实现的受物理启发的生成人工智能模型
摘要: 量子扩散模型（QDMs）是生成性人工智能中的一种新兴范式，旨在利用量子特性来改善其经典对应物的性能。然而，由于近期量子设备的限制，现有算法不易扩展。在我们之前关于QDMs的工作基础上，我们在这里提出并实施了两种受物理启发的协议。在第一种方法中，我们使用量子随机行走的形式主义，展示了前向过程中量子和经典动力学的特定相互作用产生了统计上更加稳健的模型，生成了一组MNIST图像，其Fréchet Inception Distance（FID）比完全使用经典动力学更低。在第二种方法中，我们实现了一种算法，通过利用IBM真实量子硬件的固有噪声来生成图像，仅使用四个量子比特。我们的工作可能是铺平道路的起点，为量子生成性人工智能中的大规模算法开辟新的场景，其中量子噪声既不被减轻也不被校正，而是被利用作为一种有用的资源。
更新时间: 2026-10-09 12:41:30
领域: quant-ph,cond-mat.dis-nn,cs.AI,cs.CV,cs.LG
RESCUE: Is This Alarm Worth Acting On? Counterfactual Evaluation of VLA Runtime Monitors
Runtime failure monitors for robot policies are typically evaluated by how accurately they distinguish failed from successful executions. Yet an accurate alarm may arrive too late for recovery, while an unnecessary intervention may disrupt an execution that would otherwise succeed. We introduce RESCUE, a counterfactual framework for evaluating whether a monitor's alarms enable useful interventions. By restoring simulator and policy states at execution checkpoints, RESCUE compares alternative actions, identifies the latest checkpoint satisfying a specified recovery criterion, and evaluates whether acting on alarms reduces expected task-completion cost. We evaluate RESCUE with OpenVLA-OFT and $π_{0.5}$ over LIBERO-Spatial and LIBERO-Plus using 128,340 counterfactual branches. Recoverability depends on the state--intervention pair rather than on robot state alone: at $τ=0.5$, 14 of 66 OpenVLA-OFT/LIBERO-Plus failures contain a recoverable checkpoint under continuation, compared with 34 of 66 under continuation or scripted recovery, while recovery reduces recoverability in other conditions. On the primary strong-policy conditions, favourable detection scores do not establish lower intervention cost. Under the evaluated stop-and-reset model, even an idealised detector can afford only a 2.0--3.3% false-alarm rate before unnecessary resets outweigh the savings from stopping failures early. These results show that monitor value depends jointly on the intervention available and the workload in which it is deployed.
Updated: 2026-10-09 12:38:35
标题: RESCUE: 运行时监控的反事实评估：这个警报值得采取行动吗？
摘要: 机器人策略的运行时故障监控通常通过准确区分失败和成功的执行来评估。然而，准确的警报可能会来得太晚无法恢复，而不必要的干预可能会干扰本来会成功的执行。我们引入了RESCUE，一个反事实框架，用于评估监控的警报是否能够实现有用的干预。通过在执行检查点处恢复模拟器和策略状态，RESCUE比较了替代行动，识别满足指定恢复标准的最新检查点，并评估是否根据警报行动可以降低预期任务完成成本。我们使用128,340个反事实分支对OpenVLA-OFT和$π_{0.5}$在LIBERO-Spatial和LIBERO-Plus上进行了RESCUE评估。可恢复性取决于状态-干预对而不仅仅是机器人状态：在$τ=0.5$时，66个OpenVLA-OFT/LIBERO-Plus失败中有14个包含在继续中的可恢复检查点，而在继续或脚本恢复下有34个可恢复，而在其他条件下恢复会降低可恢复性。在主要的强策略条件下，有利的检测分数并不能确定更低的干预成本。在评估的停止和重置模型下，即使是理想化的检测器在不必要的重置超过停止故障的节省之前也只能承受2.0-3.3%的误报率。这些结果表明，监控的价值取决于可用的干预和部署的工作负载。
更新时间: 2026-10-09 12:38:35
领域: cs.RO,cs.AI
Medical foundation models converge less under label supervision
Diagnostic classifiers and imaging biomarkers are fitted on the embeddings of medical foundation models. These models are replaced as new versions appear. This practice assumes that different models represent images alike. We tested this assumption with more than 750,000 images from 14 datasets in 5 imaging modalities, 18 public models, and 101 models trained on chest radiographs and histopathology that differ in pretraining objective, label type, size, random seed, initialization, training patients, or training recipe. Agreement was measured as the mutual k-nearest-neighbor overlap, the share of an image's nearest neighbors common to two models. Two public models shared on average 0.091 of the 10 nearest neighbors of a chest radiograph. A randomly initialized network shared 0.036 with them. In the four other modalities, they shared at most 0.300. On chest radiographs, agreement depended more on the source dataset and the radiographic view than on the findings. It was not significantly associated with accuracy. A linear mapping between the embeddings of two models, fitted on 4,096 unlabeled radiographs, transferred classifiers for 14 chest findings with 0.987 of their original area under the receiver operating characteristic curve. Convergence depended on the pretraining objective. Label-supervised models converged least at both model sizes in chest radiography and histopathology. On chest radiographs, two label-supervised models that differed only in their random seed shared 0.062 to 0.100 of their nearest neighbors. Two models trained with self-distillation, masked image modeling, or contrastive pretraining shared 0.411 to 0.802. Developers can therefore choose convergence when they choose the pretraining objective. Before replacing a model, users could test on unlabeled images whether its successor returns the same nearest neighbors.
Updated: 2026-10-09 12:36:46
标题: 医学基础模型在标签监督下收敛较少
摘要: 诊断分类器和影像生物标志物被装配在医学基础模型的嵌入上。随着新版本的出现，这些模型被替换。这种做法假设不同模型代表相似的图像。我们用来自14个数据集、5种成像方式、18个公共模型以及101个在胸部X光和组织病理学上训练的模型的超过75万张图像来测试这一假设，这些模型在预训练目标、标签类型、大小、随机种子、初始化、训练患者或训练配方上有所不同。协议度被测量为相互k最近邻重叠，即两个模型共同拥有一张图像的最近邻的比例。两个公共模型平均共享胸部X光的10个最近邻的0.091。一个随机初始化的网络和它们共享0.036。在其他四种成像方式中，它们最多共享0.300。在胸部X光上，协议度更依赖于数据集来源和放射学视图，而不是结果。它与准确性没有明显关联。在4096张未标记的X光片上，两个模型的嵌入之间的线性映射，转移了14个胸部结果的分类器，其原始受试者工作特征曲线下面积为0.987。收敛取决于预训练目标。在胸部放射学和组织病理学中，标签监督模型在两个模型尺寸上收敛效果最差。在胸部X光片上，两个仅在随机种子上有所不同的标签监督模型共享0.062至0.100个最近邻。两个通过自蒸馏、遮蔽图像建模或对比预训练训练的模型共享0.411至0.802。因此，开发者可以根据选择的预训练目标来选择收敛。在替换模型之前，用户可以在未标记图像上测试其继任者是否返回相同的最近邻。
更新时间: 2026-10-09 12:36:46
领域: cs.CV,cs.AI,cs.CL,cs.LG
Decoupling the Declarative from the Procedural in Vision-Language-Action Models
Deploying generalist robotic agents in the real world requires transferable skills. Specifically, a policy trained to clone a behavior from object-specific demonstrations must generalize beyond that object, otherwise data collection requirements become intractable. Recently, fine-tuning of pre-trained billion-parameter Vision-Language Models (VLMs), initially on large-scale robot datasets and then on fewer scenario-specific demonstrations, has emerged as the predominant paradigm for designing Vision-Language-Action (VLA) models. While these policies achieve state-of-the-art manipulation performance in-distribution, they remain brittle to minor spatial, semantic, and task variations. In this work, we address the inability of current models to decouple the declarative (i.e., concepts and entity semantics) from the procedural knowledge (i.e., how to do something) encoded in their parameters, which is a fundamental bottleneck for zero-shot skill transfer to novel objects. To address this, we propose w$^{2}$VLA, a new VLA model with restructured information flow. Rather than feeding all multimodal tokens from the VLM encoder into a large, opaque transformer-based action expert, our approach modulates the robot state sequence with visual, spatial, and skill information in a compositional and interpretable manner. Unlike popular, state-of-the-art VLAs, we show that our modular approach successfully decouples knowledge representations, enabling robust behavior cloning and unprecedented zero-shot skill transfer capabilities across dissimilar, unseen objects.
Updated: 2026-10-09 12:35:44
标题: 将视觉-语言-行动模型中的声明性与过程性分离
摘要: 在现实世界中部署通用机器人代理需要可转移技能。具体而言，一个从特定对象演示中克隆行为的策略必须超越该对象，否则数据收集需求将变得难以处理。最近，对预训练的十亿参数视觉语言模型（VLMs）进行微调，首先在大规模机器人数据集上，然后在更少的场景特定演示中进行微调，已经成为设计视觉语言动作（VLA）模型的主导范式。虽然这些策略在分布中实现了最先进的操纵性能，但它们对轻微的空间、语义和任务变化仍然很脆弱。在这项工作中，我们解决了当前模型无法将参数中编码的陈述性（即概念和实体语义）与程序性知识（即如何做某事）解耦的问题，这是零-shot技能转移到新对象的根本瓶颈。为了解决这个问题，我们提出了一种新的VLA模型w$^{2}$VLA，其信息流程进行了重组。我们的方法并非将所有来自VLM编码器的多模态令牌全部输入到一个庞大的不透明的基于变压器的动作专家中，而是以一种组合和可解释的方式调节机器人状态序列与视觉、空间和技能信息。与流行的、最先进的VLAs不同，我们展示了我们的模块化方法成功地解耦了知识表示，实现了对不同、看不见的对象之间的强大行为克隆和前所未有的零-shot技能转移能力。
更新时间: 2026-10-09 12:35:44
领域: cs.RO,cs.AI,cs.CV,cs.LG
Retrieval-Augmented Multimodal Language Model with Heterogeneous Evidence Integration for Clinical Prognosis
Clinical prognosis requires integrating structured measurements, physiological waveforms, longitudinal history, and population context. We introduce PRISM (Population-Retrieved Inference for Structured Multimodal Prognosis), a multimodal language model that combines per-feature numerical tokenization, recency-weighted aggregation of longitudinal ECGs, and a dual-modality population memory bank. Retrieval operates in representation space: the model receives mean-pooled embeddings of similar training-cohort encounters, while no individual retrieved record enters the prompt. On the 1,443-task MDS-ED benchmark, PRISM achieves AUROCs of 86.2% for diagnosis, 93.1% for deterioration, 91.8% for ICU admission, and 92.8% for mortality. These results use a 96-hour pre-arrival window unavailable to reported baselines. With benchmark-window inputs only, PRISM reaches 0.8895 AUROC, compared with 0.8937 for UniPACT and 0.8889 for MDS-ED. A matched-input compact fusion model using the same five evidence streams and frozen encoders reaches 0.7652 AUROC, versus 0.9092 for PRISM. Replacing numerical EHR encoding with text serialization reduces AUROC by 3.66 points; removing pre-arrival history reduces it by 1.97 points; and zeroing retrieval reduces it by 0.99 points. Genuine retrieval outperforms random-neighbor retrieval by 0.97 macro-average points, although a paired task-level sign test does not establish a uniform advantage (p=0.067). On the external MC-MED cohort, supervised fine-tuning reaches 0.8611 AUROC, compared with 0.8227 for locally trained gradient boosting and 0.7841 for source-only PRISM. These results support structure-aware multimodal integration and supervised external adaptation.
Updated: 2026-10-09 12:35:22
标题: 检索增强的多模态语言模型与异质证据集成用于临床预后
摘要: 临床预后需要整合结构化测量、生理波形、纵向历史和人群背景。我们引入了PRISM（结构化多模式预后的人群检索推理），这是一个结合了基于特征数值记号化、纵向心电图的最新加权聚合和双模式人群记忆库的多模态语言模型。检索操作在表示空间中进行：模型接收类似训练队列遭遇的均值池化嵌入，而没有个别检索记录进入提示。在1,443个任务的MDS-ED基准测试中，PRISM的诊断AUROC为86.2%，恶化为93.1%，ICU入院为91.8%，死亡率为92.8%。这些结果使用了对已报告基线不可用的96小时到达前窗口。仅使用基准窗口输入，PRISM达到0.8895 AUROC，而UniPACT为0.8937，MDS-ED为0.8889。使用相同的五个证据流和冻结编码器的匹配输入紧凑融合模型达到0.7652 AUROC，而PRISM为0.9092。用文本序列化替换数值EHR编码会使AUROC降低3.66个点；去除到达前历史可使其降低1.97个点；将检索设为零可使其降低0.99个点。真实检索优于随机邻居检索0.97个宏平均点，尽管成对任务级符号检验并没有建立统一优势（p=0.067）。在外部MC-MED队列中，受监督的微调达到0.8611 AUROC，而本地训练的梯度提升为0.8227，仅源PRISM为0.7841。这些结果支持结构感知的多模态整合和受监督的外部适应。
更新时间: 2026-10-09 12:35:22
领域: cs.AI,cs.CL
Beyond Temporal Proximity: Explanatory World Models for Credit Assignment in Reinforcement Learning
Generative Recommenders increasingly use Reinforcement Learning (RL) to optimize outcomes that may appear hundreds or thousands of interactions after a decision. Standard RL propagates reward according to temporal proximity rather than explicit relevance, attenuating the signal assigned to distant but important actions and creating a source of sample inefficiency. We introduce Explanatory World Models (EWM), an attribution framework that uses explanations to identify the earlier events that plausibly contributed to a realized outcome. We instantiate EWM with a frontier language model and use its explanations as learning signals using a conservative hierarchical operator that recursively redistributes reward while preserving the undiscounted episode return. To investigate what constitutes a useful explanation, we operationalize three complementary variants inspired by Halpern's actual causation, Lombrozo's explanatory simplicity, and Deutsch's hard-to-vary explanations. We evaluate EWM on two controlled long-horizon RL benchmarks, a public sequential-recommendation benchmark, and an industry-scale generative recommender. On Crafter and WordCraft, EWM substantially outperforms PPO and LaRe, with its largest gains when rewards are most delayed. On KuaiRand, EWM outperforms SFT and LaRe offline and A2C online. Finally, in an offline evaluation of an industry generative recommender, EWM considerably improves Recall and increases behavioral proxies. These results suggest that explanations can provide structured and reusable credit for learning from delayed outcomes.
Updated: 2026-10-09 12:30:14
标题: 超越时间接近性：强化学习中的信用分配的解释世界模型
摘要: 生成式推荐系统越来越多地使用强化学习（RL）来优化决策后可能出现的数百或数千次交互后的结果。标准的RL根据时间接近性传播奖励，而不是显式相关性，削弱了赋予远距但重要动作的信号，并造成样本效率低下的问题。我们引入了解释性世界模型（EWM），这是一个利用解释来识别可能导致实现结果的早期事件的归因框架。我们用一个前沿语言模型实例化EWM，并使用其解释作为学习信号，使用一种保守的分层算子，递归地重新分配奖励，同时保持未折现的回合收益。为了探讨什么构成有用的解释，我们操作化了受Halpern实际因果、Lombrozo解释简单性和Deutsch难以变异解释启发的三种互补变体。我们在两个受控的长时间跨度RL基准测试、一个公共的顺序推荐基准测试和一个行业规模的生成式推荐系统上评估了EWM。在Crafter和WordCraft上，EWM在奖励最延迟时取得了比PPO和LaRe更显著的效果。在KuaiRand上，EWM在离线时表现优于SFT和LaRe，在在线时优于A2C。最后，在对一个行业生成式推荐系统的离线评估中，EWM显著提高了召回率并增加了行为代理。这些结果表明，解释可以为学习延迟结果提供结构化和可重复使用的学分。
更新时间: 2026-10-09 12:30:14
领域: cs.LG,cs.AI,cs.IR
Defense-as-Skill: Evolving Runtime Guard Skill for Skill-Augmented Agents
Skill-augmented agents load reusable skills as persistent runtime context, improving task performance but also giving malicious skills a durable channel for steering future actions. Such skills may leak secrets, corrupt code, bypass approvals, or stage data for exfiltration only after a concrete user task and workspace state make the unsafe action appear useful. This makes pre-install vetting insufficient and calls for runtime, task-conditioned protection. We propose Defense-as-Skill, a defense paradigm that implements the runtime guard itself as an installable, inspectable, and editable skill. Our guard, SkillSonar, runs alongside untrusted task skills and checks sensitive actions against the user's task boundary, routing each action to an allow, replan, or confirmation decision without modifying the underlying agent runtime. To study this setting, we construct SCOPE-R, a task-conditioned dataset covering 6 risk families and 21 sub-categories, with 206 attack-confirmed malicious instances and 43 benign tasks. We then improve SkillSonar on the SCOPE-R training subset using runtime guard-skill evolution, a Monte-Carlo Tree Search procedure that evolves the on-disk guard skill from feedback on the rollouts. Across Claude Code and OpenClaw, the evolved guard substantially reduces attack success while maintaining a favorable safety-utility trade-off; on repeated GLM-5 runs, ID ASR drops from 0.482 to 0.104 and OOD ASR from 0.606 to 0.115, with gains transferring across victim models, held-out risk families, and external benchmarks. These results show that runtime guards can be skill-native, transparent, and evolvable policy artifacts for skill-augmented agents.
