|
创作新主题 |
社区所有版块导航
DATA
| docker Elasticsearch |
WEB开发
| linux MongoDB Redis DATABASE NGINX 其他Web框架 web工具 zookeeper tornado NoSql Bootstrap js peewee Git bottle IE MQ Jquery |
机器学习
| 机器学习算法 |
产品
| 短视频 |
印度
| 印度 |
一周十大热门主题
点击阅读原文访问arxivdaily.com，涵盖CS|物理|数学|经济|统计|金融|生物|电气领域，更有搜索、收藏等功能！
cs.LG 方向，今日共计238篇
大模型相关(26篇)
【1】Towards Efficient and Evidence-grounded Mobility Prediction with LLM-Driven Agent
标题：利用LLM驱动的代理实现高效且基于证据的移动性预测
链接：https://arxiv.org/abs/2606.05130
摘要：
摘要：
【2】STaR-Quant: State-Time Consistent Post-Training Quantization for Diffusion Large Language Models
标题：STaR-Quant：扩散大型语言模型的状态时间一致训练后量化
链接：https://arxiv.org/abs/2606.04945
摘要：扩散大语言模型（DLLM）最近已经成为自回归LLM的一个很有前途的替代方案，它通过双向上下文的迭代掩蔽去噪来生成文本。然而，它们的大模型尺寸和迭代去噪过程引入了大量的内存和计算开销，从而激励了训练后量化以实现有效部署。在本文中，我们确定了低比特DLLM量化的两个关键挑战：状态依赖的激活差异和时间误差积累。掩蔽和未掩蔽的令牌在每个去噪步骤中表现出不同的激活分布，而量化误差可以在迭代解码期间跨步骤累积。为了解决这些挑战，我们提出STaR-Quant，DLLM的状态时间一致的PTQ框架。STaR-Quant引入了状态引导激活转换（SGAT），通过统一的静态权重侧转换将屏蔽和非屏蔽令牌分配到不同的激活转换空间。它还引入了时间注意力补偿（TAC），通过一个轻量级的块对角仿射映射来纠正量化的注意力表示。在代表性DLLM上的实验表明，STaR-Quant在强PTQ基线上持续改进低位权重激活量化，同时在FP 16部署上提供高达1.69倍的加速和3.14倍的内存节省。
摘要：Diffusion large language models (DLLMs) have recently emerged as a promising alternative to autoregressive LLMs by generating text through iterative masked denoising with bidirectional context. However, their large model sizes and iterative denoising process introduce substantial memory and computational overhead, motivating post-training quantization for efficient deployment. In this paper, we identify two key challenges for low-bit DLLM quantization: state-dependent activation disparity and temporal error accumulation. Masked and unmasked tokens exhibit different activation distributions within each denoising step, while quantization errors can accumulate across steps during iterative decoding. To address these challenges, we propose STaR-Quant, a state-time consistent PTQ framework for DLLMs. STaR-Quant introduces State-Guided Activation Transformation (SGAT) to assign masked and unmasked tokens to different activation transformation spaces with a unified static weight-side transformation. It further introduces Temporal Attention Compensation (TAC) to correct the quantized attention representation via a lightweight block-diagonal affine mapping. Experiments on representative DLLMs demonstrate that STaR-Quant consistently improves low-bit weight-activation quantization over strong PTQ baselines, while delivering up to 1.69x speedup and 3.14x memory saving over FP16 deployment.
【3】Sequential Data Poisoning in LLM Post-Training
标题：LLM后训练中的序列数据中毒
链接：https://arxiv.org/abs/2606.04929
摘要：LLM后培训通过多个阶段进行，例如，监督微调（SFT），然后是来自人类反馈的强化学习（RLHF）或直接偏好优化（DPO），其中每个阶段从不同的，可能不可信的来源提取数据。现有文献假设数据中毒攻击可能发生在每个训练阶段，但忽略了多个攻击者的可能性。为了研究整个后训练管道的可信度，我们提出了顺序数据中毒的威胁模型，其中多个对手分别毒害SFT和偏好数据集。在这种威胁模型下，我们确定了单一攻击者的错觉：孤立评估的每个对手似乎构成了可以忽略不计的威胁。然而，当对手跨阶段合作时，真正的脆弱性就会暴露出来。在SFT $\到$ DPO管道中，它们的贡献是相加的：将固定的毒物预算分配到各个阶段的效果优于将其单独集中在任何一个阶段。在SFT $\to$ PPO管道中，它们的贡献是互补的：SFT和奖励模型中毒都没有单独成功，但它们的组合却成功了。这些研究结果表明，安全分析的个人培训后阶段系统地低估了复合漏洞，只出现在他们的互动。代码可在https://github.com/jcksanderson/sequential-poisoning上获得。
摘要：LLM post-training proceeds through multiple stages, e.g., supervised fine-tuning (SFT) followed by reinforcement learning from human feedback (RLHF) or direct preference optimization (DPO), where each stage draws data from different, potentially untrusted sources. Existing literature assumes data poisoning attacks may occur at each training stage, but neglects the possibility of multiple attackers. To study the trustworthiness of the entire post-training pipeline, we propose the threat model of sequential data poisoning, where multiple adversaries separately poison the SFT and preference datasets. Under this threat model, we identify the single-attacker illusion: each adversary, evaluated in isolation, appears to pose a negligible threat. Yet when adversaries collaborate across stages, the true vulnerability is revealed. In the SFT $\to$ DPO pipeline, their contributions are additive: splitting a fixed poison budget across stages outperforms concentrating it in either stage alone. In the SFT $\to$ PPO pipeline, their contributions are complementary: neither SFT nor reward model poisoning succeeds individually, yet their combination does. These findings show that security analyses of individual post-training stages systematically underestimate compound vulnerabilities that emerge only from their interaction. Code is available at https://github.com/jcksanderson/sequential-poisoning.
【4】Data Attribution in Large Language Models via Bidirectional Gradient Optimization
标题：通过双向梯度优化实现大型语言模型中的数据属性
链接：https://arxiv.org/abs/2606.04928
备注：Presented at the AI Governance (AIGOV) Workshop at AAAI 2026
摘要：大型语言模型（LLM）越来越多地部署在不同的应用程序中，这引发了治理，问责制和数据来源的关键问题。了解哪些训练数据对模型的输出影响最大仍然是一个基本的开放问题。我们通过扩展逆公式，通过自回归LLM的训练数据属性（TDA）来解决这一挑战：如果模型在训练过程中看到了生成的输出，训练数据会受到怎样的影响？我们的方法在生成的文本样本上使用双向梯度优化（梯度上升和下降）来扰动基础模型，并测量训练样本之间的损失变化。我们的框架支持在任意数据粒度的属性，使事实和风格的属性。我们根据具有已知数据集的预训练模型的基线评估了我们的方法，并表明它优于以前在影响力指标方面的工作，从而增强了模型的可解释性，这是负责任的人工智能系统的基本要求。
摘要：Large Language Models (LLMs) are increasingly deployed across diverse applications, raising critical questions for governance, accountability, and data provenance. Understanding which training data most influenced a model's output remains a fundamental open problem. We address this challenge through training data attribution (TDA) for auto-regressive LLMs by expanding upon the inverse formulation: How would training data be affected if the model had seen the generated output during training? Our method perturbs the base model using bidirectional gradient optimization (gradient ascent and descent) on a generated text sample and measures the resulting change in loss across training samples. Our framework supports attribution at arbitrary data granularity, enabling both factual and stylistic attribution. We evaluate our method against baselines on pretrained models with known datasets, and show that it outperforms previous work on influence metrics, thereby enhancing model interpretability, an essential requirement for accountable AI systems.
【5】Geometry-Aware Distillation for Prompt Tuning Biomedical Vision-Language Models
标题：几何感知蒸馏用于快速调整生物医学视觉语言模型
链接：https://arxiv.org/abs/2606.04922
备注：Preprint. Code is available at https://github.com/tientrandinh/OGKD
摘要：当前基于神经网络和基于适配器的视觉语言模型（VLM）的调整对于医学成像是有吸引力的，其中临床数据敏感性有利于冻结的骨干和注释是有限的。然而，这些方法通常仅优化地面实况类，将所有其他类视为同样不正确，忽略临床上有意义的类关系，并在有限监督设置中产生不稳定的决策边界。我们提出了全几何知识蒸馏（OGKD），一个新的框架，注入类关系结构到教师产生方向性的目标，保留地面真理，同时尊重类间几何。使用这些目标，我们开发了两个蒸馏损失：全局几何感知蒸馏（GAD）对全局图像令牌进行操作，标签引导几何蒸馏（LGD）将相同的几何应用于注意的补丁令牌，以提高细粒度对齐。通过对11个广泛使用的医学数据集进行全面的实验和分析，以进行基础到新的和Few-Shot评估，我们的OGKD实现了更好的性能，与所有现有的最先进的VLM适配对应物相比，平均绝对增益为1.7%-2.8%，持续提高准确性。它还可以鲁棒地推广到看不见的类，并产生比其他方法更可靠的预测。我们的代码可以在https://github.com/tientrandinh/OGKD上找到。
摘要
：Current prompt-based and adapter-based tuning of vision-language models (VLMs) is attractive for medical imaging, where clinical data sensitivity favors frozen backbones and annotations are limited. However, these methods typically optimize only the ground-truth class, treating all other classes as equally incorrect, ignoring clinically meaningful class relations and yielding unstable decision boundaries in limited-supervision settings. We propose Omni-Geometry Knowledge Distillation (OGKD), a new framework that injects class-relation structure into the teacher to produce directional targets that preserve the ground truth while respecting inter-class geometry. Using these targets, we develop two distillation losses: Global Geometry-Aware Distillation (GAD) operates on the global image token, and Label-Guided Geometry Distillation (LGD) applies the same geometry to attentive patch tokens to improve fine-grained alignment. Across comprehensive experiments and analyses on 11 widely-used medical datasets for base-to-novel and few-shot evaluations, our OGKD achieves substantially better performance, consistently improving accuracy by an average absolute gain of 1.7%-2.8% over all prior state-of-the-art VLM adaptation counterparts. It also robustly generalizes to unseen classes and yields more reliable predictions than other approaches. Our code is available at https://github.com/tientrandinh/OGKD.
【6】Beyond Objective Equivalence: Constraint Injection for LLM-Based Optimization Modeling on Vehicle Routing Problems
标题：超越目标等效：基于LLM的车辆路径问题优化建模的约束注入
链接：https://arxiv.org/abs/2606.04816
备注：28 pages
摘要：大型语言模型（LLM）越来越多地将自然语言优化问题转化为可执行的求解器代码。然而，对于约束密集的运筹学（OR）问题，现有的数据过滤和训练管道在很大程度上依赖于目标等价信号，如差分测试和答案一致性，程序可以通过这些信号，同时添加虚假约束或默默地忽略所需的约束，只要这些约束对测试实例没有约束力。我们提出了约束注入，它使用可行的探针暴露虚假的过约束和一个约束违反探针，揭示沉默的约束遗漏。与差分测试相结合，它形成了一个双重验证器。我们实例化和评估车辆路径问题（VRP），一个代表性的约束密集的组合优化测试平台，耦合操作约束。我们开发了VRPCoder，这是一个8B端到端模型，可以将自然语言VRP场景转换为Guidance脚本，以及一个经过专家验证的VRP基准套件，涵盖21个变体。验证器在数据合成过程中被重用为拒绝采样过滤器，并在组相对策略优化（GRPO）中作为每次推出的奖励。在四个VRP基准测试中，VRPCoder-GRPO达到93%的平均Pass@1，在三个基准测试中优于Gemini-3.1-Pro Preview，超过Claude-Sonnet-4.5 28个平均点，并超过之前的OR-LLM 78个平均点。
摘要：Large language models (LLMs) increasingly translate natural-language optimization problems into executable solver code. Yet for constraint-dense operations research (OR) problems, existing data-filtering and training pipelines largely rely on objective-equivalence signals such as differential testing and answer agreement, which a program can pass while adding spurious constraints or silently omitting required ones, whenever those constraints are non-binding on the tested instance. We propose constraint injection, which uses feasible probes to expose spurious over-constraint and one-constraint-violating probes to reveal silent constraint omission. Combined with differential testing, it forms a dual verifier. We instantiate and evaluate it on vehicle routing problems (VRPs), a representative constraint-dense combinatorial optimization testbed with coupled operational constraints. We develop VRPCoder, an 8B end-to-end model that translates natural-language VRP scenarios into Gurobi scripts, together with an expert-verified VRP benchmark suite covering 21 variants. The verifier is reused as a rejection-sampling filter during data synthesis and as a per-rollout reward in group relative policy optimization (GRPO). Across four VRP benchmarks, VRPCoder-GRPO reaches 93\% average Pass@1, outperforms Gemini-3.1-Pro Preview on three benchmarks, exceeds Claude-Sonnet-4.5 by 28 average points, and surpasses prior OR-LLMs by 78 average points.