Updated: 2026-10-09 12:29:56
标题: 《作为技能的防御：为技能增强代理演化运行时防护技能》
摘要: 技能增强代理程序将可重复使用的技能加载为持久的运行时上下文，提高任务性能，但也为恶意技能提供了一个持久的渠道，用于引导未来的行动。这些技能可能会泄霏秘密，损坏代码，绕过批准，或者仅在具体用户任务和工作空间状态使得不安全行动看起来有用之后，为数据阶段化而出逃。这使得预安装审查不足，并呼吁运行时、任务条件的保护。我们提出了一种防御范式，即Defense-as-Skill，它将运行时保护本身实施为一个可安装、可检查和可编辑的技能。我们的保护程序SkillSonar与不受信任的任务技能一起运行，并根据用户任务边界对敏感行动进行检查，将每个行动路由到允许、重新规划或确认决策，而不修改底层代理程序运行时。为了研究这一设置，我们构建了一个名为SCOPE-R的任务条件数据集，涵盖了6个风险家族和21个子类别，其中包括206个攻击确认的恶意实例和43个良性任务。然后，我们利用运行时保护技能进化对SCOPE-R训练子集上的SkillSonar进行改进，这是一种蒙特卡洛树搜索过程，通过对模拟结果的反馈来演化磁盘上的保护技能。在Claude Code和OpenClaw上，进化后的保护程序显著降低了攻击成功率，同时保持了有利的安全性-效用权衡；在重复的GLM-5运行中，ID ASR从0.482下降到0.104，OOD ASR从0.606下降到0.115，这些收益在受害者模型、保留的风险家族和外部基准测试中传递。这些结果表明，运行时保护程序可以成为技能增强代理程序的技能本地、透明和可演化的策略工件。
更新时间: 2026-10-09 12:29:56
领域: cs.CR,cs.AI
Agentic Peer-to-Peer Networks: From Content Distribution to Capability and Action Sharing
The ongoing shift of Artificial Intelligence (AI) from centralized cloud application programming interfaces (APIs) to local inference on edge devices is enabling Client-Side Autonomous Agents (CSAAs) - persistent personal agents that can plan, access local context, and invoke tools on behalf of users. As these agents collaborate directly, they form Agentic Peer-to-Peer (P2P) Networks. Unlike classic overlays exchanging static, hash-indexed content (e.g., BitTorrent), agentic P2P exchanges dynamic, heterogeneous capabilities that are potentially unsafe, requiring semantic discovery, drift management, and risk-aware execution. We outline the networking foundations for such collaboration. To organize this design space, we propose a plane-based reference architecture decoupling connectivity, semantic discovery, and execution. To support discovery under drift, we introduce signed, soft-state capability descriptors to support intent-aware routing. Finally, to cope with adversarial settings, we present a tiered verification spectrum - ranging from reputation signals (Tier 1) and canary probes (Tier 2) to cryptographic evidence (Tier 3) - that aligns trust costs with task risks. Results from a proxy-level discrete-event simulation suggest improved workflow success under poisoning and drift with modest control-plane overhead.
Updated: 2026-10-09 12:29:50
标题: 主观对等网络：从内容分发到能力和行动共享
摘要: 人工智能（AI）正在从集中式云应用程序编程接口（API）转向边缘设备上的本地推理，这使得客户端自主代理（CSAA）得以实现 - 持久的个人代理，可以为用户规划、访问本地内容，并代表用户调用工具。由于这些代理直接协作，它们形成了代理对等网络（P2P）。与传统的静态、哈希索引内容的覆盖网络（例如BitTorrent）不同，代理P2P交换动态、异构能力，可能存在安全隐患，需要进行语义发现、漂移管理和风险感知执行。我们概述了这种协作的网络基础。为了组织这个设计空间，我们提出了一个基于平面的参考架构，将连通性、语义发现和执行分离。为了支持漂移下的发现，我们引入了签名的、软状态的能力描述符，以支持意图感知路由。最后，为了应对敌对环境，我们提出了一个分层验证谱系 - 从声誉信号（Tier 1）和金丝雀探针（Tier 2）到密码证据（Tier 3） - 将信任成本与任务风险对齐。代理级离散事件模拟的结果表明，在中毒和漂移情况下，工作流成功率有所提高，而控制平面开销较小。
更新时间: 2026-10-09 12:29:50
领域: cs.NI,cs.AI
Humanize MLE Agents: Multi-Agent Session Alternation Beyond Single-Agent Limits
Existing LLM-based agent approaches for long-horizon machine learning engineering (MLE) often use context engineering, implemented through custom harnesses that manage tools, skills, and accumulated experience. However, we find that as single-session execution lengthens, the agent's reasoning intensity declines alongside slowing progress, summarized as long-horizon deceleration, and the benefits of these custom harnesses vanish as the underlying model improves, summarized as harness staleness. These findings suggest that their benefits may not persist across versions, which motivates a workflow-level approach in which agents build on one another's work across sessions. We model these alternating agent sessions as a context-reset semi-Markov decision process (CR-SMDP) and establish sufficient conditions for alternation to exceed single-agent reference limits, derive an explicit experiment-budget bound, and characterize when session caps preserve this advantage. Guided by this analysis, we introduce Humanize MLE Agents (HMA), which alternates two agents on a shared workspace, limits each session to at most k experiments and resets the context between sessions, without additional task-specific harness or prior knowledge supplements. Across 75 Kaggle tasks in MLE-bench, HMA reaches a state-of-the-art any-medal rate of 78.2%, exceeding the strongest open-source baseline by 8.0 points with a quarter of its time budget. Moreover, among six configurations spanning five heterogeneous agent pairs, HMA improves any-medal rate over the constituent baselines by 6.4 points on average, demonstrating cross-pair generality. A supplementary case on kernel optimization reaches 878 cycles, within 1.6% of the community leaderboard leader, demonstrating applicability beyond MLE.
Updated: 2026-10-09 12:26:34
标题: 使MLE代理人更具人性化：超越单一代理人限制的多代理人会话交替
摘要: 现有的基于LLM的长期视野机器学习工程代理方法通常使用上下文工程，通过自定义工具、技能和积累的经验来实现。然而，我们发现随着单次执行时间的延长，代理的推理强度下降，进展放缓，总结为长期视野减速，而这些自定义工具的好处随着基础模型的改进而消失，总结为工具陈旧。这些发现表明它们的好处可能不会跨版本持续存在，这激励了一种工作流级别的方法，在这种方法中，代理在不同会话中建立在彼此的工作基础上。我们将这些交替的代理会话建模为上下文重置半马尔可夫决策过程（CR-SMDP），并建立了超越单一代理参考限制的充分条件，推导出明确的实验预算上限，并表征了会话上限何时保持这种优势。在这个分析的指导下，我们引入了Humanize MLE Agents（HMA），它在共享工作空间上交替使用两个代理，将每个会话限制在最多k次实验，并在会话之间重置上下文，没有额外的任务特定工具或先前的知识补充。在MLE-bench的75个Kaggle任务中，HMA达到了78.2%的最新奖牌率，超过了最强的开源基线8.0个百分点，仅用了四分之一的时间预算。此外，在涵盖五对异构代理的六种配置中，HMA平均提高了6.4个百分点的最新奖牌率，展示了跨对通用性。一个关于内核优化的补充案例达到了878个周期，与社区排行榜领导者相差不到1.6%，展示了适用性超越MLE。
更新时间: 2026-10-09 12:26:34
领域: cs.AI
EviRepair: Evidence-Driven Agents for Reliable Vulnerability Repair
Recent advances in large language models (LLMs) have enabled agent-based approaches for repository-level automated vulnerability repair (AVR), showing promising repair capabilities. However, an important issue has received limited attention: how can we ensure that the evidence generated during the repair process is reliable enough to support accurate vulnerability understanding and patch generation? In this paper, we propose EviRepair, an evidence-driven multi-agent framework for repository-level AVR. EviRepair introduces evidence review into the repair process by enabling agents to collaboratively collect vulnerability evidence, verify evidence consistency, and construct a reviewed vulnerability model for patch synthesis and validation. Extensive experiments on PatchEval and SecBench benchmarks with multiple agent frameworks and LLMs demonstrate that EviRepair consistently outperforms existing agent-based baselines, achieving average improvements of 16.37% and 40.33% in resolved rates on PatchEval and SecBench, respectively. Codes are available at https://github.com/lly-s/EviRepair
Updated: 2026-10-09 12:22:43
标题: EviRepair：可靠漏洞修复的基于证据的代理程序
摘要: 最近大型语言模型（LLMs）的进展使得基于代理的存储库级自动漏洞修复（AVR）方法成为可能，显示出令人期待的修复能力。然而，一个重要问题受到了有限的关注：在修复过程中生成的证据如何确保足够可靠，以支持准确的漏洞理解和补丁生成？在本文中，我们提出了EviRepair，一个基于证据驱动的存储库级 AVR 多代理框架。EviRepair通过使代理能够协作收集漏洞证据、验证证据的一致性，并构建一个经过审查的漏洞模型，用于补丁的合成和验证。在 PatchEval 和 SecBench benchmarks 上进行了广泛的实验，使用多个代理框架和 LLMs，结果表明 EviRepair 在解决率上始终优于现有的基于代理的基线，分别在 PatchEval 和 SecBench 上平均改进了 16.37% 和 40.33%。代码可在 https://github.com/lly-s/EviRepair 上找到。
更新时间: 2026-10-09 12:22:43
领域: cs.CR
Branching On-Policy Distillation
On-policy distillation (OPD) provides teacher supervision on student-generated trajectories, but a single trajectory realizes only one continuation at each decision, limiting supervision coverage under a finite sampling budget. We propose BranchOPD (Branching On-Policy Distillation), which selects informative decision points using student uncertainty and teacher-student disagreement, resamples student-policy continuations from shared prefixes, and applies token-level teacher supervision to both main trajectories and branches. BranchOPD operates at token-level decisions in mathematics and within-turn decisions in agent tasks. Shared-prefix reuse reduces student-generated token costs by approximately 19% on ALFWorld and 31% on WebShop relative to seven independent trajectories. We also introduce protocol-boundary repair (PBR), which uses targeted supervision under protocol constraints to mitigate repetition and wasted generation. BranchOPD outperforms Vanilla OPD and Multi-Rollout OPD (n=7) across all evaluated benchmarks, with gains over Vanilla OPD of 18.55 percentage points in mean mathematics accuracy over 16 sampled answers (avg@16) and 13.8 points in mean ALFWorld ID/OOD success. Mechanistic analyses and selection ablations show that branching at uncertain decisions expands subsequent behavior coverage and outperforms random selection and turn-initial branching.
Updated: 2026-10-09 12:13:34
标题: 基于策略的分支蒸馏
摘要: On-policy distillation (OPD)提供了对学生生成的轨迹的教师监督，但是单个轨迹在每个决策点只能实现一种延续，从而在有限的采样预算下限制了监督覆盖范围。我们提出了BranchOPD（Branching On-Policy Distillation），它利用学生的不确定性和教师-学生的分歧选择信息决策点，从共享前缀重新采样学生策略的延续，并对主轨迹和分支都应用基于标记的教师监督。BranchOPD在数学的标记级别决策和代理任务的回合内决策中运行。相对于七个独立轨迹，共享前缀重用在ALFWorld上减少了约19%的学生生成的标记成本，在WebShop上减少了约31%。我们还介绍了协议边界修复（PBR），它利用协议约束下的有针对性的监督来减少重复和浪费的生成。BranchOPD在所有评估基准中均优于Vanilla OPD和Multi-Rollout OPD（n=7），在平均16个样本答案（avg@16）的数学准确性上超过Vanilla OPD 18.55个百分点，在ALFWorld ID/OOD成功率上超过13.8个百分点。机械分析和选择消融表明，在不确定决策点进行分支扩展了后续行为覆盖范围，并且优于随机选择和回合初始分支。
更新时间: 2026-10-09 12:13:34
领域: cs.AI
A Local Agentic AI Framework to Generate Edge Case Scenarios for Safety Assurance
Generating edge case scenarios is crucial for training, testing and safety assuring autonomous systems. This identifies hazardous failures to increase system reliability and robustness, ensures the system can still function safely on the boundaries of its operational design domain and learns the expected system behaviour for these edge cases. The rapid advancement in the generation and analysis capabilities of language models (LMs) has expanded their applicability and recent work has investigated LM scenario generation. However, many recent approaches rely on human-in-the-loop or rule-based strategies. This restricts the LM's ability to generate diverse and edge case scenarios; the long tail of the distribution. Our work uses local LMs in a laptop-based agentic AI framework for adaptive scenario generation. This is cheap, fully-controlled, ensures data privacy and security. It uses think, act, observe, update, and continue loops with lightweight JSON encoding to generate scenarios. We use the system requirements, system description and environment description as knowledge bases in contextual prompts for an LM to ensure scenario realism. A second, independent LM critiques each scenario using domain specific criteria so it can be refined. The refined scenarios are run in simulation and evaluation metrics collected. These metric scores are analysed and similar scenarios are created for edge cases (exploitation) or novel scenarios are generated to explore the scenario space. Post-hoc clustering using metric scores and thematic-type analysis of the rich-text scenario descriptions produced by the LM pinpoints functional gaps and their triggering conditions. This creates traceable evidence for the safety assurance case. Our work highlights the potential of a flexible and adaptive local agentic AI framework for automating edge case generation and producing traceable safety assurance evidence.
Updated: 2026-10-09 12:13:17
标题: 一个用于生成边缘案例场景以确保安全的本地代理人AI框架
摘要: 生成边缘案例场景对于训练、测试和确保自主系统的安全至关重要。这有助于识别危险故障，提高系统的可靠性和鲁棒性，确保系统仍然可以在其操作设计领域的边界上安全运行，并学习这些边缘案例的预期系统行为。语言模型（LMs）生成和分析能力的快速发展扩大了它们的适用性，最近的研究已经调查了LM场景生成。然而，许多最近的方法依赖于人为干预或基于规则的策略。这限制了LM生成多样化和边缘案例场景的能力；分布的长尾。我们的工作在基于笔记本的智能AI框架中使用本地LMs进行自适应场景生成。这种方法价格便宜，完全可控，确保数据隐私和安全性。它使用思考、行动、观察、更新和继续循环以轻量级JSON编码生成场景。我们利用系统要求、系统描述和环境描述作为LM的上下文提示的知识库，以确保场景的真实性。第二个独立的LM根据领域特定标准对每个场景进行评价，以便进行改进。改进后的场景在模拟中运行，并收集评估指标。这些指标分数被分析，类似的场景被创建用于边缘案例（开发），或者生成新颖的场景来探索场景空间。使用指标分数进行事后聚类和由LM生成的丰富文本场景描述的主题类型分析，可以确定功能缺陷及其触发条件。这为安全保障案例提供了可追溯的证据。我们的工作突显了灵活和自适应的本地智能AI框架的潜力，用于自动化边缘案例生成并产生可追溯的安全保障证据。
更新时间: 2026-10-09 12:13:17
领域: cs.AI,cs.RO
When MLLMs miss what humans read: diagnosing and learning to read squint text
Multimodal LLMs can fail to read text embedded in an image's coarse structure (text a human recovers by squinting) and report that no text is present. We name this class Squint Text and study it as a recognition deficiency, motivated by cases from a production moderation pipeline. We introduce SquintBench, comprising 1076 production-derived moderation images, and SquintCtrl, a controlled synthetic suite for mechanism diagnosis. Frozen-weight analysis shows that hidden text leaves a coarse signal while glyph identity is weakly represented; identity becomes substantially more accessible when the correct spatial regions are supplied, implicating spatial selection as a major bottleneck. Suppressing high-frequency content also improves reading at a fixed visual-token budget, while visual token count may also contribute. These findings motivate Squint OPSD: a frozen teacher receives a downscaled view and the ground-truth transcription as privileged context, and its token-level distributions supervise a student along on-policy trajectories. At inference, the student reads the original image alone. Squint OPSD reaches 88.61 ANLS on SquintBench, versus 26.50 for the unadapted backbone, 24.44 for the best general-purpose model on the original view, and 68.60 at best for an off-the-shelf model under a downscaled view. These results substantially narrow the gap between available visual evidence and its generated read-out without auxiliary views at inference.
Updated: 2026-10-09 12:02:22
标题: 当MLLMs错过人类阅读的内容时：诊断和学习阅读斜视文本
摘要: 多模态LLMs可能无法读取嵌入在图像粗糙结构中的文本（通过眯眼可以恢复的文本），并报告未发现文本。我们将这一类别称为眯眼文本，并将其作为一种识别缺陷进行研究，这是受到生产审核流程中的案例的启发。我们引入了SquintBench，包括1076个生产衍生的审核图像，以及SquintCtrl，一个用于机制诊断的受控合成套件。冻结权重分析表明，隐藏文本会留下一个粗糙信号，而字形标识表示较弱；当提供正确的空间区域时，标识变得更容易获得，暗示空间选择是一个主要瓶颈。抑制高频内容也可以在固定的视觉代币预算下改善阅读，而视觉代币数量也可能起作用。这些发现促使了Squint OPSD：一个冻结的教师接收一个缩小的视图和地面真实的抄写作为特权上下文，其代币级分布沿着策略轨迹监督一个学生。在推断时，学生仅读取原始图像。Squint OPSD在SquintBench上达到了88.61 ANLS，而未经调整的骨干网络为26.50，在原始视图上的最佳通用模型为24.44，在缩小视图下的现成模型最佳为68.60。这些结果在推断时极大地缩小了可用视觉证据与其生成的读出之间的差距，而不需要辅助视图。
更新时间: 2026-10-09 12:02:22
领域: cs.CV,cs.AI
SqueezeBench: Saturation-Resilient Evaluation of Mathematical Capability
AI systems are making rapid progress in mathematics, but this progress is hard to measure. Benchmarks of solved problems saturate, and in benchmarks of open problems each problem can be resolved for the first time only once. We present SqueezeBench, a benchmark that is resilient to saturation: a success raises the bar instead of removing the problem. Each problem fixes an integer function $f(n)$ whose values are known only for small $n$, and the benchmark keeps a lower and an upper record on it. A record is a computable Lean function with a machine-checked proof that it bounds $f$. A submission proposes a new function and proves that it bounds $f$, that it is no worse than the current record at any $n$, and that it is strictly better at some $n$. Once accepted, it becomes the record that the next submission has to beat. The system thus chooses the bound as well as its proof, and it may improve the record at a single $n$ or its growth rate. At every $n$, the gap between the two records shows how far the certified bounds are from determining $f(n)$. Every proof is checked by two independent Lean kernels, so no referee needs to check the mathematics. The checks also reject a function that computes $f$ by exhaustive search at small $n$, and a maintainer only reads the submitter's report and looks for such a search at larger $n$. We release ten problems from extremal combinatorics with twenty proved initial records and the admission tools. As a baseline, agents based on a frontier language model, with at most one hour per record, had 82 submissions accepted and improved every record, without going beyond the published literature. Agents based on two newer models then started from these records, had 60 more submissions accepted and improved 19 of the 20 records again. For eight of the ten problems, at least one record is still asymptotically behind the best published bound.
Updated: 2026-10-09 11:59:46
标题: SqueezeBench: 数学能力的饱和恢复评估
摘要: 人工智能系统在数学领域取得了快速进展，但这种进展很难衡量。已解决问题的基准已经饱和，而在未解决问题的基准中，每个问题只能在第一次解决时解决。我们提出了SqueezeBench，这是一个对饱和具有弹性的基准：成功会提高标准而不是消除问题。每个问题都固定了一个整数函数$f(n)$，其值仅在小$n$时已知，基准会保留它的一个下限和一个上限记录。记录是一个可计算的Lean函数，具有一个经过机器检查的证明，证明它限制了$f$。提交提出一个新函数，并证明它限制了$f$，在任何$n$时都不比当前记录差，并且在某些$n$时严格更好。一旦被接受，它就成为下一个提交者必须超越的记录。系统因此选择了边界以及其证明，并且可以改进单个$n$上的记录或其增长速率。在每个$n$处，两个记录之间的差距显示了认证边界与确定$f(n)$有多远。每个证明都由两个独立的Lean内核检查，因此无需裁判检查数学。检查还会拒绝通过在小$n$处进行穷举搜索来计算$f$的函数，维护者只需阅读提交者的报告，并查看在更大的$n$处是否存在这样的搜索。我们发布了来自极端组合学的十个问题，其中有二十个经过证明的初始记录和准入工具。作为基准，基于前沿语言模型的代理，每个记录最多一个小时，有82份提交被接受，并改进了每个记录，而不超出已发表的文献。然后，基于两个更新模型的代理从这些记录开始，有60份提交被接受，再次改进了20个记录中的19个。对于十个问题中的八个，至少一个记录仍然在渐近地落后于最佳已发表边界。
更新时间: 2026-10-09 11:59:46
领域: cs.AI,cs.LG,cs.LO
Agentic Large Language Models
Large language models (LLMs) have taken the world by storm. Their ability to communicate with us in our own language, and to reason about our world, has made them a favorite tool for writing, answering questions, and much more. Their answers already influence our decisions and our actions. What would happen if LLMs themselves could act in the world directly? What would happen if LLMs could interact with each other? Agentic LLMs combine the power of large language models with the ability to act in the real world. In this book we will discuss how agentic LLMs work, how they act, how they interact, and what this means for us, as users.
Updated: 2026-10-09 11:57:04
标题: 主动大型语言模型
摘要: 大型语言模型(LLMs)已经风靡全球。它们能够用我们自己的语言与我们交流，并且能够推理我们的世界，使它们成为写作、回答问题等工具的首选。它们的答案已经影响我们的决定和行动。如果LLMs本身能够直接在世界中行动会发生什么？如果LLMs能够相互交互会发生什么？机动LLMs将大型语言模型的能力与在现实世界中行动的能力结合在一起。在这本书中，我们将讨论机动LLMs是如何工作的，它们如何行动，它们如何相互交互，以及这对我们作为用户意味着什么。
更新时间: 2026-10-09 11:57:04
领域: cs.AI
From SGD to Muon: Adaptive Optimization via Schatten-p Norms
Modern optimizers, like Muon, impose matrix-wise geometry constraints on their updates. These matrix-wise constraints can be unified under Linear Minimization Oracle (LMO) theory. However, all current methods impose fixed LMO geometries for the update rules, chosen by-design or empirically, which are not necessarily optimal according to the problem's geometry. We introduce a novel efficient datadriven criterion for dynamically choosing proxy-optimal update LMO geometries on individual Deep Neural Network layers. Derived in closed form from gradient and activation statistics using a single-step random feature regression surrogate model, our criterion navigates a design space interpolating from SGD to Muon updates. Moreover, integrating parameter-wise preconditioning allows our framework to recover SGD, Muon, Adam, and MuAdam as specific extrema. To make this adaptive approach scalable, we pair it with efficient computational strategies, achieving only a $\sim$ 3% runtime overhead on highly optimized baselines. As a proof of concept, we show that this data-driven optimizer beats or remains competitive with the performance of the best performing optimizer between Muon and AdamW across three different training scenarios. Ultimately, this work provides evidence that LMO geometry can be successfully and efficiently adapted from runtime data, opening a new pathway for optimizer design beyond static geometries.
Updated: 2026-10-09 11:47:28
标题: 从SGD到Muon：通过Schatten-p范数的自适应优化
摘要: 现代优化器，如Muon，对其更新施加矩阵几何约束。这些矩阵约束可以统一在线性最小化Oracle（LMO）理论下。然而，所有当前方法都对更新规则施加固定的LMO几何，这些几何是由设计或经验选择的，不一定是根据问题的几何性质最优的。我们引入了一种新颖的高效数据驱动准则，用于动态地选择个别深度神经网络层的代理最优更新LMO几何。从梯度和激活统计数据中使用单步随机特征回归代理模型推导出的封闭形式，我们的准则在设计空间中从SGD到Muon更新进行插值。此外，整合参数级预处理使我们的框架能够恢复SGD、Muon、Adam和MuAdam作为特定的极值。为了使这种自适应方法可扩展，我们将其与高效的计算策略配对，仅在高度优化的基线上产生大约3%的运行时开销。作为概念验证，我们展示了这种数据驱动优化器在三种不同的训练场景中击败或保持与Muon和AdamW之间表现最佳的优化器的竞争力。最终，这项工作证明了LMO几何可以成功且高效地从运行时数据中调整，为优化器设计开辟了一条超越静态几何的新途径。
更新时间: 2026-10-09 11:47:28
领域: cs.AI
Collective Autonomous Vehicle Routing Needs Social Awareness for Coexistence with Human Drivers
Collective Autonomous Vehicle (AV) route choices in real-world traffic networks can create externalities for Human-Driven Vehicles (HDVs), as routes that benefit AVs may increase travel times for other road users. To address this, we introduce a social component, the marginal cost, to the selfish travel time AV reward. Marginal cost captures an AV's impact on other road users by comparing their travel times with those in a counterfactual simulation in which the AV is removed. We study AV route choice in mixed-autonomy traffic settings using Multi-Agent Reinforcement Learning (MARL), where AVs learn routing policies through repeated interactions with the traffic environment. Experiments on the real-world networks of Ingolstadt and Provins show that incorporating marginal cost into the AV reward not only limits the increase in HDV travel times and shortens AV travel times relative to selfish AV routing, but also improves overall system performance. In Ingolstadt, selfish routing of 40\% AVs increases the travel times of other vehicles by 1.7\%, and this drops to 1.1\% when the social reward is included. Interestingly, not only does overall system performance improve, but the AVs are also better off when acting socially.This effect is consistent across tested cities, penetration rates, and replications, yet when the weight of the social component in the reward increases, the system fails to reach the desired performance. For a more scalable approach to social AV routing in mixed-autonomy traffic, we compute the marginal cost only for the AVs whose route choices most affect the traffic system. This reduces the number of counterfactual simulations and achieves similar performance at a significantly lower computational cost. Our results suggest that AV routing should go beyond selfish rewards and, for the benefit of all traffic participants, include a social component in its objective.
Updated: 2026-10-09 11:37:47
标题: 集体自主车辆路由需要社会意识以与人类司机共存
摘要: 在现实世界的交通网络中，集体自主车辆（AV）的路线选择可能会对人类驾驶车辆（HDV）产生外部性，因为有利于AV的路线可能会增加其他道路使用者的行驶时间。为了解决这个问题，我们引入了一个社会组成部分，即边际成本，将其纳入自私行驶时间AV奖励中。边际成本通过将AV的行驶时间与在AV被移除的对照模拟中的行驶时间进行比较，来捕捉AV对其他道路使用者的影响。 我们使用多智能体强化学习（MARL）研究了混合自治交通设置中的AV路线选择，其中AV通过与交通环境的重复交互学习路由策略。对Ingolstadt和Provins的真实网络进行的实验表明，将边际成本纳入AV奖励不仅限制了HDV行驶时间的增加，并且相对于自私的AV路由，也缩短了AV行驶时间，同时提高了整体系统性能。在Ingolstadt，40% AV的自私路由使其他车辆的行驶时间增加了1.7％，而当包含社会奖励时，这一数字降至1.1％。有趣的是，不仅整体系统性能得到改善，而且AV在社会行为时也更加受益。这种效果在测试的城市、渗透率和复制中是一致的，但是当奖励中的社会组成部分权重增加时，系统无法达到期望的性能。 为了在混合自治交通中实现更具可扩展性的社会AV路由方法，我们仅计算对交通系统影响最大的AV的边际成本。这减少了对照模拟的数量，并在显著降低计算成本的情况下实现了类似的性能。我们的结果表明，AV路由应该超越自私奖励，并为了所有交通参与者的利益，将社会组成部分纳入其目标中。
更新时间: 2026-10-09 11:37:47
领域: cs.MA,cs.AI
Rollout-Time Intervention: Unlocking Agentic Reinforcement Learning Beyond Policy Limits
Reinforcement learning (RL) has emerged as a key paradigm for improving large language model (LLM) agents, enabling stronger reasoning, planning, and tool-use in interactive environments. However, existing agent RL methods remain fundamentally constrained by an on-policy reasoning bottleneck: when tasks exceed the model's current capability boundary, effective trajectories become extremely sparse and training often collapses. We introduce Rollout-Time Intervention (RTI), a technique designed to elicit successful trajectories beyond the policy's intrinsic exploration capacity. RTI operates by leveraging reward-side evidence (i.e., evaluation rubrics) during rollout to convert erroneous trajectories into successful ones, providing effective learning signals for tasks that the policy rarely solves on its own. Although intervention trajectories are inherently off-policy, we find that they can effectively guide policy optimization when trajectories with excessive distributional shifts are filtered out. Extensive experiments across five benchmarks demonstrate that RTI consistently outperforms RL baselines, enabling the post-trained policy to approach the performance of substantially stronger contemporary LLMs and, more importantly, transcend its original capability boundary. The code is available at https://github.com/Alibaba-NLP/qqr.
Updated: 2026-10-09 11:37:09
标题: 部署时间干预：突破政策限制，解锁主体强化学习
摘要: 强化学习（RL）已经成为改善大型语言模型（LLM）代理的关键范式，使其在交互环境中具有更强的推理、规划和工具使用能力。然而，现有的代理RL方法仍然受到一个基于政策的推理瓶颈的制约：当任务超出模型当前能力边界时，有效的轨迹变得极度稀疏，训练往往会崩溃。我们引入了Rollout-Time Intervention（RTI），这是一种旨在引发成功轨迹超出政策固有的探索能力的技术。RTI通过利用奖励端的证据（即评估标准）在展开过程中将错误的轨迹转化为成功的轨迹，为政策很少自行解决的任务提供有效的学习信号。尽管干预轨迹在本质上是离线政策，我们发现当具有过多分布变化的轨迹被过滤掉时，它们可以有效地指导政策优化。通过对五个基准测试的广泛实验表明，RTI始终优于RL基线方法，使后训练的政策能够接近更强大的现代LLM的性能，并且更重要的是，超越了其原始能力边界。代码可在https://github.com/Alibaba-NLP/qqr找到。
更新时间: 2026-10-09 11:37:09
领域: cs.AI
Local Diagnostics of Learned Event Representations for Anomaly Detection at the HL-LHC
Anomaly detection, the identification of collision events that deviate from Standard Model expectations, is one of the most promising model-agnostic strategies for new-physics searches at the Large Hadron Collider. Searching for rare events in learned low-dimensional representations removes the need to commit to hand-crafted observables a priori, but sacrifices the immediate physical interpretation those observables provide. For anomaly detection to enter a discovery pipeline, flagged events must therefore be diagnosed locally, both to distinguish potential new physics from mismodeling or detector effects and to identify observables for targeted follow-up searches. We evaluate supervised and self-supervised neural embedding strategies on COLLIDE-2V, a 750-million-event dataset simulated under realistic High-Luminosity LHC conditions, using the reconstruction error of an autoencoder trained in the learned latent space as the anomaly score. Across the configurations and benchmark signals considered, supervised objectives, which exploit simulation process labels during training, generally outperform their self-supervised counterparts and, in their best configurations, improve on an autoencoder trained directly on kinematic features, raising the signal efficiency at 10% background acceptance on all held-out processes. We further introduce a local diagnostic pipeline that fits a Gaussian Mixture Model to the SM latent space, profiles each component through high-level observables without process labels, localizes the flagged events, and ranks observables by their local discriminating power. The ranking recovers the defining kinematics of the held-out processes, while correlating the anomaly score with the same observables quantifies how closely the score tracks them, revealing cases where the most discriminating observable is not the one the score follows most closely.
Updated: 2026-10-09 11:37:06
标题: 在HL-LHC中学习事件表示的本地诊断用于异常检测
摘要: 异常检测是识别与标准模型预期偏离的碰撞事件的过程，是大型强子对撞机中寻找新物理的最有前途的与模型无关的策略之一。在学习到的低维表示中搜索稀有事件消除了事先承诺手工设计的可观察量的需要，但牺牲了这些可观察量提供的即时物理解释。为了使异常检测进入发现流程，被标记的事件因此必须在本地被诊断，旨在区分潜在的新物理与误建模或探测器效应，并确定可用于有针对性的后续搜索的可观察量。我们在COLLIDE-2V上评估了监督和自监督的神经嵌入策略，这是一个在真实高光度LHC条件下模拟的7.5亿事件数据集，使用在学习的潜变量空间中训练的自动编码器的重建误差作为异常评分。在考虑的不同配置和基准信号中，监督目标通常优于自监督目标，并且在它们最好的配置中，提高了直接在动力学特征上训练的自动编码器在所有保留过程上的10%背景接受度的信号效率。我们进一步引入了一个本地诊断流程，该流程拟合了一个高斯混合模型到SM潜变量空间中，通过高级可观察量对每个组件进行描述，而无需过程标签，定位被标记的事件，并根据它们的本地区分能力对可观察量进行排名。该排名恢复了保留过程的定义动力学，同时将异常评分与相同的可观察量相关联，量化评分如何紧密跟踪它们，揭示了最具区分力的可观察量并不一定是评分最密切跟踪的可观察量的情况。
更新时间: 2026-10-09 11:37:06
领域: cs.AI,hep-ex,hep-ph,physics.data-an
Towards a Measure Theory of Semantic Information
A classic account of the quantification of semantic information is that of Bar-Hiller and Carnap. Their account proposes an inverse relation between the informativeness of a statement and its probability. However, their approach assigns the maximum informativeness to a contradiction: which Floridi refers to as the Bar-Hillel-Carnap paradox. He developed a novel theory founded on a distance metric and parabolic relation, designed to remove this paradox. Unfortunately is approach does not succeed in that aim. In this paper I critique Floridi's theory of strongly semantic information on its own terms and show where it succeeds and fails. I then present a new approach based on the unit circle (a relation that has been the basis of theories from basic trigonometry to quantum theory). This is used, by analogy with von Neumann's quantum probability to construct a measure space for informativeness that meets all the requirements stipulated by Floridi and removes the paradox. In addition, while contradictions and tautologies have zero informativeness, it is found that messages which are contradictory to each other are equally informative. The utility of this is explained by means of an example.