【7】Rethinking Continual Experience Internalization for Self-Evolving LLM Agents
标题：重新思考自我发展的LLM代理人的持续经验内化
链接：https://arxiv.org/abs/2606.04703
备注：10 pages, 8 figures
摘要：经验内化将过去交互的上下文经验转换为可重用的参数能力，为大型语言模型（LLM）的持续学习提供了一条有前途的道路。虽然以前的工作主要集中在单迭代传输，我们发现，在多迭代的经验学习，现有的方法遭受渐进的能力崩溃，而不是复合改进。我们通过经验内化的三个重要维度系统地研究了这种失败：（1）经验粒度：我们发现原则层面的经验比实例层面的经验更持久，因为它有效地从特定的细节中提取了可转移的策略。(2)经验注入模式：我们的分析表明，逐步注入显着优于全球注入对齐经验与中间决策状态，一个属性，这是至关重要的长期工具的使用。(3)内部化机制：我们证明了高质量教师轨迹上的非政策上下文蒸馏提供了比政策上下文蒸馏更稳定的训练信号，这本质上受到学生引起的有缺陷状态的局部校正的限制。总之，这些见解为稳定和可持续的经验内化提供了一个简单而强大的配方，为工程自我发展和不断学习LLM提供了具体的指导。
摘要：Experience internalization converts contextual experience from past interactions into reusable parametric capability, offering a promising path toward continual learning in large language models (LLMs). While prior work has predominantly focused on single-iteration transfer, we discover that under multi-iteration experience learning, existing methods suffer from a progressive capability collapse rather than compounding improvement. We systematically examine this failure through three vital dimensions of experience internalization: (1) Experience Granularity: We find that principle-level experience is more durable than instance-level experience, as it effectively abstracts transferable strategies away from trajectory-specific details. (2) Experience Injection Pattern: Our analysis reveals that step-wise injection significantly outperforms global injection by aligning experience with intermediate decision states, a property that is critical for long-horizon tool use. (3) Internalization Regime: We demonstrate that off-policy context-distillation on high-quality teacher trajectories provides a substantially more stable training signal than on-policy context-distillation, which is inherently limited by local corrections on student-induced flawed states. Together, these insights yield a simple yet robust recipe for stable and sustainable experience internalization, providing concrete guidance for engineering self-evolving and continually learning LLMs.
【8】VentAgent: When LLMs Learn to Breathe -- Multi-Objective Arbitration for ARDS Ventilation
标题：VentAgent：当LLM学会呼吸时--急性呼吸窘迫综合征通气的多目标仲裁
链接：https://arxiv.org/abs/2606.04632
摘要：
摘要：
【9】QuBLAST: A Framework for Quantizing Large Language Models with Block-Level Compression Approach and Activation Scaling Strategy
标题：QuAMPS：一个使用块级压缩方法和激活缩放策略量化大型语言模型的框架
链接：https://arxiv.org/abs/2606.04620
备注：10 pages, 9 figures, 5 tables
摘要：
摘要：
【10】Beyond Symmetric Alignment: Spectral Diagnostics of Modality Imbalance in Vision-Language Models in the Medical Domain
标题：超越对称对齐：医学领域视觉语言模型中情态失衡的光谱诊断
链接：https://arxiv.org/abs/2606.04613
备注：10 pages, 3 figures, 9 tables
摘要：
摘要：
【11】SparDA: Sparse Decoupled Attention for Efficient Long-Context LLM Inference
标题：SparDA：稀疏去耦合注意力以实现高效的长上下文LLM推理
链接：https://arxiv.org/abs/2606.04511
摘要：
摘要：
【12】Global Sketch-Based Watermarking for Diffusion Language Models
标题：扩散语言模型的基于全局草图的水印
链接：https://arxiv.org/abs/2606.04486
摘要：
摘要：
【13】Stateful Visual Encoders for Vision-Language Models
标题：视觉语言模型的状态视觉编码器
链接：https://arxiv.org/abs/2606.04433
备注：Project page: this https URL
摘要：
摘要：
【14】From Symbolic to Geometric: Enabling Spatial Reasoning in Large Language Models
标题：从符号到几何：在大型语言模型中实现空间推理
链接：https://arxiv.org/abs/2606.04381
摘要：
摘要：
【15】Deliberate Evolution: Agentic Reasoning for Sample-Efficient Symbolic Regression with LLMs
标题：刻意进化：利用LLM进行样本高效符号回归的显式推理
链接：https://arxiv.org/abs/2606.04360
备注：ICML 2026
摘要：
摘要：
【16】RL Excursions during Pre-Training: Re-examining Policy Optimization for LLM training
标题：预训练期间的RL游览：重新审视LLM训练的政策优化
链接：https://arxiv.org/abs/2606.04272
摘要：
摘要：
【17】Recover-LoRA for Aggressive Quantization: Reclaiming Accuracy in 2-Bit Language Models via Low-Rank Adaptation with Knowledge Distillation on Synthetic Data
标题：用于激进量化的RECE-LoRA：通过低等级自适应和合成数据的知识提炼来恢复2位语言模型的准确性
链接：https://arxiv.org/abs/2606.04238
摘要：
摘要：
【18】Supportive Token Revealing for Fast Diffusion Language Model Decoding
标题：快速扩散语言模型解码的支持性令牌揭示
链接：https://arxiv.org/abs/2606.04236
摘要：
摘要：
【19】KODA: Contrastive Representation Comparison and Alignment for Vision-Language Foundation Models
标题：KODA：视觉语言基础模型的对比表示比较和对齐
链接：https://arxiv.org/abs/2606.04180
摘要：
摘要：
【20】Large Language Models Hack Rewards, and Society
标题：大型语言模型Hack奖励和社会
链接：https://arxiv.org/abs/2606.04075
备注：14 pages, 9 figures, 7 tables
摘要：
摘要：
【21】Covert Influence Between Language Models
标题：语言模型之间的隐性影响
链接：https://arxiv.org/abs/2606.04071
摘要：
摘要：
【22】LLM Compression with Jointly Optimizing Architectural and Quantization choices
标题：LLM压缩，联合优化架构和量化选择
链接：https://arxiv.org/abs/2606.04063
摘要：
摘要：
【23】The Invisible Lottery: How Subtle Cues Steer Algorithm Choice in LLM Code Generation
标题：隐形彩票：微妙线索如何在LLM代码生成中引导算法选择
链接：https://arxiv.org/abs/2606.04057
摘要：
摘要：
【24】LiftQuant: Continuous Bit-Width LLM via Dimensional Lifting and Projection
标题：LiftQuant：通过维度提升和投影的连续位宽LLM
链接：https://arxiv.org/abs/2606.04050
备注：ICML 2026 Spotlight
摘要：
摘要：
【25】Unpredictable Safety: Domain-Dependent Compliance and the Transparency Gap in Open-Weight LLMs
标题：不可预测的安全性：领域相关合规性和开放量级LLM的透明度差距
链接：https://arxiv.org/abs/2606.04035
摘要：
摘要：
【26】PRECISE: Reducing the Bias of LLM Evaluations Using Prediction-Powered Ranking Estimation
标题：PRECISE：使用预测助力排名估计减少LLM评估的偏差
链接：https://arxiv.org/abs/2601.18777
备注：Accepted at AAAI 2026 - Innovative Applications of AI (IAAI-26)
摘要：
摘要：
Graph相关(图学习|图神经网络|图优化等)(11篇)
【1】Graph Set Transformer
标题：图形集Transformer
链接
：https://arxiv.org/abs/2606.05116
备注：10 pages, 1 figure, conference
摘要：
摘要：
【2】FLAGG: Flexible Autoregressive Graph Generation
标题：FLAGG：灵活的自回归图生成
链接：https://arxiv.org/abs/2606.05067
备注：Accepted for publication at JMLR, currently in press
摘要：深度图生成的全景图跨越了两个极端：一次性模型和连续模型。前者联合生成节点和边，而后者对它们进行自回归采样。每种方法在不同的图域中表现更好，这取决于大小和拓扑结构，但都不适用于所有的图类别。例如，一次性方法难以生成大型图，而顺序方法在较小的图上表现不佳。克服这些限制的一个可能的方法是在一个独特的系统中灵活地结合这两种方法。在这项工作中，我们提出了FLAGG（灵活的自回归图生成）框架，它顺序地生成具有一次性模型的图的部分。FLAGG可以应用任何一次性模型，使其自回归，允许灵活选择顺序策略。该策略通过随机节点移除过程来指定，插入模型学习反转该过程。我们评估FLAGG与DiGress一杆模型上的几个数据集不同的图形大小和域。我们表明，该方法优于单次和自回归基线的采样质量。
摘要：The Deep Graph Generation's panorama spans two extremes: one-shot and sequential models. The former generates nodes and edges jointly, while the latter samples them autoregressively. Each method performs better in different graph domains depending on size and topology, but neither is applicable to all graph categories. For instance, one-shot methods struggle with generating large graphs, while sequential methods underperform on smaller graphs. A possible way to overcome these limitations is to flexibly combine the two methods in a unique system. In this work, we propose the FLAGG (Flexible Autoregressive Graph Generation) framework, which sequentially generates portions of graphs with one-shot models. FLAGG can apply any one-shot model to make it autoregressive, allowing flexibility in choosing the sequential policy. This policy is specified through a stochastic node removal process, which an Insertion Model learns to reverse. We evaluate FLAGG with the DiGress one-shot model on several data sets of different graph sizes and domains. We show that the approach outperforms both one-shot and autoregressive baselines in terms of sampling quality.
【3】Graph Cascades: Contagion-Based Mesoscopic Rewiring for Structure-Aware Graph Machine Learning
标题：图级联：基于传染性的介观重新布线，用于结构感知图机器学习
链接：https://arxiv.org/abs/2606.05046
摘要：我们介绍了图级联，图神经网络（GNN）和图Transformers（GT）的中观重新布线策略，它捕获了超越纯粹局部边缘或完全全局注意力的中间尺度图结构。使用基于传染的扩散过程，Graph Cascades构造，在O（|V| +| E|时间，一个辅助图，其中由重复多跳增强支持的节点对被提升为直接邻居。我们从理论上表征时，基于重布线的帮助：充分条件下，基于重布线的边缘选择更标签对齐比直接邻接，SBM证人，其中两跳加固是完全homophilic，和正式的介观连接通过图形有效电阻。从经验上讲，在节点分类基准中，Graph Cascades改进了多个GNN和稀疏GT骨干，在异嗜性和中度到高度同嗜性图上观察到最可靠的收益。理论条件还确定了介观重新布线不太可能是有益的制度-低度规则图和结构瓶颈图-这些预测与观察到的故障相匹配。我们还观察到紧密的相关性之间的性能和结构特性的重新布线图。
摘要：We introduce Graph Cascades, a mesoscopic rewiring strategy for Graph Neural Networks (GNNs) and Graph Transformers (GTs) that captures intermediate-scale graph structure beyond purely local edges or fully global attention. Using contagion-based diffusion processes, Graph Cascades constructs, in O(|V|+|E|) time, an auxiliary graph where node pairs supported by repeated multi-hop reinforcement are promoted to direct neighbors. We theoretically characterize when reinforcement-based rewiring helps: sufficient conditions under which reinforcement-based edge selection is more label-aligned than direct adjacency, an SBM witness in which two-hop reinforcement is perfectly homophilic, and a formalization of mesoscopic connectivity via graph effective resistance. Empirically, across node-classification benchmarks, Graph Cascades improves multiple GNN and sparse-GT backbones, with the most reliable gains observed on heterophilic and moderate- to high-degree homophilic graphs. The theoretical conditions also identify regimes where mesoscopic rewiring is unlikely to be beneficial -- low-degree regular graphs and graphs with structural bottlenecks -- and these predictions match the observed failures. We additionally observe tight correlations between performance and structural properties in the rewired graphs.