Updated: 2026-10-09 11:34:48
标题: 走向语义信息的测度论
摘要: 巴尔-希勒和卡尔纳普关于语义信息量化的经典观点提出了陈述的信息量与其概率之间的反向关系。然而，他们的方法将最大的信息量赋予了矛盾：弗洛里迪将其称为巴尔-希勒-卡尔纳普悖论。他基于距离度量和抛物线关系开发了一个新理论，旨在消除这一悖论。不幸的是，他的方法并未成功实现这一目标。 本文对弗洛里迪关于强语义信息的理论进行了批判，并指出其成功和失败之处。然后，我提出了一种基于单位圆的新方法（这种关系已经成为从基本三角法到量子理论的基础）。通过类比冯·诺依曼的量子概率，利用这种方法构建了一个符合弗洛里迪规定的所有要求并消除了悖论的信息量测度空间。此外，虽然矛盾和重言的信息量为零，但发现彼此矛盾的信息同样具有同等的信息量。通过一个例子来解释这一点的实用性。
更新时间: 2026-10-09 11:34:48
领域: cs.IT,cs.AI
CNN-Transformer with Sigmoid Label for High-Resolution Sub-Degree DOA Estimation
This paper addresses the grid quantization error of deep-learning (DL)-based direction-of-arrival (DOA) estimation. Classification- based DOA methods discretize the continuous angular space into a grid, so that with a peak-based read-out, the accuracy and resolution are limited by the grid step and by the width of the label peak. To address this issue, we propose a sigmoid label that encodes the sources' DOAs as the sum of angle-shifted monotonic sigmoid functions. Differentiating the output of a CNN-Transformer network trained with the sigmoid label produces sharp spectral peaks at the source DOAs. Combined with a moderately refined grid step of 0.25 degrees, the CNN-Transformer with the proposed sigmoid label attains sub-degree accuracy and a 1-degree resolution, breaking the limit of the grid quantization error. Under the unified CNN-Transformer backbone, we compare the proposed Sigmoid label with the Gaussian label, the 0-1 hard label, and the binary cross-entropy (BCE) baseline. Numerical results demonstrate that the Sigmoid label attains the highest accuracy and superior resolution compared to other labels.
Updated: 2026-10-09 11:34:23
标题: 用Sigmoid标签的CNN-Transformer用于高分辨率亚度DOA估计
摘要: 本文讨论了基于深度学习（DL）的方向到达（DOA）估计中的网格量化误差。基于分类的DOA方法将连续的角度空间离散化为一个网格，因此通过基于峰值的读出，准确度和分辨率受到网格步长和标签峰宽度的限制。为了解决这个问题，我们提出了一个Sigmoid标签，将源的DOA编码为角度移位的单调Sigmoid函数之和。使用Sigmoid标签训练的CNN-Transformer网络的输出进行微分，可以在源的DOA处产生尖锐的频谱峰。结合适度精细化的0.25度网格步长，使用提出的Sigmoid标签的CNN-Transformer实现了亚度精度和1度分辨率，突破了网格量化误差的限制。在统一的CNN-Transformer骨干下，我们将提出的Sigmoid标签与高斯标签、0-1硬标签和二元交叉熵（BCE）基线进行了比较。数值结果表明，与其他标签相比，Sigmoid标签具有最高的准确度和优越的分辨率。
更新时间: 2026-10-09 11:34:23
领域: eess.SP,cs.AI
Two Birds One Stone: Effective Static Detection of Resource and Communication Deadlocks in Rust Programs
Although Rust claims to enable fearless concurrency, concurrency errors remain prevalent in practical applications, with deadlocks particularly prominent and harmful. Current static deadlock detection techniques either do not cater to Rust programs without incorporating Rust semantics or solely concentrate on resource deadlocks, disregarding communication deadlocks. In this paper, we present RcChecker, the first static detection method that simultaneously targets resource and communication deadlocks for Rust. At its core, RcChecker introduces the signal-lock graph to characterize the interactions between threads through lock variables and condition variables. The construction of the signal-lock graph involves utilizing flow-sensitive and context-sensitive analyses to track variable lifetime details and abstracting actions associated with condition variables into signal resource holdings and requests. By examining cycles within the signal-lock graph, we can effectively identify resource and communication deadlocks in a unified way. Compared to the existing Rust deadlock detection tool, RcChecker effectively eliminates eight resource deadlock false positives and detects seventeen errors that the existing tool fails to identify. Furthermore, RcChecker reports seven previously undiscovered issues in real-world Rust applications, including two resource deadlocks and five communication deadlocks.
Updated: 2026-10-09 11:26:30
标题: 一箭双雕：在Rust程序中有效静态检测资源和通信死锁
摘要: 尽管Rust声称可以实现无畏并发，但并发错误在实际应用中仍然普遍存在，死锁尤为突出和有害。目前的静态死锁检测技术要么不适用于未整合Rust语义的Rust程序，要么仅专注于资源死锁，忽略了通信死锁。本文介绍了RcChecker，这是针对Rust同时目标资源和通信死锁的第一个静态检测方法。在其核心，RcChecker引入了信号锁图来描述线程之间通过锁变量和条件变量进行交互的方式。信号锁图的构建涉及使用流敏感和上下文敏感分析来跟踪变量生命周期细节，并将与条件变量相关的操作抽象为信号资源持有和请求。通过检查信号锁图中的循环，我们可以有效地以统一的方式识别资源和通信死锁。与现有的Rust死锁检测工具相比，RcChecker有效地消除了八个资源死锁误报，并检测出现了现有工具未能识别的十七个错误。此外，RcChecker在现实世界的Rust应用程序中报告了七个先前未发现的问题，包括两个资源死锁和五个通信死锁。
更新时间: 2026-10-09 11:26:30
领域: cs.PL,cs.CR,cs.SE
Towards Safe World Models: Risks, Defenses, and Evaluation
World models support planning, control, learning from imagined experience, and model-mediated evaluation by predicting action-conditioned consequences. Their safety cannot be inferred from predictive accuracy alone: errors in observations, state and memory, dynamics, evaluation, execution, or feedback can propagate into unsafe decisions and persist through updates. This work organizes the literature on world-model safety around a six-stage lifecycle covering data and observations, state and memory, dynamics and imagination, evaluation and planning, action and control, and feedback and update. We synthesize non-adversarial failures, attacks, defenses, evaluation protocols, benchmarks, and deployment conditions, and distinguish evidence directness from outcome level. The current evidence corpus contains 163 cited works. We use the framework to identify where claims are supported, where attribution remains unresolved, and which measurements are needed to connect model-level behavior to closed-loop and operational safety.
Updated: 2026-10-09 10:45:59
标题: 朝着安全的世界模型迈进：风险、防御和评估
摘要: 世界模型支持通过预测动作条件下的后果来进行规划、控制、从想象的经验中学习和模型介导的评估。它们的安全性不能仅仅从预测准确性中推断：观察、状态和记忆、动态、评估、执行或反馈中的错误会传播到不安全的决策中，并通过更新持续存在。这项工作围绕着一个六阶段的生命周期对世界模型安全性的文献进行组织，涵盖数据和观测、状态和记忆、动态和想象、评估和规划、动作和控制以及反馈和更新。我们综合了非对抗性的故障、攻击、防御、评估协议、基准、部署条件，并区分了证据直接性和结果水平。当前的证据库中包含163个引用作品。我们利用这个框架来确定哪些声明得到支持，哪些归因仍未解决，以及需要哪些测量来将模型水平的行为与闭环和运行安全连接起来。
更新时间: 2026-10-09 10:45:59
领域: cs.CR
Secrets in Preamble: Practical and Robust PHY Information Hiding for Wi-Fi
Physical layer (PHY) information hiding refers to the concealment of secrets within PHY waveforms without compromising the integrity of the primary communication link, thereby enabling critical applications such as covert communication, PHY steganography, and watermarking. The widespread deployment of Wi-Fi has motivated substantial research interest in hiding secrets within Wi-Fi waveforms. However, existing Wi-Fi information hiding approaches suffer from two limitations: restricted steganographic capacity and insufficient robustness against environmental variations, both of which severely constrain practical deployment. This work proposes to embed secrets within the Wi-Fi legacy short training field (L-STF). The designed method preserves the autocorrelation properties of Wi-Fi L-STF, ensuring that packet detection and coarse frequency offset compensation functionalities remain uncompromised. The secret embedding and extraction processes employ a pair of neural networks designed to learn optimal embedding rules and maximize steganographic capacity. We developed a prototype system and conducted comprehensive experiments across both indoor and outdoor environments. Experimental results show that our method can reliably embed 46 secret bits within Wi-Fi packets across diverse channel conditions, achieving significantly enhanced steganographic capacity and environmental robustness. The hardware prototype study validates the feasibility of the proposed Wi-Fi information hiding scheme.
Updated: 2026-10-09 10:40:30
标题: 序言中的秘密：用于Wi-Fi的实用且稳健的PHY信息隐藏
摘要: 物理层（PHY）信息隐藏指的是在PHY波形中隐藏秘密，而不损害主要通信链的完整性，从而实现关键应用，如隐蔽通信、PHY隐写术和水印。Wi-Fi的广泛部署激发了在Wi-Fi波形中隐藏秘密的大量研究兴趣。然而，现有的Wi-Fi信息隐藏方法存在两个局限性：受限的隐写容量和对环境变化的不足鲁棒性，这两者严重限制了实际部署。本研究提出将秘密嵌入Wi-Fi传统的短训练场（L-STF）中。设计的方法保留了Wi-Fi L-STF的自相关性质，确保数据包检测和粗频偏补偿功能保持完整。秘密嵌入和提取过程采用了一对设计用于学习最佳嵌入规则并最大化隐写容量的神经网络。我们开发了一个原型系统，并在室内和室外环境中进行了全面实验。实验结果表明，我们的方法可以可靠地在各种信道条件下嵌入46个秘密比特到Wi-Fi数据包中，实现了显著增强的隐写容量和环境鲁棒性。硬件原型研究证实了所提出的Wi-Fi信息隐藏方案的可行性。
更新时间: 2026-10-09 10:40:30
领域: cs.CR,eess.SP
ECGWmark: Deep Watermarking for Electrocardiogram Signals
Secure transmission of electrocardiogram (ECG) signals remains a major challenge in telecardiology. While traditional mathematical watermarking can embed patient metadata directly into ECG signals, it is highly fragile against temporal desynchronization such as circular phase shifts. In contrast, existing deep watermarking approaches for temporal signals, such as audio, typically rely on lossy quantization, which compromises the high-frequency micro-morphological details critical for clinical diagnosis. To bridge this gap, we propose ECGWmark, a fully convolutional and quantization-free deep watermarking framework for medical time-series. ECGWmark operates in a continuous, full-resolution latent space and utilizes a localized frame-level mechanism to improve the robustness to severe desynchronization. Crucially, we introduce a differentiable Physiological Loss guided by a clinical classifier to preserve diagnostic semantics. Experiments on the benchmark MIT-BIH database show that ECGWmark maintains channel-bit extraction accuracy above 90% under the evaluated circular temporal shifts. Furthermore, under a rigorous inter-patient diagnostic evaluation, Class-V sensitivity changes from 71.97% to 76.50%, whereas Macro-F1 changes from 35.25% to 34.71% and Class-S sensitivity from 8.89% to 1.39%; these results indicate mixed class-level effects.
Updated: 2026-10-09 10:32:03
标题: ECGWmark：心电图信号的深度水印技术
摘要: 心电图（ECG）信号的安全传输仍然是远程心脏病学中的一个重要挑战。虽然传统的数学水印技术可以直接将患者元数据嵌入到ECG信号中，但对于诸如循环相移等时间上的不同步现象非常脆弱。相比之下，现有的用于时间信号（如音频）的深度水印技术通常依赖于有损量化，这会影响对临床诊断至关重要的高频微形态细节。为了弥补这一差距，我们提出了ECGWmark，这是一个专为医学时间序列设计的全卷积和无量化的深度水印框架。ECGWmark在连续的、全分辨率的潜在空间中运行，并利用局部帧级机制来提高对严重不同步的鲁棒性。关键是，我们引入了一个可微分的生理损失，由临床分类器指导，以保留诊断语义。对MIT-BIH数据库的实验表明，ECGWmark在评估的循环时间移位下保持了通道位提取准确率在90%以上。此外，在严格的患者间诊断评估中，Class-V敏感性从71.97%提高到76.50%，而Macro-F1从35.25%下降到34.71%，Class-S敏感性从8.89%下降到1.39%；这些结果表明了混合类别级别的影响。
更新时间: 2026-10-09 10:32:03
领域: eess.SP,cs.CR,cs.CV
LETHE: Defending against Unauthorized Image-to-Video Generation via Value Diversity Collapse
Image-to-video (I2V) diffusion models can animate a single reference image into high-quality, subject-consistent videos, posing threats of misuse as anyone can turn personal photos into fabricated videos without consent. A promising solution is to proactively add an adversarial perturbation to the image before release, such that videos generated from it would be unfaithful or visually corrupted. However, due to an insufficient understanding of what governs how subsequent frames read the reference image in modern I2V models, existing protections struggle to offer strong protection under common perturbation budgets. In this paper, we approach the problem by exploring this underlying reading mechanism. Through a convex geometric analysis, we theoretically identify the spread of the reference value vectors as a key factor governing this reading process, and empirically find that even a mild contraction of this spread can severely degrade the generated video, making it an efficient lever for protection. Guided by this insight, we propose LETHE, a novel protection method that directly disrupts this reading process. LETHE pairs two complementary objectives, where a value collapse objective drives the reference value vectors toward uniformity, while an attention attraction objective draws subsequent frames toward these collapsed tokens to amplify corruption. We further construct two human-centric I2V protection benchmarks and conduct extensive experiments with multiple modern I2V models, which demonstrate the state-of-the-art protection performance and versatility of LETHE.
Updated: 2026-10-09 10:08:39
标题: LETHE：通过价值多样性崩溃防御未经授权的图像到视频生成
摘要: 图像到视频（I2V）扩散模型可以将单个参考图像动画化为高质量、主题一致的视频，可能会带来滥用的威胁，因为任何人都可以在未经同意的情况下将个人照片转化为虚构的视频。一个有前途的解决方案是在发布前主动向图像添加对抗性扰动，使从中生成的视频不忠实或视觉上受损。然而，由于对现代I2V模型中后续帧如何读取参考图像的控制机制了解不足，现有的保护措施很难在常见的扰动预算下提供强大的保护。在本文中，我们通过探索这种潜在的阅读机制来解决这个问题。通过凸几何分析，我们理论上确定了参考值向量的传播范围作为控制这一阅读过程的关键因素，并在经验上发现，即使是轻微的这种传播范围的收缩也会严重降低生成的视频质量，使其成为一种有效的保护杠杆。在这一见解的指导下，我们提出了LETHE，一种直接扰乱这一阅读过程的新颖保护方法。LETHE配对了两个互补的目标，其中一个值坍缩目标将参考值向量推向均匀性，而一个关注吸引目标将后续帧吸引到这些坍缩的令牌上以放大损坏。我们进一步构建了两个以人为中心的I2V保护基准，并与多个现代I2V模型进行了广泛实验，展示了LETHE的最新保护性能和多功能性。
更新时间: 2026-10-09 10:08:39
领域: cs.CV,cs.CR
The Express Lane to Spam and Centralization: An Empirical Analysis of Arbitrum's Timeboost
DeFi applications are vulnerable to MEV, where specialized actors profit by reordering or inserting transactions. To mitigate latency races and internalize MEV revenue, Arbitrum introduced Timeboost, an auction-based transaction sequencing mechanism that grants short-term priority access to an express lane. In this paper, we present the first large-scale empirical study of Timeboost, analyzing over 48.5 million express lane transactions and 494 thousand auctions between January 2025 and April 2026. Our results reveal five main findings. First, express lane control is highly centralized, with three entities winning 99.74% of auctions. Second, while express lane access provides earlier inclusion, it is currently mainly used for CEX-DEX arbitrage, limiting the value of priority access for other types of MEV extraction. Third, approximately 30% of timeboosted transactions revert, indicating that Timeboost does not effectively mitigate MEV inefficiencies such as spam. Fourth, secondary markets for reselling express lane rights struggle to sustain themselves due to poor execution reliability and unsustainable economics. Finally, auction competition declined as secondary markets introduced dedicated CEX-DEX execution lanes, enabling pay-per-transaction instead of pay-per-round billing and leading to steadily reduced revenue for the Arbitrum DAO. Taken together, these findings show that Timeboost fails to deliver on its stated goals of fairness, decentralization, and internalizing long term revenue for the DAO. Instead, it currently allows for collusion and narrows adoption, highlighting the limitations of auction-based ordering as a mechanism for fair transaction sequencing in rollups.
Updated: 2026-10-09 10:07:37
标题: 通往垃圾邮件和集中化的快速通道：对Arbitrum的Timeboost的实证分析
摘要: DeFi应用程序容易受到MEV的攻击，专门的参与者通过重新排序或插入交易来获利。为了减少延迟竞赛和内部化MEV收入，Arbitrum引入了Timeboost，这是一种基于拍卖的交易排序机制，为快速通道提供了短期优先访问权限。在本文中，我们展示了对Timeboost的第一次大规模实证研究，分析了从2025年1月到2026年4月之间的4850万笔快速通道交易和49.4万次拍卖。我们的结果揭示了五个主要发现。首先，快速通道控制高度集中，三家实体赢得了99.74%的拍卖。其次，虽然快速通道访问可以提供更早的包含，但目前主要用于CEX-DEX套利，限制了其他类型MEV提取的优先访问价值。第三，大约30%的Timeboost交易会回滚，表明Timeboost并未有效减少MEV的低效率，如垃圾邮件。第四，用于转售快速通道权利的二级市场难以维持，由于执行可靠性不佳和经济不可持续。最后，随着二级市场推出专门的CEX-DEX执行通道，拍卖竞争逐渐下降，这使得Arbitrum DAO的收入稳步减少，实现了按交易而不是按轮次计费。综上所述，这些发现表明Timeboost未能实现其公平、去中心化和内部化DAO长期收入的目标。相反，它目前容许勾结并限制了采用，突显了基于拍卖排序的限制作为一个公平交易排序机制在rollups中的机制。
更新时间: 2026-10-09 10:07:37
领域: cs.CR
Quantization-Tolerant Gradient Auditing for Active Inference-Attack Detection in Federated Learning
Inside a federated learning system a participant can quietly study the shared model and learn about others' private data. The usual defense, differential privacy, often adds so much noise that the model is no longer worth training. In cross-silo FL, where participants are a few accountable organizations and the server is trusted, we ask whether the server can instead interrogate clients. Each round it sends a secret challenge batch and asks every client to return a checksum of the gradient that batch induces, together with a timing. A client that mutated the model state, as an active generative-adversarial-network discriminator attacker does, returns a wrong checksum. We make this checksum robust to benign cross-platform floating-point non-determinism with a quantized fingerprint compared under a small tolerance. It flags two distinct active attacks at a true-positive rate of $1.0$ with no false positives on the checksum axis, stable across the fingerprint's parameters and up to $50$ clients at malicious ratios to $40\%$, for $32\,\mathrm{ms}$ of server work and $4.35\,\mathrm{kB}$ per response. Against a purely passive attacker the only signal is timing, and it is not robust: a cohort-relative test detects only in low-jitter conditions while a false-positive-free per-client baseline is blind, so no configuration delivers both. An attacker that answers the challenge on an untouched copy of the model while tampering with a replica evades the audit entirely. Therefore the checksum is a reliable in-band signal against attackers that do not decouple the audited computation from the attacked one, and it composes with differential privacy, whose accuracy cost we quantify on Texas100, rather than replacing it.
Updated: 2026-10-09 09:54:54
标题: 《基于梯度审计的量化容忍性方法用于联邦学习中主动推理攻击检测》
摘要: 在一个联邦学习系统中，参与者可以悄悄地研究共享模型并了解他人的私人数据。通常的防御方法，差分隐私，往往会添加太多的噪音，使得模型不再值得训练。在跨隔离的联邦学习中，参与者是一些负责任的组织，服务器是可信的，我们询问服务器是否可以代替客户端进行询问。每一轮，服务器发送一个秘密挑战批次，并要求每个客户端返回该批次引发的梯度的校验和，以及一个时间戳。一个操纵模型状态的客户端，就像主动生成对抗网络鉴别器攻击者一样，会返回一个错误的校验和。我们使这个校验和对良性跨平台浮点非确定性具有鲁棒性，通过在小容忍度下进行量化指纹比较。在校验和轴上，它以$1.0$的真正阳性率标记了两种不同的主动攻击，在指纹参数和高达$50$个客户端处于恶意比例$40\%$的稳定性下，对于$32\,\mathrm{ms}$的服务器工作和每个响应$4.35\,\mathrm{kB}$。对于纯被动攻击者，唯一的信号是时间，而且它不是鲁棒的：在低抖动条件下，队列相对测试只能检测到，而无误报的每个客户端基线是盲目的，因此没有配置可以同时交付两者。一个攻击者在未经观察的计算机模型副本上回答挑战，同时篡改一个副本，则完全规避了审计。因此，校验和是针对那些没有将审计计算与被攻击计算分离的攻击者的可靠内部信号，并且它与差分隐私结合，我们在Texas100上对其准确性成本进行量化，而不是取代它。
更新时间: 2026-10-09 09:54:54
领域: cs.CR
Beyond Edge Coverage: Per-Task Data-Flow Extraction at Kernel Function Boundaries via LLVM
Coverage-guided kernel fuzzers such as syzkaller rely on edge coverage trace-pc as their sole feedback signal. This context-blind approach cannot distinguish execution paths that differ only in argument values. For example, two invocations of copy_from_user() with different size parameters hit identical basic blocks yet have vastly different security implications. I present, an LLVM-based instrumentation framework that extends Linux KCOV with data-flow extraction of function arguments and return values. A compiler pass emits lightweight callbacks that report one source-level parameter per call, passing the value itself in a register together with its index and byte size. A composite is instead reported by the address of the object holding it, paired with a constant table of field (offset, size) pairs derived from DWARF DICompositeType metadata, with zero source annotation. The callbacks carry no program counter. The kernel derives it from the return address, which also attributes records the compiler emitted from inside an inlined body to the callee that was inlined. A lock-free per-task record buffer delivers variable-length records to user space with no interference to existing KCOV or syzkaller infrastructure, carrying entry, return, and comparison-operand records in one unified stream. I demonstrate dual utility: (1) fuzzers gain state-aware feedback for mutation guidance into value-dependent state transitions, and (2) security analysts obtain deterministic argument records for root-cause analysis without printk or kprobe overhead.
Updated: 2026-10-09 09:27:28
标题: 超越边缘覆盖范围：通过LLVM在内核函数边界处进行基于任务的数据流提取
摘要: 覆盖引导的内核模糊器（如syzkaller）依赖于边缘覆盖trace-pc作为其唯一的反馈信号。这种无上下文的方法无法区分仅在参数值上有所不同的执行路径。例如，两个使用不同大小参数调用copy_from_user()的实例会命中相同的基本块，但具有截然不同的安全影响。我提出了一个基于LLVM的仪器化框架，通过数据流提取函数参数和返回值来扩展Linux KCOV。编译器通过轻量级回调来报告每次调用的一个源级别参数，将值本身与其索引和字节大小一起传递到寄存器中。相反，由包含它的对象的地址报告一个复合体，与来自DWARF DICompositeType元数据的字段（偏移量，大小）对派生的常数表配对，不带有源注释。这些回调不携带程序计数器。内核从返回地址中推导出程序计数器，同时还会将编译器从内联体内部发出的记录归属于被内联的被调用者。一个无锁的每任务记录缓冲区将可变长度的记录传递到用户空间，不会对现有的KCOV或syzkaller基础设施造成干扰，其中包含入口、返回和比较操作数记录在一个统一的流中。 我展示了双重效用：（1）模糊器获得了状态感知反馈，以指导变异进入依赖于值的状态转换，以及（2）安全分析人员获得了确定性参数记录，用于根本原因分析，无需printk或kprobe开销。
更新时间: 2026-10-09 09:27:28
领域: cs.CR,cs.OS
Can escalation channels redirect reward hacking toward defect disclosure?
When coding agents encounter defective test infrastructure they may reward-hack: hardcoding outputs or editing test files to pass tests they cannot legitimately satisfy, a pattern that has now appeared outside benchmarks, in a coordinated multi-agent intrusion of a major AI platform's production infrastructure. The same capability that lets an agent detect and exploit a defect could let it report one, given the right decision environment. We evaluate escalation channels, structured reporting tools available to the agent at the point of conflict, as a decision-environment intervention that both reduces reward hacking and surfaces the infrastructure defects that trigger it. A $2 \times 2$ factorial separates the contributions of an escalation tool, a standalone anti-reward-hacking policy, and their combination. Across 8 frontier models spanning 5 families, the combined intervention reduces reward hacking from 23.6% to 5.3% (mixed-effects logistic OR = 9.2, 95% CI 5.0--16.8, $p < 10^{-12}$) with no detectable cost or performance overhead, eliminating it entirely for 6 of 8 models. Escalation and hacking are near-perfectly mutually exclusive, with 98.7% of escalations involving no hacking (100% under the combined intervention). Beyond reduction, escalation channels function as diagnostic infrastructure: on top of monitoring, escalation adds +10.1 percentage points of defect detection coverage and is more accurate once it fires (99.4% vs 85.8%). Unlike containment-based approaches that risk outpacing growing model capabilities, escalation channels redirect capability toward disclosure rather than exploitation.
Updated: 2026-10-09 08:57:37
标题: 升级通道是否可以将奖励攻击转向缺陷披露？
摘要: 当编码代理遇到有缺陷的测试基础设施时，它们可能会进行奖励黑客：硬编码输出或编辑测试文件以通过它们无法合理满足的测试，这种模式现在已经出现在基准之外，在一个主要AI平台生产基础设施的协调多代理入侵中。让代理检测和利用缺陷的相同能力也可以让它报告一个缺陷，只要提供正确的决策环境。我们评估升级通道，结构化报告工具在冲突点可用于代理作为决策环境干预，既减少奖励黑客，又暴露引发它的基础设施缺陷。一个$2 \times 2$的因子分离了升级工具、独立的反奖励黑客政策以及它们的组合对贡献的影响。在涵盖5个家族的8个前沿模型中，组合干预将奖励黑客从23.6%降至5.3%（混合效应logistic OR = 9.2，95% CI 5.0-16.8，$p < 10^{-12}$），没有可检测的成本或性能开销，对8个模型中的6个完全消除了奖励黑客。升级和黑客几乎是完全互斥的，98.7%的升级不涉及黑客（在组合干预下为100%）。除了减少外，升级通道还作为诊断基础设施运作：在监控之上，升级增加了+10.1个百分点的缺陷检测覆盖，并且一旦启动后更加准确（99.4% vs 85.8%）。与基于内容的方法不同，后者有可能超过不断增长的模型能力，升级通道将能力重新导向披露而不是利用。
更新时间: 2026-10-09 08:57:37
领域: cs.AI,cs.CR,cs.CY
AutoControl Arena: Synthesizing Executable Test Environments for Frontier AI Risk Evaluation
As Large Language Models (LLMs) evolve into autonomous agents, existing safety evaluations face a fundamental trade-off: manual benchmarks are costly, while LLM-based simulators are scalable but suffer from logic hallucination. We present AutoControl Arena, an automated framework for frontier AI risk evaluation built on the principle of logic-narrative decoupling. By grounding deterministic state in executable code while delegating generative dynamics to LLMs, we mitigate hallucination while maintaining flexibility. This principle, instantiated through a three-agent framework, achieves over 98% end-to-end success and 60% human preference over existing simulators. To elicit latent risks, we vary environmental Stress and Temptation across X-Bench (70 scenarios, 7 risk categories). Evaluating 9 frontier models reveals: (1) Alignment Illusion: risk rates surge from 21.7% to 54.5% under pressure, with capable models showing disproportionately larger increases; (2) Scenario-Specific Safety Scaling: advanced reasoning improves robustness for direct harms but worsens it for gaming scenarios; and (3) Divergent Misalignment Patterns: weaker models cause non-malicious harm while stronger models develop strategic concealment. Code and data are available at https://github.com/CosmosYi/AutoControl-Arena.
Updated: 2026-10-09 08:53:10
标题: AutoControl Arena：为前沿人工智能风险评估合成可执行测试环境
摘要: 随着大型语言模型（LLMs）演变为自主代理，现有的安全评估面临着一个根本性的折衷：手动基准成本高昂，而基于LLM的模拟器可以进行扩展，但存在逻辑幻觉。我们提出了AutoControl Arena，这是一个建立在逻辑-叙事解耦原则上的前沿AI风险评估的自动化框架。通过将确定性状态基于可执行代码，同时将生成动态委托给LLMs，我们可以减轻幻觉，同时保持灵活性。这一原则通过一个三代理框架实现，实现了超过98%的端到端成功率和60%的人类偏好，优于现有模拟器。为了引发潜在风险，我们在X-Bench上变化环境压力和诱惑（70个场景，7个风险类别）。评估9个前沿模型发现：（1）对齐错觉：在压力下，风险率从21.7%激增至54.5%，具有能力的模型显示出不成比例的增加；（2）特定场景安全缩放：高级推理改善了对直接危害的鲁棒性，但恶意游戏场景恶化；（3）分歧的不对准模式：较弱的模型造成非恶意危害，而较强的模型发展出战略性隐瞒。代码和数据可在https://github.com/CosmosYi/AutoControl-Arena找到。
更新时间: 2026-10-09 08:53:10
领域: cs.AI,cs.CR
STAIR: Effective Incident Response Using an End-to-End Agentic Planning Framework
Incident response planning is critical for restoring compromised software systems. Existing AIOps methods automate fault diagnosis and remediation and adapt response using feedback, but lack a unified decision basis across planning and execution. This limits their ability to maintain incident state and align actions with evolving recovery objectives during long-horizon, multi-stage security incident response. We present STAIR, an end-to-end agentic response framework that organizes planning and execution around a shared incident state. STAIR maintains incident evidence, response actions, and execution feedback in Graph-as-State (GAS), using the latest state to determine the response stage and recovery objectives for stage-specialized planning. The framework retrieves effect-validated historical experience based on the current state and stage to guide action selection. An Execution Harness executes actions, returns feedback to update GAS, and validates action effects for future experience reuse. We construct STAIR-Bench, an executable incident response benchmark comprising 44 Docker-based cyber-range cases. On its held-out test cases, STAIR achieves an overall defense score of 0.95, improving over the strongest baseline by approximately 10.47%. In the independent CAGE4 environment with ongoing attacks and benign activity, STAIR reduces cumulative penalties by approximately 28.8% relative to the strongest baseline.
Updated: 2026-10-09 08:48:42
标题: STAIR：使用端到端代理规划框架进行有效的事件响应
摘要: 事故响应规划对于恢复受损的软件系统至关重要。现有的AIOps方法自动化故障诊断和修复，并使用反馈调整响应，但缺乏跨规划和执行的统一决策基础。这限制了它们在长期、多阶段安全事件响应过程中维护事件状态并将行动与不断发展的恢复目标保持一致的能力。我们提出了STAIR，一个端到端的主动响应框架，围绕共享的事件状态组织规划和执行。STAIR在图形状态（GAS）中保留事件证据、响应行动和执行反馈，使用最新状态确定响应阶段和阶段专门规划的恢复目标。该框架根据当前状态和阶段检索经过验证效果的历史经验来引导行动选择。执行控制器执行行动，返回反馈以更新GAS，并验证行动效果以便将来重复使用经验。我们构建了STAIR-Bench，一个包含44个基于Docker的网络攻击案例的可执行事件响应基准。在其保留的测试案例中，STAIR实现了总体防御分数为0.95，比最强基线改善了约10.47%。在独立的CAGE4环境中，有持续攻击和良性活动，STAIR相对于最强基线将累积惩罚减少了约28.8%。
更新时间: 2026-10-09 08:48:42
领域: cs.CR,cs.AI
Understanding and Mitigating Cross-Session Attacks via Persistent Artifacts in Agent Skill Composition
Agent skills extend the capabilities of AI agents by bundling instructions, code, and resources into reusable packages. During skill execution, agents may produce or modify artifacts, such as files, configurations, and cached data, that are reused by other skills in later sessions. Since these artifacts can persist across sessions and later be accessed by other skills, they introduce a new attack surface that can be exploited to influence subsequent executions. However, the security risks arising from such cross-session skill interactions remain largely unexplored. In this work, we investigate cross-session composition attacks in skill-based AI agents and develop a runtime defense against them. Specifically, we characterize three attack mechanisms, namely Instruction Propagation (IP), Data Boundary Violation (DBV), and Evidence Poisoning (EP), through which malicious effects propagate across skill executions via persistent artifacts. To evaluate whether existing agents can resist these attacks while completing legitimate tasks, we construct SkillRelayBench, a benchmark of 270 adversarial cases drawn from 30 cross-session skill-composition scenarios across ten domains. Our evaluation of seven models across Codex CLI, Gemini CLI, and Claude Code reveals attack success rates (ASR) ranging from 46.7% to 77.4% and attack chain success rates (ACSR) ranging from 35.6% to 61.1%. Successful attack chains cover all five target security consequence categories, highlighting the broad impact of these attacks. To mitigate these risks, we propose SkillRelayGuard, which restores artifact provenance context and provides pre-execution advisory review while leaving the final execution decision to the agent. Evaluation across three agent settings shows that SkillRelayGuard reduces overall ASR from 69.5% to 10.9% and ACSR from 54.8% to 10.0%, while increasing legitimate task completion from 79.6% to 87.5%.