【4】AIP: A Graph Representation for Learning and Governing Agent Skills
标题：AIP：学习和管理代理技能的图形表示
链接：https://arxiv.org/abs/2606.04781
摘要：今天的代理人技能主要包括自由形式的散文，要求代理人阅读，解释和重新推导如何在每个会话中采取行动。这带来了两个复合成本：降低了实现繁重任务的可靠性，以及技能创建和改进的困难，因为编辑散文是一个脆弱的过程，人类和代理都在努力，特别是对于模型训练中代表性不足的特定领域的程序知识。代理指令协议（AIP）通过将技能建模为有向执行图来解决这两个问题：由确定性脚本或自然语言描述支持的离散步骤作为节点，由显式类型的输入/输出边连接，并由模式验证的YAML规范管理。编译器元技能将现有的人类编写的技能转换为这种形式。好处是双重的。首先，将人类书写的技能编译为AIP将Claude Sonnet的平均任务奖励从0.60提高到0.71，并且在SkillsBench的27个真实代理任务中的通过率从53%提高到67%-这是一个统计学上显著的收益（Wilcoxon符号秩p = 0.011），赢得12个任务，以13个平局赢得2个任务-通常在更少的挂钟时间内。该图将经过审查的可运行单元交付给代理，而不是要求它从自然语言中重新派生代码，命令和工具调用。第二，在创建和改进方面，由于每项技能都是经过模式验证的，功能上可测试的，并且可以逐节点寻址，因此可以精确地诊断和修复故障。两个作者技能失败被追溯到脚本级别。在调整AIP规范并重新编译后，两者都恢复为零回归（一个任务从0/5到5/5），将技能改进转化为可测量的调整循环，而不是散文重写。同样的图结构支持语料库级别的治理和技能自省，并为技能的强化学习提供了自然的动作空间。
摘要：Agent Skills today consist largely of free-form prose requiring the agent to read, interpret, and re-derive how to act in every session. This imposes two compounding costs: reduced reliability on implementation-heavy tasks, and difficulty in skill creation and improvement, since editing prose is a fragile process that both humans and agents struggle with, particularly for domain-specific procedural knowledge underrepresented in model training. The Agent Instruction Protocol (AIP) addresses both by modeling a skill as a directed execution graph: discrete steps as nodes backed by deterministic scripts or natural-language descriptions, connected by explicit typed input/output edges, and governed by a schema-validated YAML specification. A compiler meta-skill translates existing human-written skills into this form. The benefits are twofold. First, compiling human-written skills to AIP raised Claude Sonnet's mean task reward from 0.60 to 0.71 and pass rate from 53% to 67% across 27 real agent tasks from SkillsBench - a statistically significant gain (Wilcoxon signed-rank p = 0.011), winning 12 tasks to 2 with 13 ties - often in less wall-clock time. The graph delivers vetted, runnable units to the agent rather than asking it to re-derive code, commands, and tool calls from natural language. Second, on creation and improvement, because each skill is schema-validated, functionally testable, and addressable node-by-node, failures can be diagnosed and repaired precisely. Two authored-skill failures were traced to the script level. After adjusting the AIP spec and recompiling, both recovered with zero regressions (one task going from 0/5 to 5/5), turning skill improvement into a measurable tuning loop rather than a prose rewrite. That same graph structure supports corpus-level governance and skill introspection, and provides a natural action space for reinforcement learning over skills.
【5】Graph-Guided Universum Learning in Generalized Eigenvalue Proximal SVMs for Alzheimer's Disease Classification
标题：用于阿尔茨海默病分类的广义特征值近端支持机中的图引导Universum学习
链接：https://arxiv.org/abs/2606.04699
摘要：早期准确检测阿尔茨海默病（AD）对于及时干预和疾病管理非常重要。广义特征值近似支持向量机（GEPSVM）及其基于Universum的变体在AD分类中表现出了良好的效果。然而，现有的方法处理Universum样本作为独立的点，并没有考虑它们之间的几何关系。本文提出了两个图形引导的Universum学习模型，即UG-GEPSVM和IUG-GEPSVM，AD与认知正常（CN）分类使用结构MRI数据。在建议的框架中，轻度认知障碍（MCI）的受试者被用作Universum数据，以提供AD和CN类之间的中间信息。使用高斯相似性、最小生成树连接性和多跳传播在Universum样本上构造图。从该图中，导出捕获MCI样本的几何结构的拉普拉斯矩阵。这种基于拉普拉斯的正则化被纳入到学习过程中，代替传统的独立Universum惩罚项。UG-GEPSVM将此正则化集成到广义特征值公式中，而IUG-GEPSVM使用标准特征值公式扩展了数值稳定的改进GEPSVM框架。在五个不同的噪声水平下使用ICA和PCA特征的ADNI MRI数据集变体的实验表明，所提出的模型始终优于现有的GEPSVM和基于Universum的方法。UG-GEPSVM实现了最高的平均AUC 88.07%，并在增加的噪声水平下保持稳定的性能。统计检验进一步证实了所观察到的改进的重要性。
摘要
：Early and accurate detection of Alzheimer's disease (AD) is important for timely intervention and disease management. Generalized Eigenvalue Proximal Support Vector Machine (GEPSVM) and its Universum-based variants have shown promising results for AD classification. However, existing methods treat Universum samples as independent points and do not consider the geometric relationships among them. This paper proposes two graph-guided Universum learning models, namely UG-GEPSVM and IUG-GEPSVM, for AD versus cognitively normal (CN) classification using structural MRI data. In the proposed framework, mild cognitive impairment (MCI) subjects are used as Universum data to provide intermediate information between AD and CN classes. A graph is constructed over the Universum samples using Gaussian similarity, Minimum Spanning Tree connectivity, and multi-hop propagation. From this graph, a Laplacian matrix is derived that captures the geometric structure of the MCI samples. This Laplacian-based regularization is incorporated into the learning process in place of the conventional independent Universum penalty term. UG-GEPSVM integrates this regularization into the generalized eigenvalue formulation, while IUG-GEPSVM extends the numerically stable improved GEPSVM framework using a standard eigenvalue formulation. Experiments on ADNI MRI dataset variants using ICA- and PCA-based features at five different noise levels show that both proposed models consistently outperform existing GEPSVM and Universum-based methods. UG-GEPSVM achieves the highest average AUC of 88.07% and maintains stable performance under increasing noise levels. Statistical tests further confirm the significance of the observed improvements.
【6】Learning Long Range Spatio-Temporal Representations over Continuous Time Dynamic Graphs with State Space Models
标题：使用状态空间模型学习连续时间动态图上的长范围时空表示
链接：https://arxiv.org/abs/2606.04672
备注：Accepted at ICML 2026
摘要：连续时间动态图（CTDG）提供了一个更丰富的框架，以捕捉细粒度的时间模式，在不断变化的关系数据。长距离信息传播是学习表示时的一个关键挑战，其中重要的是在长时间范围内保留和更新信息。现有的方法限制模型捕捉单跳或局部时间邻域，并未能捕捉多跳或全球结构模式。为了缓解这一问题，我们从第一原理推导出一个参数有效的连续时间动态图（CTDG-SSM）的状态空间建模框架。我们首先介绍了连续时间拓扑感知高阶多项式投影算子（CTT-HiPPO），一种新的基于内存的HiPPO重新制定联合编码的时间动态和图形结构。CTT-HiPPO的解决方案是通过将经典HiPPO解决方案投影到拉普拉斯矩阵的多项式中来获得的，从而产生拓扑感知的存储器更新，该更新允许CTDG（CTDG-SSM）的等效状态空间公式化。然后使用零阶保持方法获得一个计算效率高的离散公式。 在动态链接预测、动态节点分类和序列分类的基准测试中，CTDG-SSM实现了最先进的性能。值得注意的是，它在需要长距离时间（LRT）和空间推理的数据集上实现了巨大的性能增益。
摘要：Continuous-time dynamic graphs (CTDGs) provide a richer framework to capture fine-grained temporal patterns in evolving relational data. Long-range information propagation is a key challenge while learning representations, wherein it is important to retain and update information over long temporal horizons. Existing approaches restrict models to capture one-hop or local temporal neighborhoods and fail to capture multi-hop or global structural patterns. To mitigate this, we derive a parameter-efficient state-space modeling framework for continuous-time dynamic graphs (CTDG-SSM) from first principles. We first introduce continuous-time Topology-Aware higher order polynomial projection operator (CTT-HiPPO), a novel memory-based reformulation of HiPPO to jointly encode temporal dynamics and graph structure. The solution from CTT-HiPPO is obtained by projecting the classical HiPPO solution through a polynomial of the Laplacian matrix, yielding topology-aware memory updates that admit an equivalent state-space formulation for CTDGs (CTDG-SSM). Then a computationally efficient discrete formulation is obtained using the zero-order hold approach for model implementation. Across benchmarks on dynamic link prediction, dynamic node classification, and sequence classification, CTDG-SSM achieves state-of-the-art performance. Notably, it achieves large performance gains on datasets that require long range temporal (LRT) and spatial reasoning.