Updated: 2026-10-09 08:45:27
标题: 理解和减轻通过代理技能组合中的持续性产物进行的跨会话攻击
摘要: Agent skills扩展了AI代理的能力，通过将指令、代码和资源打包成可重复使用的包。在执行技能时，代理可能会生成或修改文件、配置和缓存数据等工件，这些工件在后续会话中被其他技能重复使用。由于这些工件可以跨会话持久存在，并且稍后可以被其他技能访问，它们引入了一个新的攻击面，可以被利用来影响后续的执行。然而，由于跨会话技能交互而产生的安全风险仍然大多未被探索。在这项工作中，我们研究基于技能的AI代理中的跨会话组合攻击，并针对这些攻击开发了一种运行时防御机制。具体来说，我们通过三种攻击机制，即指令传播（IP）、数据边界违规（DBV）和证据毒化（EP），表征了恶意效果如何通过持久工件在技能执行之间传播。为了评估现有代理在完成合法任务的同时是否能抵抗这些攻击，我们构建了SkillRelayBench，这是一个由30个跨会话技能组合场景中的270个对抗案例组成的基准。我们在Codex CLI、Gemini CLI和Claude Code上评估了七种模型，发现攻击成功率（ASR）在46.7%到77.4%之间，攻击链成功率（ACSR）在35.6%到61.1%之间。成功的攻击链涵盖了所有五个目标安全后果类别，突显了这些攻击的广泛影响。为了减轻这些风险，我们提出了SkillRelayGuard，它恢复了工件来源上下文，并在留给代理最终执行决策的同时提供了执行前的咨询审查。在三种代理设置上的评估显示，SkillRelayGuard将总体ASR从69.5%降至10.9%，ACSR从54.8%降至10.0%，同时将合法任务完成率从79.6%提高至87.5%。
更新时间: 2026-10-09 08:45:27
领域: cs.CR,cs.SE
Understanding Verification Vulnerabilities in Cross-chain Bridges: An Empirical Study
With the rapid development of blockchain technology, numerous heterogeneous blockchains have emerged, each featuring distinct architectures, consensus mechanisms, and application ecosystems. Cross-chain bridges have consequently become a critical infrastructure component, typically relying on smart contracts to facilitate message passing and asset transfers across heterogeneous blockchains. Unlike single-chain contracts, bridge contracts introduce additional verification logic to validate external-chain data, such as state proofs, message sender identities, and signatures. However, correctly implementing such logic is challenging. We refer to vulnerabilities that violate the security guarantees of cross-chain verification as Cross-chain Verification Vulnerabilities (CVVs). To characterize CVVs, we conduct the first systematic empirical study of this vulnerability class. By analyzing 588 security audit reports from 36 real-world cross-chain bridge projects, we derive 11 CVV patterns and organize them into 5 high-level categories according to the affected verification properties. Following this taxonomy, we construct a CVV dataset containing 830 vulnerability instances across 786 contract files. The dataset establishes mappings among audit reports, audited source-code versions, affected source files, and CVV-pattern labels. We then assess existing detection approaches. Current tools provide limited documented coverage of CVV patterns, while only two of four evaluated LLMs outperform an all-pattern constant baseline of 29.3% Macro-F1; the best reaches 51.2%. Together, the limited tool coverage and inconsistent LLM gains motivate detection methods tailored to cross-chain verification semantics.
Updated: 2026-10-09 07:58:02
标题: 理解跨链桥梁中的验证漏洞：一项实证研究
摘要: 随着区块链技术的快速发展，出现了众多异构区块链，每个区块链都具有独特的架构、共识机制和应用生态系统。跨链桥因此成为关键的基础设施组成部分，通常依赖智能合约来促进异构区块链之间的消息传递和资产转移。与单链合约不同，桥合约引入额外的验证逻辑来验证外部链数据，如状态证明、消息发送方身份和签名。然而，正确实现这样的逻辑是具有挑战性的。我们将违反跨链验证安全性保证的漏洞称为跨链验证漏洞（CVVs）。为了表征CVVs，我们进行了对这一漏洞类别的首次系统实证研究。通过分析36个实际跨链桥项目的588份安全审计报告，我们得出11种CVV模式，并根据受影响的验证属性将其组织成5个高级类别。根据这一分类法，我们构建了一个包含786个合约文件中830个漏洞实例的CVV数据集。该数据集建立了审计报告、审计源代码版本、受影响源文件和CVV模式标签之间的映射。然后我们评估了现有的检测方法。目前的工具对CVV模式的覆盖范围有限，而四种评估的LLM中只有两种超出了29.3% Macro-F1的所有模式常数基线；最好的达到了51.2%。综合考虑，有限的工具覆盖范围和不一致的LLM增益促使我们针对跨链验证语义定制检测方法。
更新时间: 2026-10-09 07:58:02
领域: cs.CR,cs.SE
PyCache Trap: The Inspection-Execution Gap in Agent Skill Scanners
Agent skills combine instructions with executable resources, giving third-party packages access to an agent's runtime. Existing skill scanners inspect documentation and visible source, but Python may execute a bundled bytecode cache with different behavior. We study this gap between inspection and execution through PyCache Trap, which pairs benign source with a substituted cache accepted by the loader and connects it to a task-relevant invocation. Scanner-guided rewriting changes the invocation wording while preserving the cache body, separating package admission from recognition of the concealed behavior. Across 100 skills and seven scanners, PyCache Trap achieves 94-100% attack success, with no semantic recognition of the cache-resident behavior. We propose execution-aware validation (EAV) to connect inspected instructions, scripts, imports, and runtime artifacts in a typed execution graph. EAV combines grounded behavioral analysis with trusted reproduction of compiled artifacts. It detects all 100 evaluated source-present cache substitutions and reaches 92.8% Recall at 10.0% FPR across five attack families and 200 benign skills. The results support checking the executable artifacts a runtime can select as part of skill admission, within the supported loaders and code-object normalization. The code is released at https://github.com/leo0481/PyCacheTrap.
Updated: 2026-10-09 07:56:20
标题: PyCache陷阱：代理技能扫描器中的检查执行差距
摘要: 代理技能将指令与可执行资源结合在一起，使第三方包能够访问代理的运行时。现有的技能扫描器检查文档和可见源代码，但是Python可能会执行一个捆绑的字节码缓存并具有不同的行为。我们通过PyCache Trap研究了检查和执行之间的差距，它将良性源代码与被加载器接受的替代缓存配对，并将其连接到一个与任务相关的调用。扫描器引导的重写改变了调用措辞，同时保留了缓存体，将包的准入与对隐藏行为的识别分开。在100个技能和七个扫描器中，PyCache Trap实现了94-100%的攻击成功率，而没有对缓存中的行为进行语义识别。我们提出了执行感知验证（EAV），将检查的指令、脚本、导入和运行时工件连接在一个类型化的执行图中。EAV将基于行为的分析与可信的编译工件复制相结合。它检测到了所有100个评估源代码的缓存替换，并在五个攻击系列和200个良性技能中达到了92.8%的召回率，10.0%的误报率。这些结果支持检查可执行工件，作为技能准入的一部分，支持的加载器和代码对象规范化。代码已发布在https://github.com/leo0481/PyCacheTrap。
更新时间: 2026-10-09 07:56:20
领域: cs.CR,cs.SE
Safety Graph: When Safety Responses Become Attack Surfaces in Cyber-Physical Systems
Modern industrial control systems (ICS) are susceptible to cyber attacks triggering safety failures in the operational technology (OT) layer. The traditional way to mitigate any failure in OT is to implement safety responses. However, we show in this work that these responses can open up new attack surfaces, thereby vastly expanding the scope of lateral movement. Existing frameworks model cyber-to-physical escalation but provide no mechanism to reason about how these safety responses to OT-failures can trigger new cascading cyber attacks in the Information Technology (IT) layer. To enable such reasoning, we introduce Safety Graph (SG), a formal framework unifying three paradigms - Attack Graphs (AG), Fault Trees (FT), and Event Trees (ET) - over a shared propositional state space, capturing the full bidirectional IT-OT cascade. We provide a sound and complete compilation of SG to a classical AI planning problem, performing plan search using A* to find cross-layer cascade paths. Through a Web3D modeling and simulation of the Tennessee Eastman chemical process plant, which is a widely used chemical engineering benchmark serving as a standard for testing control, monitoring and fault propagation, we demonstrate that the planning compilation of SG automatically discovers safety-response-induced lateral movements, and identifies the responsible barriers that are unidentifiable under existing models. A complexity and scalability analysis characterizes how the planning compilation of SG scales with model size and cascade depth. To the best of our knowledge, SG is the first framework to formally model IT-OT attacks and enable operators to explore the problem of cyber attacks triggering cascading failures in ICS.
Updated: 2026-10-09 06:40:17
标题: 安全图：当安全响应成为网络物理系统中的攻击面
摘要: 现代工业控制系统（ICS）易受网络攻击的影响，可能导致操作技术（OT）层的安全故障。传统的减轻OT故障的方法是实施安全响应。然而，我们在这项工作中展示，这些响应可能会打开新的攻击面，从而大大扩展了横向移动的范围。现有的框架模拟了从网络到物理的升级，但没有提供关于这些安全响应如何触发信息技术（IT）层中新的级联网络攻击的机制。为了实现这种推理，我们引入了Safety Graph（SG），这是一个形式框架，统一了三种范例 - 攻击图（AG），故障树（FT）和事件树（ET）- 在一个共享的命题状态空间上，捕获全双向的IT-OT级联。我们提供了一个对SG的声音和完整的编译，解决了古典人工智能规划问题，使用A*进行计划搜索，以找到跨层级级联路径。通过对田纳西伊斯曼化工厂的Web3D建模和仿真，该工厂是一个广泛使用的化工工程基准，用于测试控制、监测和故障传播，我们展示了SG的规划编译能自动发现安全响应引发的横向运动，并识别出在现有模型下无法识别的负责障碍。复杂性和可扩展性分析表征了SG的规划编译如何随着模型大小和级联深度的增加而扩展。据我们所知，SG是第一个正式模拟IT-OT攻击并使操作员能够探讨ICS中网络攻击触发级联故障问题的框架。
更新时间: 2026-10-09 06:40:17
领域: cs.CR
The Limits of Query Monitoring against Distributed Model Extraction
Model providers now face distillation campaigns run through thousands of accounts, each of which looks like an ordinary user. When network and payment metadata cannot link those accounts, the only evidence left is what they ask. We study what a provider can and cannot learn from those queries alone, using one million real chat turns for development and 17 later weeks that no design choice touched for a single frozen evaluation. The answer is set less by anything specific to extraction than by benign coordination: a campaign is visible only if it is more distinctive than the most distinctive group of ordinary users in the same week. We make this precise, show that a small but positive difference between campaign and users buys a detection time rather than detectability, and identify the campaign that is hardest to detect. Empirically, a calibrated monitor catches naive distributed extraction that per-account monitors miss (0.86-1.00 of weeks), and one variant - selecting accounts that move together after discounting how topical each one is - still catches campaigns padded with natural text (0.50-0.96 of weeks) where the other two spectral detectors catch none. Distilling each campaign into a small student shows that, in our testbed, campaigns that evaded every detector kept at most 42% of the capability of a detectable one. The limits are equally concrete: a campaign that ramps up slowly poisons every detector's calibration, benign evaluation pipelines look exactly like distillation, we found no deployable detector at colluder shares of 2.5% or below, and campaigns that relay real users' requests may be invisible to any query-only monitor.
Updated: 2026-10-09 06:37:01
标题: 查询监控在分布式模型提取中的限制
摘要: 模型提供商现在面临的挑战是，通过成千上万个账户进行提炼活动，每个账户看起来都像一个普通用户。当网络和支付元数据无法将这些账户联系在一起时，唯一留下的证据就是他们的查询。我们研究了一个提供商仅从这些查询中可以学到和不能学到的内容，使用了一百万真实的聊天对话用于开发，并且在之后的17周中没有任何设计选择干扰进行了单一冻结评估。答案不是由任何特定的提取因素决定，而是由良性协调决定：一个活动只有在比同一周中最突出的普通用户群体更具特色时才能被看到。我们将这一点具体化，表明活动和用户之间存在微小但积极的差异可以购买检测时间，而不是可检测性，并且确定了最难检测到的活动。校准监视器在每个账户监视器所错过的天数中捕捉到天真的分布式提取（0.86-1.00周），而一种变体 - 选择在折扣后一起移动的账户 - 仍然捕捉到填充了自然文本的活动（0.50-0.96周），而其他两种光谱检测器则没有捕捉到任何活动。将每个活动提炼为一个小型学生表明，在我们的测试中，每个逃避所有检测器的活动最多只保留了可检测活动的42％的能力。限制同样具体：缓慢升级的活动会毒害每个检测器的校准，良性评估管道看起来与提炼完全相同，我们没有找到适用于2.5％或以下的共谋者份额的可部署检测器，而将转发真实用户请求的活动可能对任何仅查询监视器都是不可见的。
更新时间: 2026-10-09 06:37:01
领域: cs.CR,cs.LG
CAPE - Context-Aware Prioritization Engine for Vulnerability Scoring and Management
Vulnerability patching is critical for safeguarding an enterprise's software ecosystem from cyber attacks. However, patching requires adequate preparation for potential system downtime and for unforeseen effects of the patched software on a live system. Thus, effective risk mitigation requires prioritization strategies that explicitly account for the operational context of the affected software. While vulnerability prioritization based on Common Vulnerability Scoring System (CVSS), Exploit Prediction Scoring System (EPSS) and CISA Known Exploited Vulnerabilities (KEV) catalog, can help, their differing goals and methodologies often lead to inconsistent, context-unaware decisions resulting in mis-allocated remediation efforts. In this work, we present the Context-Aware Prioritization Engine (CAPE), a vulnerability analysis framework that enriches each vulnerability scanner report of vulnerabilities (CVEs) with deployment-specific evidence. This evidence is used to perform vulnerability reachability, centrality, and exploitability analysis, producing a ranked priority score for the effective distribution of remediation effort. We evaluate CAPE on 30 open-source agentic applications comprising 6,138 scanner-reported CVEs. Our evaluation shows that a significant majority of CVEs are not reachable and exploitable which results in significant disparities between CAPE and context-free baselines - the CVSS, EPSS, and KEV. It also shows that CAPE consistently surfaces confirmed-exploitable CVEs at higher priority than all baselines, demonstrating that context-aware analysis provides substantially more accurate remediation guidance than context-free severity scoring alone.
Updated: 2026-10-09 06:30:34
标题: CAPE - 基于上下文的漏洞评分和管理优先级引擎
摘要: 漏洞修补对于保护企业软件生态系统免受网络攻击是至关重要的。然而，修补需要充分准备潜在的系统停机和补丁软件对实际系统的未预期影响。因此，有效的风险缓解需要优先考虑受影响软件的运营背景的策略。虽然基于通用漏洞评分系统（CVSS）、利用预测评分系统（EPSS）和CISA已知利用漏洞（KEV）目录的漏洞优先级可能有所帮助，但它们的不同目标和方法常常导致不一致的、不考虑上下文的决策，从而导致修复工作的错误分配。在本研究中，我们提出了上下文感知优先级引擎（CAPE），这是一个漏洞分析框架，它通过部署特定证据丰富每个漏洞扫描器报告的漏洞（CVE）。这些证据用于执行漏洞可达性、中心性和利用性分析，产生一个排名优先级分数，以有效分配修复工作。我们在30个开源代理应用程序上评估了CAPE，这些应用程序包括6,138个扫描器报告的CVE。我们的评估结果显示，绝大多数CVE是不可达和不可利用的，这导致CAPE与不考虑上下文的基线——CVSS、EPSS和KEV之间存在显著差异。它还表明，CAPE始终将经确认可利用的CVE放在比所有基线更高的优先级，证明上下文感知分析比单纯基于严重性评分提供更准确的修复指导。
更新时间: 2026-10-09 06:30:34
领域: cs.CR,cs.SE
Space Over Statistic, Retraining Over Ranking: Auditing Training Labels Under Label-Only Backdoors
Some backdoor attacks never touch an image. FLIP only changes labels: on CIFAR-10, flipping 3% of them makes a ResNet-32 send 99.7% of triggered source-class images to the attacker's class. The usual defenses look for the poison inside a model trained on those same labels. So when they struggle, is the problem the test they run, or the space they run it in? We answer with a controlled swap: we keep the test, the labels and the metric fixed, change only the feature space, and judge each defense by retraining the victim on what it keeps. The space decides. On dog$\to$cat the same nearest-neighbor check scores AUROC 0.546 on the poisoned model's features, 0.967 on frozen DINOv2 features that never saw a label, and 0.995 on the same network trained on clean labels, an oracle control rather than a defense. We reproduce SPECTRE, the defense FLIP's authors recommend, to 90.94 / 0.97 against its published 91.47 / 1.7, then price it: on truck$\to$deer it deletes 20% of every class, while a prototype audit on the frozen features deletes 6% for higher clean accuracy and lower attack success. On three harder pairs that audit leaves about a quarter of the attack standing. A count read off encoder geometry and clean labels, before any attack is run, separates the pair where the audit removes the attack from those where it only partly does, though on a single zero-count pair. Three results warn against the easy story. A better detection score does not mean a safer model: on two of four pairs Confident Learning scores the higher AUROC, and on dog$\to$cat leaves 2.6 times the attack success. The prototype family, our best static performer, is the first an adaptive attacker slips past. And past a point, catching more poisons stops reducing attack success. Check labels in a space that never learned them, and judge a defense by the retrained model, not by its detection score.
Updated: 2026-10-09 06:05:01
标题: 空间胜于统计，再培训胜于排名：在仅有标签后门下审计训练标签
摘要: 一些后门攻击从不触及图像。FLIP只改变标签：在CIFAR-10上，翻转其中3%的标签使得一个ResNet-32将99.7%的触发源类图像发送到攻击者的类别。通常的防御方法是在训练在相同标签上的模型中寻找毒药。因此，当它们挣扎时，问题出在它们进行的测试，还是测试的空间？ 我们通过一种受控的交换进行回答：我们保持测试、标签和度量不变，仅改变特征空间，并通过在其保留的内容上对受害者进行重新训练来评估每种防御措施。空间起决定作用。在狗$\to$猫的情况下，相同的最近邻检查在受污染模型的特征上得分AUROC 0.546，在从未看到标签的冻结DINOv2特征上得分0.967，在干净标签上训练的相同网络上得分0.995，这是一个神谕控制而不是一种防御措施。我们复现了SPECTRE，FLIP的作者推荐的防御方法，对其发布的91.47 / 1.7进行了90.94 / 0.97的价格，然后对其进行评估：在卡车$\to$鹿的情况下，它删除了每个类别的20%，而在冻结特征上的原型审计删除了6%，以获得更高的干净准确性和更低的攻击成功率。在另外三对较难的情况下，该审计留下了攻击的大约四分之一。在任何攻击运行之前，通过编码器几何和干净标签读取计数，将该审计从完全移除攻击的情况与部分移除攻击的情况分开，尽管在一个计数为零的对中。 三个结果警告我们不要轻易下结论。更好的检测得分并不意味着模型更安全：在四对中的两对中，自信学习得分更高的AUROC，并且在狗$\to$猫的情况下留下了攻击成功的2.6倍。原型家族，我们最好的静态表现者，是第一个被自适应攻击者绕过的。并且一旦超过一个点，捕捉更多的毒药并不能减少攻击成功率。在一个从未学习过它们的空间中检查标签，并通过重新训练的模型来评估一种防御措施，而不是通过其检测得分。
更新时间: 2026-10-09 06:05:01
领域: cs.CR,cs.CV,cs.LG
Provably Subexponential Algorithms for Bounded-Error LWE with Few Samples
We show that a variant of the Wagner/BKW-style algorithm for LWE of Kirchner and Fouque [KF15] solves decision and search LWE with secrets of size $n$ over prime moduli $q=n^{κ+o(1)}$, for any fixed $κ>0$, with errors bounded by $\exp((\ln n)^{o(1)})$, in time $2^{O(n/\ln\ln n)}$ given $n+O(n/\ln\ln n)$ original samples. Further, if each coordinate of the secret is also bounded by $\exp((\ln n)^{o(1)})$, the same time bound holds with errors bounded by $n^{o(1)}$ and only $O(n/\ln\ln q)$ samples. Earlier, it was known that Kirchner and Fouque's algorithm could solve LWE with these parameters in subexponential time with a much larger number of samples $m = Ω(n\log{n})$ [HKM18]. Ducas et al. [DEL25] showed that an adaptation of this algorithm could solve the corresponding dual, the $\mathrm{SIS}^{\infty}$ problem, with a suitable norm bound $q/\operatorname{polylog}(n)$ and $m=n+ω(n/\ln\ln n)$ variables in subexponential time. Whereas their analysis involves interpreting the algorithm in terms of chains of lattices, our approach is quite elementary and is based on bounding the lower moments of a Fourier statistic.
Updated: 2026-10-09 05:58:55
标题: 证明有界误差LWE问题的少样本子指数算法
摘要: 我们展示了Kirchner和Fouque提出的LWE变种的Wagner/BKW风格算法[KF15]可以解决具有大小为$n$的秘密信息的决策和搜索LWE问题，其中素数模$q=n^{κ+o(1)}$，对于任意固定的$κ>0$，并且错误受到$\exp((\ln n)^{o(1)})$的限制，时间复杂度为$2^{O(n/\ln\ln n)}$，给定$n+O(n/\ln\ln n)$个原始样本。此外，如果秘密信息的每个坐标也受到$\exp((\ln n)^{o(1)})$的限制，那么相同的时间复杂度适用于错误受到$n^{o(1)}$的限制，且仅需要$O(n/\ln\ln q)$个样本。 早期已知Kirchner和Fouque的算法可以在子指数时间内使用更多的样本数$m = Ω(n\log{n})$ [HKM18]解决具有这些参数的LWE问题。Ducas等人[DEL25]展示了该算法的适应版本可以在子指数时间内解决相应的对偶问题，即$\mathrm{SIS}^{\infty}$问题，其中适当的范数限制为$q/\operatorname{polylog}(n)$，变量数为$m=n+ω(n/\ln\ln n)$。而他们的分析涉及将算法解释为格链，我们的方法非常基础，基于对傅里叶统计的低阶矩的限制。
更新时间: 2026-10-09 05:58:55
领域: cs.CR
Post-Quantum Session Establishment for Matter over Thread: Message Size, Transport, and Protocol Options
Matter's first post-quantum change protects device attestation with ML-DSA, but its operational handshake, CASE, still derives session keys from P-256 ECDH, so recorded traffic can be decrypted once a quantum computer exists. We quantify the cost of moving CASE to ML-KEM and post-quantum signatures over Thread, where frames carry at most 127 bytes and a Matter message at most 1178 bytes. A byte-exact model of the CASE messages shows that a drop-in ML-KEM-768/ML-DSA-44 handshake grows from 1.6 kB to 22.8 kB and that no fully post-quantum configuration built from FIPS 203/204/205 algorithms fits the message limit. A Monte Carlo model of multi-hop Thread, calibrated to OpenThread and Matter source constants, shows that sending these messages as fragmented IPv6 datagrams causes spurious retransmissions whenever a message and its reply outlast Matter's retransmission timer. At three to four hops the drop-in handshake then takes 1.8 to 2.6 times longer than with a selective-repeat block transfer, with or without interference, and up to 33% of handshakes with sleepy devices fail; raising the timers does not remove the effect. Caching certificate chains and authenticating the device with a pre-distributed ML-KEM key reduce the handshake to 6.1 kB and 44 ms of device computation. As a first step, hybrid P-256+ML-KEM-512 key exchange with a cached device chain protects session keys against harvest-now, decrypt-later attacks while every message still fits the current limit.
Updated: 2026-10-09 05:32:12
标题: 量子后会话建立对于物质超线程的影响：消息大小、传输和协议选项
摘要: 物联网的第一个量子后变化通过ML-DSA保护设备认证，但其操作握手CASE仍然从P-256 ECDH派生会话密钥，因此一旦量子计算机存在，记录的流量可以被解密。我们对将CASE移至ML-KEM和后量子签名在Thread上的成本进行了量化，其中帧最多携带127字节，Matter消息最多携带1178字节。CASE消息的字节精确模型显示，一个插入ML-KEM-768/ML-DSA-44握手从1.6 kB增长到22.8 kB，并且没有完全建立在FIPS 203/204/205算法上的后量子配置符合消息限制。校准为OpenThread和Matter源常数的多跳Thread的蒙特卡罗模型显示，将这些消息作为分段的IPv6数据报发送会导致虚假重传，每当消息及其回复超过Matter的重传计时器时。在三到四跳时，插入握手的时间比选择性重复块传输长1.8到2.6倍，无论是否干扰，并且高达33%的带有休眠设备的握手失败；增加定时器并不能消除这种效果。缓存证书链并使用预分发的ML-KEM密钥对设备进行认证可将握手减少到6.1 kB，并且设备计算时间为44毫秒。作为第一步，混合P-256+ML-KEM-512密钥交换与缓存设备链可保护会话密钥免受立即收集、稍后解密攻击，同时每条消息仍然符合当前限制。
更新时间: 2026-10-09 05:32:12
领域: cs.CR,cs.NI
What Makes Adversarial Examples Transfer Across Deepfake Detectors?
Deepfake detectors remain vulnerable to transfer-based black-box attacks, in which adversarial examples are generated on a source surrogate model and transferred to a target model, unknown to the attacker. Yet how source--target compatibility shapes attack success remains poorly understood. Prior studies evaluate limited detector pools and rarely disentangle architectural from training factors. We conduct a controlled evaluation of adversarial transferability across 60 detectors spanning six backbones, two pretraining regimes, and five training-data configurations, using two attack procedures: AutoAttack (AA) and the Carlini--Wagner attack with Expectation over Transformation (CW--EOT). Matched comparisons reveal significantly higher transfer when source and target share an exact backbone, architecture family, pretraining regime, or training data. This compatibility structure is attack-dependent: exact backbone compatibility has the largest effect under AA, whereas shared pretraining and training data have the largest effects under CW--EOT. When transfer is averaged across non-target sources, mean attack success rate (ASR) is $7.21\%$ under AA and $19.52\%$ under CW--EOT. By contrast, a multi-source oracle combining both attacks attains a \(64.48\%\) mean ASR after excluding exact backbone and training-data matches, showing that source averaging can substantially understate target vulnerability. We release 240,000 adversarially perturbed images, complete pairwise transfer results, detector configurations, and evaluation code. These findings establish source--target compatibility and source-model selection as central dimensions of credible transfer-based black-box robustness evaluation.
Updated: 2026-10-09 05:06:59
标题: 深度伪造检测器之间为何会传递对抗性样本？
摘要: 深度伪造检测器仍然容易受到基于转移的黑盒攻击的影响，其中对抗性示例在源替代模型上生成，并转移到目标模型，攻击者不知情。然而，源-目标兼容性如何塑造攻击成功仍然知之甚少。先前的研究评估了有限的检测器池，并很少将架构因素与训练因素分开。我们对60个检测器进行了对抗性可转移性的受控评估，涵盖了六种骨干、两种预训练方案和五种训练数据配置，使用了两种攻击程序：AutoAttack（AA）和带有期望转换的Carlini-Wagner攻击（CW-EOT）。匹配比较显示，当源和目标共享完全相同的骨干、架构系列、预训练方案或训练数据时，转移性显著增加。这种兼容性结构取决于攻击方式：在AA下，完全相同的骨干兼容性效果最大，而在CW-EOT下，共享的预训练和训练数据效果最大。当将转移平均应用于非目标源时，平均攻击成功率（ASR）在AA下为7.21％，在CW-EOT下为19.52％。相比之下，一个结合了两种攻击的多源神谕，在排除了完全相同的骨干和训练数据匹配后，实现了64.48％的平均ASR，表明源平均可能大大低估了目标的脆弱性。我们发布了24万个对抗性扰动图像、完整的成对转移结果、检测器配置和评估代码。这些发现将源-目标兼容性和源模型选择确立为可靠的基于转移的黑盒鲁棒性评估的核心维度。
更新时间: 2026-10-09 05:06:59
领域: cs.CV,cs.AI,cs.CR,cs.LG
Trust, but Validate the Instrument: Auditing AI-Generated RTL Verification Plans on Authored Security-Regression Proxies
AI-generated RTL verification plans can satisfy a provider schema yet fail at the boundary to trusted execution. We present SecTB-RTL, an auditable framework covering 31 tasks and 124 authored hardware-security regressions. A deterministic non-AI baseline killed 36, 75, and 78 mutants at increasing resource limits. The first confirmatory run (C1-R2) failed before model execution because the provider rejected its response schema. After a schema-only repair made without viewing outcomes, a separately frozen follow-up run (C1-R3) completed 1,860 calls. The provider accepted 1,857 responses, but only nine passed the production semantic validator. The generation and execution rules did not match. We therefore preserve the run as an instrument-validation incident and report no prompt-effect estimate. This incident shows that provider or schema acceptance does not establish execution validity. Compilation and coverage are only diagnostics; the exact saved artifact must pass the full production path. A subsequent follow-up is excluded because it did not satisfy the preregistered evidence-completeness gate and is treated only as future work. We release the benchmark, failure-preserving contract, incident provenance, and governance controls needed to prevent infrastructure behavior from being misreported as model behavior.
Updated: 2026-10-09 04:59:05
标题: 信任，但验证工具：对人工智能生成的RTL验证计划在编写的安全回归代理上进行审计
摘要: 人工智能生成的RTL验证计划可以满足提供者架构，但在到可信执行的边界处失败。我们提出了SecTB-RTL，一个可审计的框架，涵盖了31个任务和124个硬件安全回归。一个确定性的非人工智能基线在不断增加的资源限制下消灭了36、75和78个突变体。第一次确认性运行（C1-R2）在模型执行之前失败，因为提供者拒绝了其响应架构。在未查看结果的情况下进行了仅架构修复后，一个单独冻结的后续运行（C1-R3）完成了1,860次调用。提供者接受了1,857个响应，但只有九个通过了生产语义验证器。生成和执行规则不匹配。因此，我们将运行保留为一个工具验证事件，并报告没有及时效果估计。这一事件表明，提供者或架构的接受并不能确立执行的有效性。编译和覆盖仅仅是诊断；确切的保存物必须通过完整的生产路径。随后的跟进被排除在外，因为它没有满足预注册的证据完整性门槛，只被视为未来工作。我们发布了基准、保留故障合同、事件来源和治理控制，以防止基础设施行为被误报为模型行为。
更新时间: 2026-10-09 04:59:05
领域: cs.CR,cs.AI,cs.CE,cs.IR,cs.LG
A Comprehensive Survey of Website Fingerprinting Attacks and Defenses in Tor: Advances and Open Challenges
The Tor network provides users with strong anonymity by routing their internet traffic through multiple relays. While Tor encrypts traffic and hides IP addresses, it remains vulnerable to traffic analysis attacks such as the website fingerprinting (WF) attack, achieving increasingly high fingerprinting accuracy even under open-world conditions. In response, researchers have proposed a variety of defenses, ranging from adaptive padding, traffic regularization, and traffic morphing to adversarial perturbation, that seek to obfuscate or reshape traffic traces. However, these defenses often entail trade-offs between privacy, usability, and system performance. Despite extensive research, a comprehensive survey unifying WF datasets, attack methodologies, and defense strategies remains absent. This paper fills that gap by systematically categorizing existing WF research into three key domains: datasets, attack models, and defense mechanisms. We provide an in-depth comparative analysis of techniques, highlight their strengths and limitations under diverse threat models, and discuss emerging challenges such as multi-tab browsing and coarse-grained traffic features. By consolidating prior work and identifying open research directions, this survey serves as a foundation for advancing stronger privacy protection in Tor.
Updated: 2026-10-09 04:05:10
标题: 《Tor中网站指纹识别攻击和防御的全面调查：进展与挑战》
摘要: Tor网络通过将用户的Internet流量通过多个中继路由来为用户提供强大的匿名性。虽然Tor加密流量并隐藏IP地址，但仍然容易受到流量分析攻击，如网站指纹识别（WF）攻击，在开放世界条件下实现越来越高的指纹识别准确性。研究人员提出了各种防御措施，包括自适应填充、流量正则化、流量变形和对抗扰动，旨在混淆或重塑流量迹象。然而，这些防御措施往往需要在隐私性、可用性和系统性能之间进行权衡。尽管进行了大量研究，但尚缺乏一个统一WF数据集、攻击方法和防御策略的综合调查。本文通过系统地将现有WF研究分类为三个关键领域：数据集、攻击模型和防御机制。我们对技术进行了深入的比较分析，突出了它们在不同威胁模型下的优势和局限性，并讨论了诸如多标签浏览和粗粒度流量特征等新兴挑战。通过整合先前的工作并确定开放的研究方向，这项调查为在Tor中加强隐私保护奠定了基础。
更新时间: 2026-10-09 04:05:10
领域: cs.CR
WaveMark: Robust Wavelet-Domain Image Watermarking for Diffusion Models
Image watermarking enables provenance tracing for photorealistic images generated by diffusion models. However, state-of-the-art (SOTA) removal attacks can suppress embedded signals with little perceptual degradation. This vulnerability limits the effectiveness of watermarks for intellectual property protection and misuse attribution. We propose WaveMark, a novel wavelet-domain watermarking method that embeds messages in a diffusion model's initial noise. To support robust recovery from perturbed carriers, WaveMark distributes keyed replicas of message bits across spatial positions, wavelet orientations, and scales. Each carrier groups coefficients from four latent channels and encodes a randomized bit through magnitude projection, providing an explicit decoding margin. We evaluate WaveMark with SD2.1 on SDP and MS-COCO Captions under 14 attacks. WaveMark achieves 100.0% bit accuracy (BA) without attacks, while maintaining image quality comparable to the evaluated baselines. Under SOTA MarkNull and MarkNull-A, it retains over 88% BA, outperforming all baselines.
Updated: 2026-10-09 03:46:45
标题: WaveMark：用于扩散模型的稳健小波域图像水印技术
摘要: 图像水印技术使扩散模型生成的逼真图像的来源可以被追踪。然而，当前最先进的去水印攻击可以在几乎没有感知损坏的情况下抑制嵌入的信号。这种脆弱性限制了水印在知识产权保护和滥用归因方面的有效性。我们提出了WaveMark，一种新颖的小波域水印方法，它将消息嵌入到扩散模型的初始噪声中。为了支持对受扰传输载体的强大恢复，WaveMark在空间位置、小波方向和比例上分布了密钥复制的消息位。每个载体从四个潜在通道中组合系数，并通过幅度投影编码一个随机位，提供了明确的解码边界。我们在SDP和MS-COCO Captions上使用SD2.1评估了WaveMark在14种攻击下的性能。WaveMark在没有攻击的情况下实现了100.0%的位准确性（BA），同时保持了与评估基线相当的图像质量。在SOTA MarkNull和MarkNull-A下，它保留了超过88%的BA，优于所有基线。
更新时间: 2026-10-09 03:46:45
领域: cs.CR,cs.CV
AgentValve: Sybil-Resilient Payment Authorization for Permissionless AI Markets
AI agents increasingly select and pay service providers without human review. In emerging permissionless markets, however, adversaries can create inexpensive identities and fabricate feedback to inflate provider reputation, making reputation-based payment authorization vulnerable to Sybil attacks and exposing clients to unbounded financial loss. This paper presents AgentValve, a protocol for Sybil-resilient payment authorization in permissionless AI markets. AgentValve builds on a trust network that represents reputation as finite capacity in a directed graph and authorizes payments through trust paths with sufficient available capacity. A smart contract atomically reserves the capacity supporting each admitted payment, bounding honest agents' payment exposure by the capacity they extend into the adversarial region, and thus creating identities or interactions entirely within that region cannot increase this bound. To limit capacity depletion by malicious senders, AgentValve combines payment-backed capacity creation with bounded route length, preventing attackers from depleting honest capacity without incurring corresponding payment costs. AgentValve further decouples off-chain route discovery from on-chain verification and introduces a witness-commit mechanism that reduces route verification to linear work while requiring only constant persistent state per admitted payment. We implement a AgentValve prototype and evaluate it using six months of Ethereum data and synthetic workloads. Experimental results show that under attacks with up to 10,000 synthetic Sybil identities and with chosen attackers, the payment exposure and capacity depletion remain bounded. Regarding practicality, the witness-commit mechanism lowers admission cost by up to 40.8% compared with baselines, and admitting common payments costs approximately \$0.04-\$0.21.