【7】ALINC: Active Learning for Inductive Node Classification via Graph Sampling
标题：ALISC：通过图采样进行归纳节点分类的主动学习
链接：https://arxiv.org/abs/2606.04647
备注：Accepted at ECML PKDD 2026
摘要：
摘要：
【8】Scaling Novel Graph Generation via Lightweight Structure-Guided Autoregressive Models
标题：通过轻量级结构引导自回归模型扩展新型图生成
链接：https://arxiv.org/abs/2606.04287
摘要：
摘要：
【9】Bayesian Membership Privacy for Graph Neural Networks
标题：图神经网络的Bayesian成员隐私
链接：https://arxiv.org/abs/2606.04069
摘要：
摘要：
【10】QPredSGG: Hybrid Quantum Predicate Learning for Long-Tailed Scene Graph Generation
标题：QPredSGG：用于长尾场景图生成的混合量子预测学习
链接：https://arxiv.org/abs/2606.04689
备注：11 pages, 5 figures
摘要：
摘要：
【11】Structure-Aware Prediction of PROTAC-Mediated Protein Degradability via Graph Neural Networks
标题：通过图神经网络对PROTAC介导的蛋白质降解性的结构感知预测
链接：https://arxiv.org/abs/2606.04021
备注：10 pages, 5 figures, ACM-BCB 2026 Main Conference Full Paper
摘要：
摘要：
Transformer(8篇)
【1】An Open-Source Two-Stage Computer Vision Pipeline for Fine-Grained Vehicle Classification using Vision Transformers
标题：使用视觉转换器进行细粒度车辆分类的开源两级计算机视觉管道
链接：https://arxiv.org/abs/2606.05149
备注：24 pages, 10 figures, venue TBD
摘要：
摘要：
【2】An Empirical Audit of Input Encoders for Multi-Channel Signal Transformers
标题：多通道信号Transformer输入编码器的实证审计
链接：https://arxiv.org/abs/2606.04752
备注：21 pages, 1 figure, 8 tables. Code: https://github.com/OssiLehtinen/channel-encoder-audit
摘要：使用多通道标量信号的Transformers必须将$C$同时值嵌入到每个时间步长的$d_{\text{model}}$维向量中。我们根据经验审计了八个输入编码器-跨越共享标量基线，每通道线性投影，正交正则化，非线性MLP干，块分割级联，通道独立和通道作为令牌架构，以及投影位置编码-在设计用于使通道身份信息化的合成基准上，并在ETTh 1上作为真实数据检查，在下一步负对数似然（NLL）中测量。标题是在一个广泛的“顶层”中实际接近等价的一个：标准的每通道线性投影（nn.Linear（C，$d_{\text{model}}$））匹配该层中的每一个备选方案，直到很小的，统计上真实但实际上适度的差异。两个编码器决定性地输了：共享标量基线，它崩溃的信息理论的原因，我们明确，和通道无关的补丁TST精神基线，这两个基准表现不佳，并普遍过拟合的合成。配对测试解决了两个小的差距：通过一个学习的线性层边缘的正弦位置编码投影其余的小$C$，与直接的几何探针显示的机制是位置通道正交化;一个非线性MLP干边缘他们在最大的$C$，我们测试，与更多的训练数据下的差距缩小。实际的建议是默认使用nn.Linear（C，$d_{\text{model}}$），只有当手头的任务给出了这样做的真正原因时，才使用更详细的东西。本文中再现每个实验的代码和数据可在https://github.com/OssiLehtinen/channel-encoder-audit上获得
摘要
：Transformers consuming multi-channel scalar signals must embed $C$ simultaneous values into one $d_{\text{model}}$-dimensional vector per time step. We empirically audit eight input encoders -- spanning a shared-scalar baseline, per-channel linear projections, an orthogonality regulariser, a nonlinear MLP stem, block-partitioned concatenation, channel-independent and channel-as-token architectures, and a projected positional encoding -- on a synthetic benchmark designed to make channel identity informative and on ETTh1 as a real-data check, measured in next-step negative log-likelihood (NLL). The headline is one of practical near-equivalence within a wide "top tier": the standard per-channel linear projection (nn.Linear(C, $d_{\text{model}}$)) matches every alternative in that tier up to small, statistically real but practically modest, differences. Two encoders lose decisively: the shared-scalar baseline, which collapses for information-theoretic reasons we make explicit, and the channel-independent PatchTST-spirit baseline, which underperforms on both benchmarks and overfits universally on the synthetic one. Paired tests resolve two small gaps: projecting the sinusoidal positional encoding through a learned linear layer edges the rest at small $C$, with a direct geometric probe showing the mechanism is positional-channel orthogonalisation; a nonlinear MLP stem edges them at the largest $C$ we test, with the gap shrinking under more training data. The practical recommendation is to use nn.Linear(C, $d_{\text{model}}$) by default and reach for something more elaborate only when the task at hand gives a real reason to do so. Code and data to reproduce every experiment in this paper are available at https://github.com/OssiLehtinen/channel-encoder-audit
【3】Test-Time Compute Scaling for ASR with Depth-Conditioned Looped Transformers
标题：具有深度调节环路Transformer的ASB的测试时间计算缩放
链接：https://arxiv.org/abs/2606.04678
摘要：端到端ASR系统通常在推理时使用固定深度的声学编码器，这使得在不训练更大模型的情况下，很难用额外的测试时间计算来改善识别。一种自然的方法是循环地重用共享的Transformer块，但是我们发现朴素循环不能完全利用额外的循环计算。我们介绍了LARM，一个深度调节的循环Transformer，将循环编码器深度变成一个可控的测试时间计算轴。LARM结合了稀疏CTC检查点，监督时钟嵌入，电影深度调节和延迟软后验反馈。这些组件将循环结构化为由潜在细化阶段分隔的识别检查点，并允许共享权重在循环步骤中专门化。在LibriSpeech上，LARM随着推理循环数量的增加而提高了WER，并实现了与更深的非共享参数基线竞争的性能。我们的研究结果表明，测试时计算缩放可以扩展到连续非自回归语音识别的自回归语言模型推理。
摘要：End-to-end ASR systems typically use fixed-depth acoustic encoders at inference, making it difficult to trade additional test-time computation for improved recognition without training a larger model. A natural approach is to reuse a shared Transformer block recurrently, but we find that naive looping does not fully exploit additional recurrent compute. We introduce LARM, a depth-conditioned looped Transformer that turns recurrent encoder depth into a controllable test-time compute axis. LARM combines sparse CTC checkpoints, supervision-clock embeddings, FiLM depth conditioning, and delayed soft-posterior feedback. These components structure the loop into recognition checkpoints separated by latent refinement phases and allow shared weights to specialize across recurrent steps. On LibriSpeech, LARM improves WER as the number of inference loops increases and achieves performance competitive with deeper unshared-parameter baselines. Our results show that test-time compute scaling can extend beyond autoregressive language-model reasoning to continuous non-autoregressive speech recognition.
【4】ChessMimic: Per-Rating Transformer Models for Human Move, Clock, and Outcome Prediction in Online Blitz Chess
标题：ChessMimic：在线闪电战国际象棋中人类动作、时钟和结果预测的按评级Transformer模型
链接：https://arxiv.org/abs/2606.04473
摘要：
摘要：
【5】RowNet: A Memory Transformer for Tabular Regression
标题：RowNet：表格回归的内存Transformer
链接：https://arxiv.org/abs/2606.04445
备注：Retrieval-based neural architecture for real estate valuation. Related to TabR (arXiv:2307.14338) and retrieval-augmented tabular learning
摘要：
摘要：
【6】Low-Rank Decay for Grokking in Scale-Invariant Transformers: A Spectral-Geometric View
标题：比例不变Transformer中Grokking的低等级衰变：光谱-几何视图
链接：https://arxiv.org/abs/2606.04405
摘要：
摘要：
【7】MeshTok: Efficient Multi-Scale Tokenization for Scalable PDE Transformers
标题：MeshTok：可扩展的DTETransformer的高效多规模代币化
链接：https://arxiv.org/abs/2606.04366
备注：ICML2026
摘要：
摘要：
【8】Do Transformers Need Three Projections? Systematic Study of QKV Variants
标题：Transformer需要三个投影吗？QKV变种的系统研究
链接：https://arxiv.org/abs/2606.04032
备注：Accepted at ICML 2026 (PMLR vol. 306). 26 pages, 12 figures, 16 tables. Code: this https URL
摘要：
摘要：
GAN|对抗|攻击|生成相关(12篇)
【1】Generating Financial Time Series by Matching Random Convolutional Features
标题：通过匹配随机卷积特征生成金融时间序列
链接：https://arxiv.org/abs/2606.05138
摘要：
摘要：
【2】MusaCoder: Native GPU Kernel Generation with Full-Stack Training on Moore Threads GPU
标题：MusaCoder：在Moore Threads图形处理器上进行全栈训练的原生图形处理器内核生成
链接：https://arxiv.org/abs/2606.04847
摘要：原生GPU内核生成将高级张量程序转换为可执行的高效低级代码。现有的大型语言模型（LLM）难以完成这项任务，而基于执行的强化学习则面临着奖励稀疏、奖励黑客和训练不稳定的问题。我们介绍了MusaCoder，这是一个用于在CUDA和MUSA后端上生成本机GPU内核的全栈训练框架。MusaCoder通过MooreEval（一个分布式验证器和奖励环境）结合了渐进式面向内核的数据合成、多样性保持拒绝微调和执行反馈强化学习（RL）。为了稳定RL，MusaCoder引入了PrimeEcho用于第一回合锚定的多回合奖励，Buffered Dynamic Repeat用于从所有失败的硬样本中恢复信号，以及用于关闭策略序列过滤的Repeat Pop。在KernelBench和一个移植了MUSA的变体上的实验表明，MusaCoder在正确性和经验加速方面都优于强大的开源和专有基线，9B模型匹配或超过了前沿的闭源模型，27B模型建立了一个新的技术水平。这些结果不仅证明了全栈执行反馈训练对本机内核生成的有效性，而且Moore Threads GPU支持完整LLM后训练堆栈的能力，为新兴加速器上的大模型训练和优化提供了实用基础。
摘要
：Native GPU kernel generation turns high-level tensor programs into executable, efficient low-level code. Existing Large Language Models (LLMs) struggle with this task, while execution-based reinforcement learning suffers from sparse rewards, reward hacking, and training instability. We present MusaCoder, a full-stack training framework for native GPU kernel generation on CUDA and MUSA backends. MusaCoder combines progressive kernel-oriented data synthesis, diversity-preserving rejection fine-tuning, and execution-feedback Reinforcement Learning (RL) through MooreEval, a distributed verifier and reward environment. To stabilize RL, MusaCoder introduces PrimeEcho for first-turn-anchored multi-turn rewards, Buffered Dynamic Retry for recovering signals from all-failed hard samples, and MirrorPop for off-policy sequence filtering. Experiments on KernelBench and a MUSA-ported variant show that MusaCoder outperforms strong open-source and proprietary baselines in both correctness and empirical speedup, with the 9B model matching or exceeding frontier closed-source models and the 27B model establishing a new state of the art. These results demonstrate not only the effectiveness of full-stack execution-feedback training for native kernel generation, but also the capability of Moore Threads GPUs to support the complete LLM post-training stack, providing a practical foundation for large-model training and optimization on emerging accelerators.
【3】Scenario Generation for Risk-Aware Reinforcement Learning with Probably Approximately Safe Guarantees
标题：具有可能大致安全保证的风险意识强化学习的场景生成
链接：https://arxiv.org/abs/2606.04812
备注：8 pages, preprint
摘要：保证安全性对于在现实世界中部署强化学习（RL）代理至关重要，特别是使用深度RL学习的策略可能会对导致未知或不安全行为的过渡扰动表现出敏感性。策略验证的一种方法是通过对安全约束的策略轨迹进行采样来构建概率屏障证书，从而将已知的安全行为与未知的行为区分开来。如果策略易受转换不确定性或扰动的影响，使代理处于未充分探索的状态，则很难获得违反这些约束的概率的严格上限和下限。为了解决这个问题，我们使用变分自动编码器（VAE）来近似遇到的状态空间的分布，并使用状态的潜在特征构建上界和下界屏障证书，以高置信度优化已知安全行为的区域。我们在我们的工作框架作为一个双重优化问题的下限障碍证书提出了一个更保守的估计比上限障碍证书的安全区域。在训练过程中，采样状态位于两者的集合差内，即非鲁棒区域，使我们能够收紧上限和下限，以提供更清晰的安全概率保证。在我们的研究中，我们描述了放置的保证，并证明了我们的界限实验的紧密性。
摘要：Guaranteeing safety is critical to the deployment of reinforcement learning (RL) agents in the real-world, especially as policies learned using deep RL may demonstrate susceptibility to transition perturbations that result in unknown or unsafe behaviour. A method of policy verification is to construct probabilistic barrier-certificates by sampling policy trajectories with respect to safety constraints, thereby demarcating known safe behaviour from unknown behaviour. Obtaining tight upper and lower bounds on the probability of violation of these constraints may be difficult if the policy is susceptible to transition uncertainty or perturbation that places the agent in insufficiently explored states. To address this, we approximate the distribution of the encountered state-space using a variational autoencoder (VAE) and construct upper and lower-bound barrier-certificates using latent characteristics of states to optimize for regions of known, safe behaviour with high confidence. We frame this in our work as a dual optimization problem where the lower-bound barrier-certificate presents a more conservative estimate of the safe region than the upper-bound barrier-certificate. Sampling states that lie within the set difference of the two during training, i.e. the non-robust region, allows us to tighten the upper and lower bounds to provide sharper probabilistic guarantees on safety. Within our study, we describe the guarantees placed and demonstrate the tightness of our bounds experimentally.
【4】The Right Measure for Physics-Constrained Generation: A Co-Area Correction for Posterior-Consistent PDE Inverse Problems
标题：物理约束生成的正确衡量标准：后相容的PDL逆问题的共面积修正
链接：https://arxiv.org/abs/2606.04804
摘要：生成模型-扩散和流量匹配-越来越多地用于解决偏微分方程（PDE）逆问题，强制执行管理物理作为硬约束（通过投影或指导），并报告所得到的样本作为校准不确定性的贝叶斯后验。我们表明，这种广泛采用的配方样本错误的分布。在硬PDE约束条件下调节生成先验是在测量零流形上调节--一种本质上模糊的操作（Borel-Kolmogorov悖论），其物理正确的分辨率，小残留噪声限制，带有投影和基于制导的方法默默忽略的共区（Fixman）雅可比因子$[det（JJ^{\top}）]^{-1/2}$。我们使偏差精确，表明它的增长与约束灵敏度的异质性，并验证它对\n {i.i.d.}的控制问题。地面真相仲裁者省略的因子不是二阶细节：去除它会使后验误差膨胀到20倍于采样噪声基底;最小位移投影（如PCFM）偏向于最低价的9倍;我们引入了\textbf{CoCoS}，一个测量感知的约束采样器，目标是正确的共同区域后验，并表明它与采样噪声内的黄金标准后验相匹配。我们的研究结果意味着“满足物理学”并不等同于“对后验进行采样”，并对不确定性意识的科学推理进行了原则性的修正。
摘要：Generative models -- diffusion and flow matching -- are increasingly used to solve partial differential equation (PDE) inverse problems, enforcing the governing physics as a \emph{hard constraint} (via projection or guidance) and reporting the resulting samples as a Bayesian posterior with calibrated uncertainty. We show that this widely adopted recipe samples the wrong distribution. Conditioning a generative prior on a hard PDE constraint is conditioning on a measure-zero manifold -- an operation that is intrinsically ambiguous (the Borel--Kolmogorov paradox) and whose physically correct resolution, the small-residual-noise limit, carries a co-area (Fixman) Jacobian factor $[det(JJ^{\top})]^{-1/2}$ that projection- and guidance-based methods silently omit. We make the bias precise, show that it grows with the heterogeneity of the constraint sensitivity, and validate it on controlled problems against an \emph{i.i.d.} ground-truth arbiter. The omitted factor is not a second-order detail: removing it inflates the posterior error to $20\times$ the sampling-noise floor; minimal-displacement projection (as in PCFM) is biased at $9\times$ the floor; and a naive scalar reweighting does not fix it. We introduce \textbf{CoCoS}, a measure-aware constrained sampler that targets the correct co-area posterior, and show that it matches the gold-standard posterior to within sampling noise. Our results imply that ``satisfying the physics'' is not the same as ``sampling the posterior,'' and give a principled correction for uncertainty-aware scientific inference.
【5】Inference-Time Vulnerability Beyond Shallow Safety: Alignment Along Generation Trajectories
标题：超越浅层安全的推理时漏洞：沿着世代轨迹的一致
链接：https://arxiv.org/abs/2606.04778
摘要：安全对齐的大型语言模型（LLM）在推理过程中仍然容易受到干预，这些干预将生成重定向到有害输出。最近的工作将其归因于浅层安全性，其中对齐集中在前几个输出令牌中。我们表明，浅层安全是一个更广泛的推理时间漏洞的特殊情况下，在任何生成步骤的短令牌注入可以大大改变随后的安全行为。我们还发现，一个模型的对齐拒绝方向在其隐藏状态不预测其鲁棒性，这样的注入，揭示内部状态单独不确定扰动下的生成行为。为了解决这个问题，我们直接对齐模型的生成轨迹构建模拟中间序列扰动，并表明，这提高了鲁棒性的中间序列注入和推广到攻击，利用早期令牌生成。我们的工作认为，强大的安全调整需要对发电过程本身进行培训，而不仅仅是其输出。
摘要：Safety-aligned Large Language Models (LLMs) remain vulnerable to interventions during inference that redirect generation toward harmful outputs. Recent work attributes this to shallow safety, where alignment concentrates in the first few output tokens. We show that shallow safety is a special case of a broader inference-time vulnerability, in which short token injections at any generation step can substantially alter subsequent safety behavior. We also find that a model's alignment with refusal directions in its hidden states does not predict its robustness to such injection, revealing that internal state alone does not determine generation behavior under perturbation. To address this, we align models directly on generation trajectories constructed by simulating mid-sequence perturbation, and show that this improves robustness to mid-sequence injection and generalizes to attacks that exploit early-token generation. Our work argues that robust safety alignment requires training on the generation process itself, not only its outputs.
【6】Activation Steering of Video Generation Models via Reduced-Order Linear Optimal Control
标题：通过降阶线性最优控制对视频生成模型进行激活引导
链接：https://arxiv.org/abs/2606.04775
摘要：在大规模网络数据上训练的文本到视频（T2 V）模型可能会生成不受欢迎的内容，从而激励干预措施，在不牺牲视觉质量的情况下减少有害输出。激活转向提供了对微调和即时过滤的有吸引力的机械替代方案，但是现有的T2 V转向方法仍然有限，通常应用可能导致过度转向和内容降级的粗糙的、非预期的干预。为了缩小这一差距，我们提出了潜在激活线性二次调节器（LA-LQR），一个降阶的最优控制框架，微创T2 V转向。LA-LQR将T2 V推理公式化为一个动态系统，并计算闭环反馈干预，将激活转向所需的功能设定点，同时惩罚不必要的扰动。为了使高维视频激活的最优控制可行，我们将激活投影到一个低维的，任务相关的子空间上，该子空间来自对比提示对，估计该潜在空间中的局部线性动态，并解决潜在的LQR问题，以获得时间步长和层特定的转向信号。我们提供的理论界限相关的潜在的设定点跟踪到原始激活空间的功能控制，并根据经验验证的保真度减少潜在的动态。在概念转向和视频安全基准方面，LA-LQR相对于基线减少了不安全的生成，同时保持了即时保真度和视觉质量。
摘要
：Text-to-video (T2V) models trained on large-scale web data can generate undesired content, motivating interventions that reduce harmful outputs without sacrificing visual quality. Activation steering offers an attractive mechanistic alternative to finetuning and prompt filtering, but existing T2V steering methods remain limited, typically applying coarse, non-anticipative interventions that can lead to oversteering and content degradation. To close this gap, we propose Latent Activation Linear-Quadratic Regulator (LA-LQR), a reduced-order optimal control framework for minimally invasive T2V steering. LA-LQR formulates T2V inference as a dynamical system and computes closed-loop feedback interventions that steer activations toward desired feature setpoints while penalizing unnecessary perturbations. To make optimal control feasible for high-dimensional video activations, we project activations onto a low-dimensional, task-relevant subspace derived from contrastive prompt pairs, estimate local linear dynamics in this latent space, and solve a latent LQR problem to obtain timestep- and layer-specific steering signals. We provide theoretical bounds relating latent setpoint tracking to raw activation-space feature control, and empirically validate the fidelity of the reduced latent dynamics. On concept steering and video safety benchmarks, LA-LQR reduces unsafe generations relative to baselines, while preserving prompt fidelity and visual quality.
【7】Toward a Generalized Defense Across Sparse, Continuous, and Structured Parameter Attacks
标题：迈向跨稀疏、连续和结构化参数攻击的广义防御
链接：https://arxiv.org/abs/2606.04317
摘要：
摘要：
【8】Latent Anchor-Driven Test Generation for Deep Neural Networks
标题：深度神经网络的潜在锚点驱动测试生成
链接：https://arxiv.org/abs/2606.04310
摘要：
摘要：
【9】LazyAttention: Efficient Retrieval-Augmented Generation with Deferred Positional Encoding
标题：LazyAttention：具有延迟位置编码的高效检索增强生成
链接：https://arxiv.org/abs/2606.04302
备注：ICML 2026
摘要：
摘要：
【10】SymTRELLIS: Symmetry-Enforced Voxel Latents for 3D Generation
标题：SymDRELLIS：适用于3D生成的对称性增强体素潜伏期
链接：https://arxiv.org/abs/2606.04108
摘要：
摘要：
【11】Dive into the Scene: Breaking the Perceptual Bottleneck in Vision-Language Decision Making via Focus Plan Generation
标题：深入场景：通过焦点计划生成打破视觉语言决策中的感知瓶颈
链接：https://arxiv.org/abs/2606.04046
备注：Accepted at ICML 2026
摘要：
摘要：
【12】CaloTrilogy: Toward a Breakthrough in One-Step, End-to-End, Physics-Guided Shower Generation for Modern Calorimeters
标题：CaloTrilogy：迈向现代量热计一步、端到端、物理引导淋浴生成的突破
链接：https://arxiv.org/abs/2606.04165
摘要：
摘要：
半/弱/无/有监督|不确定性|主动学习(13篇)
【1】BBOmix: A Tabular Benchmark for Hyperparameter Optimization of Unsupervised Biological Representation Learning
标题：BBOmix：无监督生物表示学习超参数优化的表格基准
链接：https://arxiv.org/abs/2606.05139
摘要：
摘要：
【2】Activation-Based Active Learning for In-Context Learning: Challenges and Insights
标题：基于激活的主动学习用于情境学习：挑战和见解
链接：https://arxiv.org/abs/2606.05134
备注：9 pages, 3 figures
摘要：
摘要：
【3】Learning What Not to Impute: An Uncertainty-Aware Diffusion Framework for Meaningful Missingness
标题：学习什么不应受到影响：一个具有不确定性意识的有意义失踪的扩散框架
链接：https://arxiv.org/abs/2606.05073
摘要：
摘要：
【4】Uncertainty-Aware End-to-End Co-Design of Neural Network Processors: From Training and Mapping to Fabrication
标题：神经网络处理器的不确定性感知端到端协同设计：从训练和映射到制造
链接：https://arxiv.org/abs/2606.04850
备注：14 pages
摘要：设计神经网络处理器是一个端到端的协同设计问题：网络架构和训练预算决定了推理工作量;硬件映射决定了芯片面积、延迟和能量;这些特性决定了制造产量和制造成本。在实践中，这些决策是在不同的阶段做出的，现有的协同设计方法与特定的算法紧密耦合，因此很难在不重新设计整个管道的情况下改进一个组件。本文提出了一个统一的框架，在单调协同设计理论的基础上，由四个可互操作的设计模块跨越网络训练，芯片映射，晶圆级制造和计算资源分配。每个块只向系统的其余部分公开一个功能资源接口，因此可以细化任何块，而无需在其他地方进行结构更改。一个核心的贡献是不确定性的处理：而不是崩溃的随机结果到点估计，该框架引入了信心，成功概率的倒数，作为一个明确的和可优化的资源，成本，时间和功率。三个案例研究验证了这种方法。第一个恢复帕累托最优的实现跨异构应用程序的场景。第二个证实了置信度功能是一个连续可调的设计旋钮，而不是一个事后诊断。第三个演示了改进单个块的实现集自动传播到全局Pareto前沿，而无需修改协同设计图。
摘要：Designing a neural network processor is an end-to-end co-design problem: network architecture and training budget determine the inference workload; hardware mapping decisions determine chip area, latency, and energy; and these characteristics govern fabrication yield and manufacturing cost. In practice, these decisions are made in separate stages, and existing co-design methodologies are tightly coupled to specific algorithms, making it difficult to improve one component without reworking the entire pipeline. This paper presents a unified framework, grounded in monotone co-design theory, that composes four interoperable design blocks spanning network training, chip mapping, wafer-level fabrication, and compute resource allocation. Each block exposes only a functionality-resource interface to the rest of the system, so any block can be refined without structural changes elsewhere. A central contribution is the treatment of uncertainty: rather than collapsing stochastic outcomes into point estimates, the framework introduces Confidence, the inverse of success probability, as an explicit and optimizable resource alongside cost, time, and power. Three case studies validate the approach. The first recovers Pareto-optimal implementations across heterogeneous application scenarios. The second confirms that Confidence functions as a continuously tunable design knob rather than a post-hoc diagnostic. The third demonstrates that improving a single block's implementation set automatically propagates to the global Pareto front, without modifying the co-design diagram.