Updated: 2026-10-09 03:45:31
标题: AgentValve：面向无许可AI市场的Sybil-鲁棒支付授权
摘要: 人工智能代理越来越频繁地在没有人类审查的情况下选择并支付服务提供商。然而，在新兴的无许可市场中，对手可以创建廉价的身份并伪造反馈，以提高提供商的声誉，使基于声誉的付款授权易受Sybil攻击影响，并使客户面临无限的财务损失。 本文提出了AgentValve，这是一个用于无许可AI市场中的Sybil抵抗付款授权的协议。AgentValve建立在一个信任网络之上，将声誉表示为有限容量的有向图，并通过具有足够可用容量的信任路径授权支付。一个智能合约会原子地保留支持每笔承认的支付的容量，通过延伸到对抗区域的容量来限制诚实代理的支付风险，因此在该区域内创建身份或交互不会增加此限制。为了限制恶意发送者对容量的耗尽，AgentValve将支持支付的容量创建与有界路线长度相结合，防止攻击者耗尽诚实容量而不产生相应的支付成本。AgentValve进一步将链下路由发现与链上验证分离，并引入了一个见证-提交机制，将路由验证降低为线性工作，同时每笔承认的支付只需要恒定的持久状态。 我们实现了一个AgentValve原型，并使用六个月的以太坊数据和合成工作负载进行评估。实验结果表明，在具有高达10,000个合成Sybil身份和选择的攻击者的攻击下，支付风险和容量耗尽保持有限。就实用性而言，见证-提交机制将承认成本降低了最多40.8%，并且承认常见支付大约花费0.04美元至0.21美元。
更新时间: 2026-10-09 03:45:31
领域: cs.CR
Sample-Preserving Search-to-Decision Reduction for Noisy Linear Equations over Large Moduli
The Noisy Linear Equations problem involves finding the solution to a random linear system over a finite field given noisy evaluations. This is a generalisation of various learning problems widely used in cryptography, including Learning with Errors (LWE), Learning with Rounding (LWR), and Learning Parity with Noise (LPN). We present a new sample-preserving search-to-decision reduction for Noisy Linear Equations that has complexity $\mathrm{poly}(n,m,\log q, q/σ)$, where $n$ is the dimension of the secret, $m$ is the number of samples, $q$ is the modulus, and $σ$ is the magnitude of the noise (defined suitably for the kind of noise involved). In particular, this implies search-to-decision reductions for LWE and LWR that run in time $\mathrm{poly}(n)$ even if the modulus $q$ is exponentially large, as long as the noise-to-modulus ratio $σ/q$ is non-negligible. Our reduction works for prime moduli $q$. If $q$ is composite, it instead produces a list that contains a small multiple of the solution. Prior to this work, known search-to-decision reductions for LWE and LWR either did not preserve the parameters of the problem (i.e., $m$, $n$, $σ$, and $q$), or had complexity that scaled polynomially with either $σ$ or the largest prime factor of $q$.
Updated: 2026-10-09 03:04:07
标题: 大模数上嘈杂线性方程的保留样本搜索到决策缩减
摘要: Noisy Linear Equations问题涉及在有噪声的情况下找到有限域上随机线性系统的解。这是密码学中广泛使用的各种学习问题的泛化，包括带错误的学习（LWE）、四舍五入学习（LWR）和带噪声的奇偶校验学习（LPN）。 我们提出了一个新的保留样本的搜索到决策约简，用于Noisy Linear Equations，其复杂度为$poly(n,m,\log q, q/σ)$，其中$n$是秘密维度，$m$是样本数量，$q$是模数，$σ$是噪声的幅度（适当定义了所涉噪声的类型）。特别是，这意味着对于LWE和LWR的搜索到决策约简可以在时间$poly(n)$内运行，即使模数$q$指数级大，只要噪声与模数比$σ/q$是非可忽略的。我们的约简适用于素数模$q$。如果$q$是复合数，则会产生包含解的小倍数的列表。 在这项工作之前，对于LWE和LWR的已知搜索到决策约简要么不能保留问题的参数（即$m$，$n$，$σ$和$q$），要么其复杂度随着$σ$或$q$的最大质因数多项式地增长。
更新时间: 2026-10-09 03:04:07
领域: cs.CR
Vulnerable Models in the AI Supply Chain: An Empirical Study of Adoption, Propagation, and Usage Patterns
The rapid integration of Artificial Intelligence (AI) models has established a complex supply chain, but many of these shared artifacts harbor severe vulnerabilities such as insecure serialization and architectural backdoors. While prior efforts have extensively scanned model registries to flag malicious artifacts, a critical gap remains in understanding how developers actually adopt and integrate these unsafe models into downstream software systems. We bridge this gap by conducting a large-scale empirical study on the usage and propagation of 16,977 known vulnerable models on HuggingFace flagged by InsightDB. We track their prevalence across GitHub repositories, software packages, fine-tuning chains, and deployed applications. Our analysis reveals that, despite being publicly flagged, these vulnerable models appear in over 21,000 code locations across more than 5,000 open-source repositories, with 3,757 distinct flagged models (22.1\%) actively referenced in at least one project. To understand how these models are integrated in code and identify recurring usage patterns, we categorize and analyze the recurring patterns surrounding them via a customized pipeline that combines Code Property Graphs extraction with a tailored Frequent Subgraph Mining approach. Across all patterns, we find no evidence of provenance verification or model sanitization logic, with developers frequently bypassing security parameters or delegating trust to permissive framework defaults. These findings underscore that scanning model files alone is insufficient, highlighting the urgent need for context-aware security analysis and stricter integration guardrails at the application layer of the AI supply chain.
Updated: 2026-10-09 02:27:26
标题: 人工智能供应链中的脆弱模型：采纳、传播和使用模式的实证研究
摘要: 人工智能（AI）模型的快速整合已经建立了一个复杂的供应链，但许多这些共享的工件存在严重的漏洞，如不安全的序列化和架构后门。尽管先前的努力已经广泛扫描模型注册表以标记恶意工件，但仍存在一个关键差距，即了解开发人员实际如何采用和集成这些不安全的模型到下游软件系统中。我们通过对HuggingFace上由InsightDB标记的16,977个已知有漏洞的模型的使用和传播进行了大规模实证研究来弥补这一差距。我们跟踪它们在GitHub存储库、软件包、微调链和部署应用程序中的普及程度。我们的分析显示，尽管被公开标记，这些有漏洞的模型出现在超过5,000个开源存储库中的21,000多个代码位置中，其中有3,757个不同的被标记模型（22.1\%）至少在一个项目中被引用。为了了解这些模型如何在代码中集成并识别重复使用模式，我们通过一个结合了代码属性图提取和定制频繁子图挖掘方法的自定义流水线对围绕它们的重复模式进行分类和分析。在所有模式中，我们发现没有来自溯源验证或模型消毒逻辑的证据，开发人员经常绕过安全参数或将信任委托给宽松的框架默认值。这些发现强调，仅仅扫描模型文件是不够的，强调了在AI供应链的应用层急需具有上下文意识的安全分析和更严格的集成防护。
更新时间: 2026-10-09 02:27:26
领域: cs.CR,cs.SE
The False Allure of "Security by Obscurity": A Game-Theoretic Approach
"Security by obscurity," or the idea that hiding how a system works helps protect it from adversaries, has long been poorly regarded in the security community. Yet its intuitive appeal persists, and it has resurfaced repeatedly in industry practices and public debates over the years. The logic, on the one hand, is that open scrutiny of systems is the best way to find and respond to problems; and on the other hand, that too much openness can increase risk by giving adversaries a roadmap for attack. This isn't purely a technical question, but a human one: it's about how attackers and good-faith security researchers respond to openness in the real world. As such, security experts base their stance on decades of community experience; however, the field lacks empirical or theoretical foundations to back up these long-standing beliefs. We provide the first rigorous modeling of "security by obscurity," aiming to capture how rational actors will respond to openness or obscurity in fallible systems. We prove that in equilibrium under realistic model conditions, "obscured" systems are more likely to suffer successful attacks than "open" systems --- establishing theoretical support for the longstanding received wisdom.
Updated: 2026-10-09 01:33:16
标题: “‘安全性靠深藏不露’的虚假诱惑：一种博弈论方法”
摘要: “安全性通过不透明性”，或者隐藏系统如何工作的想法有助于保护系统免受敌对方的攻击，长期以来在安全社区中受到质疑。然而，它的直观吸引力仍然存在，并且多年来在工业实践和公众辩论中反复出现。一方面，逻辑是系统的公开审查是发现和解决问题的最佳方式；另一方面，过度的开放性可能会增加风险，因为这会给敌对方提供攻击的路线图。这不仅仅是一个技术问题，而是一个人类问题：攻击者和善意的安全研究人员如何在现实世界中对待开放性。因此，安全专家基于几十年的社区经验来制定立场；然而，该领域缺乏实证或理论基础来支持这些长期以来的信念。我们提供了对“安全性通过不透明性”的第一个严格建模，旨在捕捉理性行为者将如何响应有缺陷系统中的开放性或不透明性。我们证明，在现实模型条件下的均衡状态下，“不透明”系统比“开放”系统更有可能遭受成功攻击 --- 为长期以来的普遍接受的智慧提供了理论支持。
更新时间: 2026-10-09 01:33:16
领域: cs.GT,cs.CR,cs.CY,econ.TH
A Multi-Scenario UAV RF Dataset with Real-World Acquisition and Signal Processing Benchmarking
We present a real-world multi-scenario unmanned aerial vehicle (UAV) radio frequency (RF) dataset, namely DRFF-R2, which is collected using a dedicated acquisition platform under diverse operational conditions. All signals are acquired within a unified framework to ensure consistency in hardware configuration and environmental settings. The dataset is systematically organized into seven well-defined subsets corresponding to different operational and signal composition scenarios to facilitate structured experimentation. Each file follows a clearly annotated naming convention to enable convenient data indexing and reproducible analysis. The dataset contains RF recordings from 26 UAV units spanning 8 distinct models, captured across varying flight states, altitudes, speeds, acquisition days, and receiver configurations. By covering diverse acquisition settings and signal compositions, the dataset provides a comprehensive resource for future UAV RF signal research, including RF fingerprinting (RFF) identification, model-level recognition, flight state analysis, time-varying RFF study, and interference-aware signal processing.
Updated: 2026-10-09 01:31:00
标题: 一个具有真实世界获取和信号处理基准测试的多场景无人机射频数据集
摘要: 我们提供了一个真实世界的多场景无人机（UAV）无线电频率（RF）数据集，即DRFF-R2，该数据集是在不同操作条件下使用专用采集平台收集的。所有信号都是在统一框架内获取的，以确保硬件配置和环境设置的一致性。该数据集被系统地组织成七个明确定义的子集，对应不同的操作和信号组合场景，以促进结构化实验。每个文件都遵循明确注释的命名惯例，以便进行方便的数据索引和可重复的分析。该数据集包含来自26个无人机单位的RF记录，跨越8种不同型号，捕获了不同飞行状态、高度、速度、采集天数和接收器配置。通过覆盖多样的采集设置和信号组合，该数据集为未来无人机RF信号研究提供了全面的资源，包括RF指纹（RFF）识别，型号级别识别，飞行状态分析，时变RFF研究和干扰感知信号处理。
更新时间: 2026-10-09 01:31:00
领域: eess.SP,cs.CR
Single-Shuffle Full-Open Card-Based Secure Computation Protocols for Any Function
A card-based secure computation protocol is a method for $n$ parties to compute a function $f$ on their private inputs $(x_1,\ldots,x_n)$ using physical playing cards, in such a way that the suits of revealed cards leak no information beyond the value of $f(x_1,\ldots,x_n)$. A \textit{single-shuffle full-open} protocol is a minimal model of card-based secure computation in which, after the parties place face-down cards representing their inputs, a single shuffle operation is performed and then all cards are opened to derive the output. Despite the simplicity of this model, the class of functions known to admit single-shuffle full-open protocols has been limited to a few small examples. In this work, we prove for the first time that every function admits a single-shuffle full-open protocol. We present two constructions that offer a trade-off between the number of cards and the complexity of the shuffle operation. These feasibility results are derived from a novel connection between single-shuffle full-open protocols and a cryptographic primitive known as \textit{Private Simultaneous Messages} protocols, which has rarely been studied in the context of card-based cryptography. We also present variants of single-shuffle protocols in which only a subset of cards are revealed. These protocols reduce the complexity of the shuffle operation compared to existing protocols in the same setting.
Updated: 2026-10-09 00:05:31
标题: 单洗牌全开卡片基础的安全计算协议，适用于任何函数
摘要: 一种基于卡片的安全计算协议是一种方法，用于$n$方计算一个函数$f$，使用实体纸牌表示他们的私有输入$(x_1,\ldots,x_n)$，使得揭示的卡片花色不泄露除$f(x_1,\ldots,x_n)$值外的任何信息。一个\textit{单洗牌全公开}协议是基于卡片的安全计算的最小模型，在该模型中，当各方放置代表他们输入的牌时，进行一次洗牌操作，然后打开所有卡片以导出输出。尽管这个模型简单，已知能够接受单洗牌全公开协议的函数类被限制在少数几个小例子中。在这项工作中，我们首次证明了每个函数都能接受一个单洗牌全公开协议。我们提出了两种构造，介于卡片数量和洗牌操作复杂度之间的权衡。这些可行性结果源于单洗牌全公开协议和一种称为\textit{私密同时消息}协议之间的新颖联系，这种协议在基于卡片的密码学中很少被研究。我们还提出了单洗牌协议的变体，其中仅揭示部分卡片。与相同环境中的现有协议相比，这些协议减少了洗牌操作的复杂性。
更新时间: 2026-10-09 00:05:31
领域: cs.CR
Reward Hacking as a Transmissible Strategy in Multi-Agent Coding Systems
Multi-agent systems are increasingly used to solve coding tasks by coordinating agents that share code and plans. These agents are often optimized with reinforcement learning (RL) using automated evaluators, but bypassing evaluation can earn reward without producing a genuine solution. Prior work has largely treated reward hacking as a single-agent failure. In this work, we show that a reward-hacking strategy elicited in one agent can propagate to others and affect system coordination. Across four open-weight backbones, coding RL elicits and amplifies reward hacking in two settings: explicit mechanism descriptions in the system prompt and mid-training that models upstream exposure to facts about the coding environment. Using post-RL agents from the mid-training setting as sources, we observe propagation through code and shared memory, and across roles through natural-language plans. Even agents based on small open-weight models induce reward hacking in stronger closed-weight targets. On hard tasks, propagation can intensify along a transmission chain after the original source leaves as the code acquires more persuasive justifications. We also theoretically analyze conditions under which reward hacking amplifies, persists, or decays across agent interactions. Beyond propagation, reward hacking can bias task routing toward reward-hacking agents and survive candidate aggregation. As potential mitigations, we evaluate code review by the coder or a separate verifier, both of which reduce delivered hacks. Together, these results establish reward hacking as a transmissible strategy whose effects extend beyond the agent in which it first emerges.
Updated: 2026-10-09 00:02:59
标题: 奖励欺骗作为多智能编码系统中可传播的策略
摘要: 多智能体系统越来越被用来通过协调共享代码和计划的智能体来解决编码任务。这些智能体通常通过使用自动评估器进行强化学习（RL）进行优化，但绕过评估可以获得奖励而不产生真正的解决方案。先前的工作主要将奖励黑客视为单个智能体的失败。在这项工作中，我们展示了一种在一个智能体中引发的奖励黑客策略可以传播到其他智能体并影响系统协调。在四个开放权重骨干中，编码RL在两个设置中引发和放大奖励黑客：系统提示中的显式机制描述和模拟上游对编码环境事实的暴露。使用来自中间训练设置的后RL智能体作为资源，我们观察到通过代码和共享内存传播，并通过自然语言计划跨角色传播。即使基于小型开放权重模型的智能体也会引发较强的封闭权重目标中的奖励黑客。在难任务中，传播可以在原始来源离开后沿着传输链加剧，因为代码获得了更具说服力的理由。我们还在理论上分析了奖励黑客在智能体相互作用中放大、持续或衰减的条件。除了传播，奖励黑客还可能偏向于奖励黑客智能体的任务路由，并在候选聚合中存活。作为潜在的缓解措施，我们评估了由编码者或单独验证者进行的代码审查，这两种方法都会减少交付的黑客。综上所述，这些结果将奖励黑客确定为一种可传播的策略，其影响超出了首次出现的智能体。
更新时间: 2026-10-09 00:02:59
领域: cs.AI,cs.CR,cs.MA