【5】Uncertainty-Aware (Un)Supervised Few-Shot User Adaptation for On-Device Personalized Human Activity Recognition
标题：不确定性感知（无）监督的Few-Shot用户适应，用于设备上个性化人类活动识别
链接：https://arxiv.org/abs/2606.04798
备注：6 pages, 4 figures, 2 tables, 2 algorithms
摘要：基于传感器的人类活动识别（HAR）模型通常会由于个体运动模式和传感器放置引起的域偏移而在看不见的用户上降级。因此，实际的可穿戴HAR系统需要轻量级的个性化方法，无论校准数据是标记的、未标记的还是不可用的，都是适用的，并且在有限的校准下是鲁棒的。我们提出了一个无梯度的框架，使用先前的原型将预训练的HAR分类器重新用作原型网络，保留了zero-shot性能并正则化自适应。对于标记标定，我们引入了封闭形式的贝叶斯原型估计，并将相同的原理扩展到无标记标定。每个活动只有3秒的校准数据（一次），监督自适应在四个数据集上将宏F1提高了+2.76到+33.44个百分点，而无监督自适应提高了+0.56到+32.13个百分点。由于自适应只需要封闭形式的原型更新，该框架使现有的HAR分类器的有效和强大的设备上的个性化。
摘要：Sensor-based Human Activity Recognition (HAR) models often degrade on unseen users due to domain shifts caused by individual movement patterns and sensor placement. Practical wearable HAR systems therefore require personalization methods that are lightweight, applicable whether calibration data is labeled, unlabeled, or unavailable, and robust under limited calibration. We present a gradient-free framework that repurposes pretrained HAR classifiers as Prototypical Networks using using prior prototypes, which preserve zero-shot performance and regularize adaptation. For labeled calibration, we introduce closed-form Bayesian prototype estimation and extend the same principle to unlabeled calibration. With only 3 seconds of calibration data per activity (one shot), supervised adaptation improves macro-F1 by +2.76 to +33.44 percentage points across four datasets, while unsupervised adaptation improves by +0.56 to +32.13 points. Since adaptation requires only closed-form prototype updates, the framework enables efficient and robust on-device personalization of preexisting HAR classifiers.
【6】Towards Accurate Model Selection in Deep Unsupervised Domain Adaptation
标题：深度无监督领域适应中的准确模型选择
链接：https://arxiv.org/abs/2606.04665
备注：upload to arxiv for record
摘要：
摘要：
【7】Distributional Approximate Nearest Neighbour Search for Uncertainty-Aware Retrieval
标题：分布式逼近最近邻搜索不确定性意识检索
链接：https://arxiv.org/abs/2606.04603
摘要：
摘要：
【8】Neetyabhas: A Framework for Uncertainty-Aware Public Policy Optimization in Rational Agent-Based Models
标题：Neetyabhas：基于理性主体的模型中具有不确定性的公共政策优化框架
链接：https://arxiv.org/abs/2606.04562
摘要：
摘要：
【9】GeoMin: Data-Efficient Semi-Supervised RLVR via Geometric Distribution Modeling
标题：GeoMin：通过几何分布建模的数据高效半监督的WLVR
链接：https://arxiv.org/abs/2606.04516
摘要：
摘要：
【10】Expectations vs. Realities: The Cost of MSE-Optimal Forecasting Under Conditional Uncertainty
标题：预期与现实：条件不确定性下MBE最优预测的成本
链接：https://arxiv.org/abs/2606.04342
备注：12 pages, Accepted for KDD 2026 Research track
摘要：
摘要：
【11】ADAPTOOD: Uncertainty-Aware Fine-Tuning for Out-of-Distribution ECG Time Series Models
标题：ADAPTOOD：不确定性意识的微调，用于超出分布的心电图时间序列模型
链接：https://arxiv.org/abs/2606.04164
备注：11 pages
摘要：
摘要：
【12】Stein Kernelized Molecular Dynamics for Active Learning of Interatomic Potentials
标题：Stein核心化分子动力学用于原子间势的主动学习
链接：https://arxiv.org/abs/2606.04100
摘要：
摘要：
【13】Bayes-Sufficient Representations in Supervised Learning
标题：监督学习中的贝足表示
链接：https://arxiv.org/abs/2606.04045
摘要：
摘要：
迁移|Zero/Few/One-Shot|自适应(4篇)
【1】U-Net-Accelerated Quality-Diversity Optimization for Climate-Adaptive Urban Layouts
标题：U-Net加速的气候适应性城市布局质量多样性优化
链接：https://arxiv.org/abs/2606.04658
摘要：
摘要：
【2】TITAN-FedAnil+: Trust-Based Adaptive Blockchain Federated Learning for Resource-Constrained Intelligent Enterprises
标题：泰坦-FedAnil+：用于资源受限智能企业的基于信任的自适应区块链联合学习
链接：https://arxiv.org/abs/2606.04388
备注：8 pages, 5 figures; code available at this https URL
摘要：
摘要：
【3】Adaptive Patching Is Harder Than It Looks For Time-Series Forecasting
标题：自适应修补比时间序列预测看起来更难
链接：https://arxiv.org/abs/2606.04074
摘要：
摘要：
【4】SpurAudio: A Benchmark for Studying Shortcut Learning in Few-Shot Audio Classification
标题：SpurAudio：一个用于研究Few-Shot音频分类的学习算法的基准
链接：https://arxiv.org/abs/2605.13672
摘要：
摘要：
强化学习(15篇)
【1】Reinforcement Learning from Rich Feedback with Distributional DAgger
标题：使用分布式DAger从丰富反馈中进行强化学习
链接：https://arxiv.org/abs/2606.05152
摘要：
摘要：
【2】Reproducing, Analyzing, and Detecting Reward Hacking in Rubric-Based Reinforcement Learning
标题：基于条目的强化学习中的复制、分析和检测奖励黑客行为
链接：https://arxiv.org/abs/2606.04923
备注：23 pages, 7 figures
摘要：基于规则的强化学习（RL）使用LLM作为判断者（LaaJ）根据规则对模型输出进行评分作为奖励。然而，政策模型可能会利用法官的潜在偏见，导致奖励黑客和无效或不安全的培训结果。在现实世界的基于规则的强化学习中，此类黑客行为通常很微妙，并且与多种判断偏见纠缠在一起，使其难以分析、检测和缓解。在本文中，我们介绍CHERRL，一个可控的黑客环境，基于规则的RL。通过将已知的偏差注入LaaJ，CHERRL能够稳定地复制奖励黑客行为，明确观察奖励分歧，并精确识别黑客攻击的开始。这为研究基于规则的RL中奖励黑客的机制和缓解措施提供了一个干净的实验测试平台。为了证明其实用性，我们分析了不同的判断偏见的角度的可重复性和可利用性，并探讨了一个基于代理的系统，自动检测奖励黑客攻击发作从训练日志。代码和环境可在https://github.com/THUAIS-Lab/CHERRL上公开获得。
摘要：Rubric-based reinforcement learning (RL) uses an LLM-as-a-Judge (LaaJ) to score model outputs according to rubrics as rewards. However, policy models may exploit latent biases in the judge, leading to reward hacking and ineffective or unsafe training outcomes. In real-world rubric-based RL, such hacking behaviors are often subtle and entangled with multiple judge biases, making them difficult to analyze, detect, and mitigate. In this paper, we introduce CHERRL, a controllable hacking environment for rubric-based RL. By injecting known biases into LaaJ, CHERRL enables stable reproduction of reward hacking, explicit observation of reward divergence, and precise identification of hacking onset. This provides a clean experimental testbed for studying the mechanisms and mitigations of reward hacking in rubric-based RL. To demonstrate its utility, we analyze different judge biases from the perspectives of discoverability and exploitability, and explore an agent-based system for automatically detecting reward hacking onset from training logs. The code and environment are publicly available at https://github.com/THUAIS-Lab/CHERRL.
【3】Fog of Love: Engineering Virtuous Agent Behavior with Affinity-based Reinforcement Learning in a Game Environment
标题：Fog of Love：Engineering Virtuous Agent Behavior with Affinity-Based Reinforcement Learning in a Game Environment（爱之雾：在游戏环境中使用基于亲和力的强化学习设计有道德的代理行为）
链接：https://arxiv.org/abs/2606.04750
摘要：在人工智能中灌输良性行为已经引起了越来越多的兴趣。其中一种技术被称为基于仿射的强化学习，它使用目标函数上的策略正则化来激励良性行为，而不完全依赖于奖励函数的设计。到目前为止，这种技术已被证明是有效的网格世界和玩具问题的环境中最小的状态和动作空间。为了将这项研究扩展到更复杂的环境中，我们引入了一个基于角色扮演棋盘游戏的两人多智能体环境，称为雾的爱。在这种环境中，两个代理人竞争以实现他们的个人美德，同时也合作以满足他们的关系。考虑到多智能体的性质，这是一个复杂的问题，其中多智能体深度确定性策略梯度智能体既不竞争也不成功合作。我们提出的证据表明，本地化的亲和力提高代理性能，在实现竞争和合作的目标，在这两个领域的优异的整体成绩。这不仅导致了道德选择，而且澄清了代理人的目的论，并使其行为在人类层面上可解释。
摘要：Instilling virtuous behavior in artificial intelligence has seen increasing interest. One of the techniques proposed is known as affinity-based reinforcement learning, which uses policy regularization on the objective function to incentivize virtuous actions without being fully dependent on the reward function design. Thus far, this technique has been demonstrated to be effective in grid worlds and toy-problem environments with minimal state and action spaces. To expand this research to more sophisticated environments, we introduce a two-player multi-agent environment based on the role-playing board game known as Fog of Love. In this environment, two agents compete to fulfill their individual virtues, while also cooperating to satisfy their relationship. Given the multi-agent nature, this is a complex problem where multi-agent deep deterministic policy gradient agents neither compete nor cooperate successfully. We present evidence that localized affinities enhance agent performance in achieving both competitive and cooperative objectives, resulting from superior overall scores in both domains. This not only results in virtuous choices but also clarifies an agent's teleology and makes its behavior human-level interpretable.
【4】COP-Q: Safety-First Reinforcement Learning for Robot Control via Cholesky-Ordered Projection
标题：COP-Q：通过Cholesky有序投影进行机器人控制的安全第一强化学习
链接：https://arxiv.org/abs/2606.04749
备注：7 pages, 6 figures, 2 tables
摘要：安全的机器人控制要求在满足安全约束的同时最大化回报。在非策略安全强化学习中，奖励和安全Q值通常由单独的批评者集合学习，每个目标的不确定性独立处理。这种客观的处理忽略了目标间的相关性，并可能导致过于保守的值估计，从而降低样本效率。为了解决这个问题，我们提出了Cholesky有序投影Q学习（COP-Q），这是一种安全第一的方法，将目标间协方差纳入向量值Q值估计。COP-Q在联合Q值空间中构造了一个广义置信界，并使用Cholesky分解以序列形式编码目标优先级。这保持了安全性的保守性，同时自适应地减少了奖励目标的过度保守性。所得到的估计用于时间差目标计算和演员优化。COP-Q的计算开销最小，并且与大多数现有的深度Q学习框架兼容。Brax中的机器人运动实验和Safety-Gymnasium中的安全导航实验（涵盖硬安全和软安全设置）表明，COP-Q实现了强大的安全性能，以及相对于代表性基线的竞争力或改进的样本效率。
摘要：Safe robot control requires maximizing return while satisfying safety constraints. In off-policy safe reinforcement learning, reward and safety Q-values are commonly learned by separate critic ensembles, with uncertainty handled independently for each objective. This objective-wise treatment neglects inter-objective correlation and can lead to overly conservative value estimates, thereby reducing sample efficiency. To address this issue, we propose Cholesky-Ordered Projection Q-learning (COP-Q), a safety-first method that incorporates inter-objective covariance into vector-valued Q-value estimation. COP-Q constructs a generalized confidence bound in the joint Q-value space and uses Cholesky factorization to encode objective priority in a sequential form. This preserves conservatism on safety while adaptively reducing excessive conservatism on the reward objective. The resulting estimate is used in both temporal-difference target computation and actor optimization. COP-Q incurs minimal computational overhead and is readily compatible with most existing deep Q-learning frameworks. Experiments on robot locomotion in Brax and safe navigation in Safety-Gymnasium, covering both hard- and soft-safety settings, demonstrate that COP-Q achieves strong safety performance together with competitive or improved sample efficiency relative to representative baselines.
【5】Trace-Mediated Peak Bias: Bridging Temporal Credit Assignment and Cognitive Heuristics in Deep Reinforcement Learning
标题：Trace-mediated Peak Bias：Bridging Temporal Credit Assignment and Cognitive Heuristics in Deep Reinforcement Learning追踪中介的峰值偏差：在深度强化学习中桥接时间信用分配和认知启发
链接：https://arxiv.org/abs/2606.04735
摘要：时间信用分配是生物和人工智能的核心，但它与非线性函数近似的相互作用知之甚少。我们确定了深度强化学习（RL）中的一种系统性故障模式，称为跟踪介导的峰值偏差（TMPB）。在中间资格跟踪深度，代理非理性地喜欢轨迹与高幅度的奖励“峰值”的替代品具有较高的累积回报。这提供了峰终法则的一个机械解释：人类记忆的偏见，经验是根据最强烈的时刻而不是综合效用来判断的。我们表明，TMPB的出现，因为痕迹放大远端时间差误差到“梯度冲击”，固定步长的随机梯度下降不能正常化，导致全球高估。相反，自适应优化器通过二阶矩归一化来减轻这种病态。我们的研究结果表明，人类一样的显着性失真可能会自然出现在分布式系统中的信用分配的数学约束，自适应优化是合理的价值估计的理论必要性。
摘要
：Temporal credit assignment is central to both biological and artificial intelligence, yet its interaction with non-linear function approximation is poorly understood. We identify a systematic failure mode in deep reinforcement learning (RL) termed Trace-Mediated Peak Bias (TMPB). At intermediate eligibility trace depths, agents irrationally prefer trajectories with high-magnitude reward ``peaks'' over alternatives with higher cumulative returns. This provides a mechanistic account of the Peak-End Rule: a human memory bias where experiences are judged by their most intense moments rather than integrated utility. We show that TMPB emerges because traces amplify distal Temporal Difference errors into ``gradient shocks'' that fixed-step-size Stochastic Gradient Descent cannot normalize, leading to global overestimation. Conversely, adaptive optimizers mitigate this pathology via second-moment normalization. Our results suggest that human-like saliency distortions may emerge naturally from the mathematical constraints of credit assignment in distributed systems, and that adaptive optimization is a theoretical necessity for rational value estimation.
【6】Explainably Safe Reinforcement Learning
标题：可以解释的安全强化学习
链接：https://arxiv.org/abs/2606.04634
摘要：
摘要：
【7】Dynamic Multi-Pair Trading Strategy in Cryptocurrency Markets with Deep Reinforcement Learning
标题：采用深度强化学习的加密货币市场中的动态多对交易策略
链接：https://arxiv.org/abs/2606.04574
备注：61 pages, 37 figures, 16 tables
摘要：
摘要：
【8】Episodic Memory Temporal Consistency for Cooperative Multi-Agent Reinforcement Learning
标题：协作多智能体强化学习的情景记忆时间一致性
链接：https://arxiv.org/abs/2606.04492
备注：Under Review
摘要：
摘要：
【9】AgentJet: A Flexible Swarm Training Framework for Agentic Reinforcement Learning
标题：AgentJet：一个用于强化学习的灵活群训练框架
链接：https://arxiv.org/abs/2606.04484
备注：Technical report, 27 pages
摘要：
摘要：
【10】From Ticks to Flows: Dynamics of Neural Reinforcement Learning in Continuous Environments
标题：从滴答到流动：连续环境中的神经强化学习动力学
链接：https://arxiv.org/abs/2606.04275
备注：Presented at ICLR 2026: this https URL
摘要：
摘要：
【11】Exact Unlearning in Reinforcement Learning
标题：强化学习中的精确去学习
链接：https://arxiv.org/abs/2606.04182
备注：ICML Spotlight
摘要：
摘要：
【12】Smart Transportation Without Neurons -- Fair Metro Network Expansion with Tabular Reinforcement Learning
标题：没有神经元的智能交通--利用表格强化学习进行公平的地铁网络扩展
链接：https://arxiv.org/abs/2606.04167
备注：16 pages
摘要：
摘要：
【13】RUBAS: Rubric-Based Reinforcement Learning for Agent Safety
标题：RUBAS：基于条目的强化学习用于代理安全
链接：https://arxiv.org/abs/2606.04051
摘要：
摘要：
【14】Position: Deployed Reinforcement Learning should be Continual
标题：位置：部署的强化学习应该是连续的
链接：https://arxiv.org/abs/2606.04029
备注：Accepted to the ICML 2026 Position Paper Track. See this https URL
摘要：
摘要：
【15】Assistax: A Multi-Agent Hardware-Accelerated Reinforcement Learning Benchmark for Assistive Robotics
标题：Assistax：辅助机器人的多智能体硬件加速强化学习基准
链接：https://arxiv.org/abs/2507.21638
备注：Accepted at the Reinforcement Learning Conference 2026
摘要：
摘要：
医学相关(4篇)
【1】Modeling and Interpreting Teamwork Dynamics in Cancer Care Outcome Prediction
标题：癌症护理结果预测中团队动力学的建模和解释
链接：https://arxiv.org/abs/2606.04499
摘要：
摘要：
【2】Radiomic Feature Selection Using Gradient Loss of Deep Neural Network for Lung Cancer Stage Detection
标题：利用深度神经网络梯度损失选择放射体特征用于肺癌分期检测
链接：https://arxiv.org/abs/2606.04453
摘要：
摘要：
【3】Early Detection of Alzheimer's Disease Using Explainable Machine Learning on Clinical Biomarkers: A Multi-Class Classification Study Using the Alzheimer's Disease Neuroimaging Initiative (ADNI) Dataset
标题：使用临床生物标志物的可解释机器学习早期检测阿尔茨海默病：使用阿尔茨海默病神经成像倡议（ADNI）数据集的多类别分类研究
链接：https://arxiv.org/abs/2606.03995
摘要：
摘要：
【4】SC-TauPath: A Structural Connectivity Attribution Framework for Mapping Tau Propagation Pathways in Alzheimer's Disease
标题：SC-TauPath：绘制阿尔茨海默病Tau传播途径的结构连接性归因框架
链接：https://arxiv.org/abs/2606.04066
摘要：
摘要：
蒸馏|知识提取(2篇)
【1】Invariant Gradient Alignment for Robust Reasoning Distillation
标题：稳健推理蒸馏的不变梯度对齐
链接：https://arxiv.org/abs/2606.05025
备注：30 Pages
摘要：大型语言模型（LLM）遭受捷径学习：即使逻辑结构相同，它们也会在语义表面与训练数据不同的分布外（OOD）输入上系统地失败。这破坏了将思维链推理转移给较小学生的知识蒸馏管道。我们引入了不变梯度对齐（IGA），一个训练框架，通过三个创新在语义多样但逻辑同构的例子中对齐梯度更新：（i）逻辑异构体集，在不同的语义域中共享相同逻辑结构的问题组（数学、医学、法律、科学）;（ii）可微的\n {连续梯度冲突掩码}，其抑制具有高跨域梯度方差的参数维度，同时保持不变方向;以及（iii）将掩蔽梯度截断SVD投影回到LoRA低秩流形上，始终保持参数效率。理论上，IGA产生更严格的OOD推广界比ERM，标度与异构体域的数量，并收敛于标准的SGD率下温和的规则性。从经验上讲，IGA在四个基准测试中的表现优于八个基线，准确性比ERM-SFT高出14.3个百分点，逻辑一致性得分为0.031比0.142 --代表性不变性提高了四倍。
摘要：Large language models (LLMs) suffer from shortcut learning: they systematically fail on out-of-distribution (OOD) inputs whose semantic surface differs from training data, even when the logical structure is identical. This undermines knowledge distillation pipelines that transfer chain-of-thought reasoning to smaller students. We introduce Invariant Gradient Alignment (IGA), a training framework that aligns gradient updates across semantically diverse but logically isomorphic examples via three innovations: (i) Logical Isomer Sets, groups of problems sharing identical logical structure across distinct semantic domains (mathematics, medicine, law, science); (ii) a differentiable \emph{Continuous Gradient Conflict Mask}, that suppresses parameter dimensions with high cross-domain gradient variance while preserving invariant directions; and (iii) a truncated SVD projection of the masked gradient back onto the LoRA low-rank manifold, maintaining parameter efficiency throughout. Theoretically, IGA yields tighter OOD generalization bounds than ERM, scaling with the number of isomer domains, and converges at the standard SGD rate under mild regularity. Empirically, IGA outperforms eight baselines across four benchmarks with accuracy gains up to 14.3 pp over ERM-SFT and a Logical Consistency Score of 0.031 versus 0.142 -- a fourfold improvement in representational invariance.
【2】Hyper-ICL: Attention Calibration with Hyperbolic Anchor Distillation for Multimodal In-Context Learning
标题：Hyper-ICL：使用双曲锚点蒸馏进行注意力校准，用于多模式内上下文学习
链接：https://arxiv.org/abs/2606.04434
备注：Accepted at the 43rd International Conference on Machine Learning (ICML 2026)
摘要：
摘要：
聚类(2篇)
【1】UniFair: A unified fair clustering approach based on separation and compactness
标题：UniFair：基于分离性和紧凑性的统一公平集群方法
链接：https://arxiv.org/abs/2606.04777
备注：17 pages, 6 Figures
摘要：聚类越来越多地用于支持高影响力的决策，但标准的目标，如$k$-means可以产生聚类，对待人口统计群体不平等。现有的公平聚类方法通常优化一个单一的公平性概念，往往忽略了聚类成本与几何形状的诱导决策边界的相互作用。我们提出\textsc{UniFair}，一个统一的框架，共同优化\textsc {分离公平}和\textsc {社会公平}。分离公平鼓励受保护的群体远离诱导决策边界，而社会公平通过惩罚群体聚类成本来减少群内失真的差异。我们开发了基于梯度的分离公平和统一的$k$-均值目标的优化程序，并通过在自动编码器的潜在空间中执行相同的标准将其扩展到深度聚类。在表格和图像数据集上的实验表明，\textsc{UniFair}减少了与边界相关的和基于成本的组差异，而仅适度增加了聚类损失。
摘要：Clustering is increasingly used to support high-impact decisions, yet standard objectives such as $k$-means can produce clusterings that treat demographic groups unequally. Existing fair clustering methods typically optimize a single notion of fairness and often overlook how clustering costs interact with the geometry of the induced decision boundaries. We propose \textsc{UniFair}, a unified framework that jointly optimizes \emph{separation fairness} and \emph{social fairness}. Separation fairness encourages protected groups to lie farther from the induced decision boundaries, while social fairness reduces disparities in within-cluster distortion by penalizing group-wise clustering costs. We develop gradient-based optimization procedures for separation-fair and unified $k$-means objectives, and extend them to deep clustering by enforcing the same criteria in the latent space of an autoencoder. Experiments on tabular and image datasets show that \textsc{UniFair} reduces both boundary-related and cost-based group disparities with only a modest increase in clustering loss.
【2】Contrastive Learning and Correlation Clustering for Sequences of Network Telescope Data
标题：网络望远镜数据序列的对比学习和相关聚集
链接：https://arxiv.org/abs/2606.04733
备注：Code: https://github.com/JannikPresberger/Contrastive_Learning_and_Correlation_Clustering_for_Sequences_of_Network_Telescope_Data
摘要：了解互联网扫描器的活动是具有挑战性的，它往往需要识别源之间的关系，语义注释是稀缺的任务。这项工作研究是否可以通过对比学习来估计网络流记录序列之间有语义意义的成对关系，而无需预训练和注释。为此，我们提出了一个Transformer模型，它嵌入了网络流记录的最小预处理序列，并使用对比学习对其进行训练。从这个模型中获得的相似性，我们状态的相关聚类问题，并解决它的本地。通过实验，我们发现：平均而言，来自相同来源的序列的学习相似性高于来自不同来源的序列，并且该属性推广到未知来源的未知序列。此外，相关聚类产生与扫描仪标签一致的聚类。完整的源代码的算法和再现的实验是公开的。
摘要：Understanding activities of Internet scanners is challenging; it often requires identifying relationships between sources, a task for which semantic annotations are scarce. This work investigates whether semantically meaningful pairwise relationships between sequences of network flow records can be estimated by contrastive learning, without pretraining and without annotations. To this end, we propose a transformer model that embeds minimally preprocessed sequences of network flow records and train it using contrastive learning. With the similarities obtained from this model, we state a correlation clustering problem and solve it locally. Experimentally, we show: Learned similarities are higher on average for sequences originating from the same source than for sequences originating from different sources, and this property generalizes to unseen sequences of unseen sources. Moreover, correlation clustering yields clusters consistent with scanner labels. The complete source code of the algorithms and for reproducing the experiments is publicly available.
自动驾驶|车辆|车道检测等(1篇)
【1】CADET: A Modular Platform for Evaluating Distributed Cooperative Autonomy in Connected Autonomous Vehicles
标题：CADET：评估互联自动驾驶汽车中分布式协作自主的模块化平台
链接：https://arxiv.org/abs/2606.04072
摘要：
摘要：
联邦学习|隐私保护|加密(1篇)
【1】Federated Learning for Multi-Center Sepsis Early Prediction with Privacy-Preserving
标题：保护隐私的联合学习多中心脓毒症早期预测
链接：https://arxiv.org/abs/2606.04338
摘要：
摘要：
推理|分析|理解|解释(8篇)
【1】Failed Reasoning Traces Tell You What Is Fixable (But Not by Reading Them)
标题：失败的推理痕迹告诉你什么是可以修复的（但不是通过阅读它们）
链接：https://arxiv.org/abs/2606.05145
摘要：
【2】In-Context Graphical Inference
标题：上下文图形推理
链接：https://arxiv.org/abs/2606.05042
备注：19 Pages
摘要：Marginal inference in discrete graphical models forces a choice between exactness and scalability: exact algorithms are intractable for high-treewidth graphs, while iterative approximations (Belief Propagation, variational methods) sacrifice convergence guarantees on frustrated topologies. We argue that this dichotomy stems from a mismatched inductive bias: iterative methods abandon the sequential elimination structure that makes exact inference correct. We introduce In-Context Graphical Inference (ICG-I), an autoregressive Graph Transformer that restores this structure by mimicking Variable Elimination with learned, Tensor- Train-compressed intermediate factors, paired with a Dirichlet output layer and Weighted Conformal Prediction for calibrated, distribution-free coverage guarantees under topological shift. We prove that TT compression errors propagate at most lincarly through the autoregressive chain, that the Dirichlet-Multinomial loss is a proper scoring rule, and that WCP maintains coverage with a quantifiable degradation under estimated density ratios. We conducted intensive experiments to evaluate ICG-I and achieved state-of-the-art performance across all benchmarks. ICG-I reduces MAE from 0.041 (best baseline) to 0.020 on standard instances and achieves 0.048 on N=500 frustrated spin glasses where BP diverges entirely.
【3】Enhancing the MADDPG Algorithm for Multi-Agent Learning via Action Inference and Importance Sampling
标题：通过动作推理和重要性抽样增强多智能体学习的MADDPG算法
链接：https://arxiv.org/abs/2606.05021
摘要：We investigate multi-agent deep reinforcement learning and propose two enhancements to the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm. First, we introduce a novel Action Inference mechanism that enables each agent to predict other agents' intended actions, thereby improving the accuracy and stability of its own policy. Second, we apply an importance sampling strategy, using geometric distribution, in the replay buffer to prioritize more recent and informative experiences, which helps mitigate the non-stationarity inherent in multi-agent environments. We evaluate both modifications on the discrete-action Predator-Prey task provided by the PettingZoo library, a flexible Python interface for general multi-agent reinforcement learning benchmarks. Our results indicate that Action Inference is effective in improving learning stability and inter-agent cooperation and that importance sampling using geometric distribution can lead to significant improvements in exploration efficiency over standard MADDPG. Code available at https://github.com/shaashwathsivakumar/MARL_Proj
【4】NLLog: Lightweight, Explainable SOC Anomaly Detection via Log-to-Language Rewriting
标题：NLLog：通过日志到语言重写进行轻量级、可解释的SOC异常检测
链接：https://arxiv.org/abs/2606.04957
备注：15 pages, 11 figures, 12 tables; submitted to ACSAC 2026
摘要：System-generated logs underpin security monitoring, yet their rigid template-based format hinders both automated analysis and human comprehension. We present NLLog (Natural-Language Log), a lightweight pipeline that deterministically rewrites parsed templates into WHO-WHAT-SEVERITY sentences, pools them with term-frequency-inverse-document-frequency weighting, classifies sessions with tree ensembles, and back-projects evidence with TreeSHAP for analyst review. On Hadoop Distributed File System (HDFS) and Blue Gene/L (BGL) corpora, NLLog exceeds two reproduced matched-protocol baselines; across HDFS, BGL, and the AIT Alert Data Set, it sustains low false-positive rates with commodity-hardware latency suitable for security operations center triage. Coverage, sparse-versus-dense, faithfulness, and adversarial ablations show that fallback sufficiency is corpus-dependent, that an enrollment-time coverage check can surface refinement requirements before deployment, and that an auditable deterministic rewrite combined with lightweight dense encoding provides a measurable representation layer for log-anomaly detection and triage.
【5】Smart Picks in the Dark: Towards Efficient RLVR for Reasoning via Tracing Metacognitive Pivots
标题：黑暗中的智能选择：通过跟踪元认知推测实现高效的RL VR推理
链接：https://arxiv.org/abs/2606.04503
摘要：
【6】Neural Galerkin Normalizing Flows for Bayesian Inference of Diffusions with Inaccessible Boundaries
标题：具有不可及边界扩散的Bayesian推理的神经Galerkin正规化流
链接：https://arxiv.org/abs/2606.04324
备注：27 pages, 12 figures
摘要：
【7】UltraEP: Unleash MoE Training and Inference on Rack-Scale Nodes with Near-Optimal Load Balancing
标题：UltraEP：在机架级节点上释放MoE训练和推理，实现近乎最优的负载平衡
链接：https://arxiv.org/abs/2606.04101
摘要：
【8】Counterfactual Explanations for Deep Two-Sample Testing
标题：深度双样本测试的反事实解释
链接：https://arxiv.org/abs/2606.04009
备注：17 pages
摘要：
检测相关(6篇)
【1】HYolo: An Intelligent IoT-Based Object Detection System Using Hypergraph Learning
标题：Hyolo：使用超图学习的基于物联网的智能对象检测系统
链接：https://arxiv.org/abs/2606.04345
备注
：8 pages, multiple figures;
摘要：
【2】DetectZoo: A Unified Toolkit for AI-Generated Content Detection Across Text, Audio, and Image Modalities
标题：DetectZoo：用于跨文本、音频和图像模式的人工智能生成内容检测的统一工具包
链接：https://arxiv.org/abs/2606.04205
摘要：
【3】Cross-Prompt Generalization in Detecting AI-Generated Fake News Using Interpretable Linguistic Features
标题：使用可解释语言特征检测人工智能生成的假新闻的交叉提示概括
链接：https://arxiv.org/abs/2606.04199
摘要：
【4】MimeLens: Position-Agnostic Content-Type Detection for Binary Fragments
标题：MimeLens：位置无关的二进制片段内容类型检测
链接：https://arxiv.org/abs/2606.04171
备注：18 pages, 2 figures, 15 tables. Models released on Hugging Face (this https URL reference training code at this https URL
摘要：
【5】veriFIRE: an Industrial Case Study in Verifying Consistency Properties for a DNN-Based Wildfire Detection System
标题：veriFIRE：基于DNN的野火检测系统的NPS一致性属性工业案例研究
链接：https://arxiv.org/abs/2606.04121
备注：To appear in The 9th International Symposium on AI Verification (SAIV)
摘要：
【6】TPA-AD: A Two-Stage Pseudo Anomaly-Guided Method for Bearing Time-Series Anomaly Detection
标题：TPA-AD：一种用于方位时间序列异常检测的两阶段伪异常引导方法
链接：https://arxiv.org/abs/2606.04073
摘要：
分类|识别(1篇)
【1】Representation Matters in Randomized Smoothing for Audio Classification
标题：音频分类随机平滑中的表示很重要
链接：https://arxiv.org/abs/2606.04210
摘要：
表征(3篇)
【1】RePercENT: Scaling Disentangled Representation Learning Beyond Two Modalities
标题：RePercent：超越两种模式扩展解开的表示学习
链接：https://arxiv.org/abs/2606.05109
摘要：
【2】The Loss Is Not Enough: Sampling Conditions and Inductive Bias in Contrastive Representation Learning
标题：损失还不够：对比表示学习中的抽样条件和归纳偏差
链接：https://arxiv.org/abs/2606.04280
摘要：
【3】PointAction: 3D Points as Universal Action Representations for Robot Control
标题：PointAction：3D点作为机器人控制的通用动作表示
链接：https://arxiv.org/abs/2606.03943
备注：Project page: this https URL
摘要：
优化|敛散性(11篇)
【1】Provably Reduced Sample Cost in Prior-Guided Hyperparameter Optimization
标题：先验引导超参数优化中明显降低的样本成本
链接：https://arxiv.org/abs/2606.04866
摘要：Large-scale hyperparameter optimization (HPO) in automated machine learning (AutoML) consumes substantial computational resources, raising growing concerns about scalability and energy efficiency. Existing methods use prior information heuristically to accelerate both black-box and multi-fidelity settings, but they lack a characterization of how prior informativeness quantitatively reduces sample complexity. In this work, we provide the first distribution-dependent sample complexity bounds for multi-fidelity HPO with priors through the formal lens of fixed-budget best-arm identification. By modeling priors directly over arm means as configuration performance, we derive explicit, distribution-dependent error bounds that quantify the relationship between priors and evaluation budget. Our analysis shows that informative priors, which concentrate probability mass on near-optimal arms, yield reductions in the number of required evaluations, whereas baseline performance is recovered with uninformative or misleading priors. We conduct proof-of-concept experiments on a synthetic benchmark and on LCBench, a common multi-fidelity HPO benchmark for deep learning, to confirm our theoretical results, achieving up to 90% budget reduction while retaining solution quality. Together, our results provide a principled foundation for prior-guided and compute-efficient green AutoML.
【2】BiasGRPO: Stabilizing Bias Mitigation in High-Variance Reward Landscapes via Group-Relative Policy Optimization
标题：BiasGRPO：通过群体相对政策优化稳定高方差回报景观中的偏差缓解
链接：https://arxiv.org/abs/2606.04807
备注：Accepted to Findings of the ACL
摘要
：Mitigating social bias in Large Language Models (LLMs) presents a distinct alignment challenge: unlike verifiable tasks, bias lacks a single ground truth, creating a high-variance, subjective reward landscape. Previous preference-based fine-tuning methods have major trade-offs: Direct Preference Optimization (DPO) is limited by the lack of exploration inherent in offline training, while Proximal Policy Optimization (PPO) can lead to training instability due to potentially unreliable critic estimates. In this paper, we propose BiasGRPO, a framework using Group Relative Policy Optimization (GRPO) to stabilize alignment by normalizing rewards across a group of sampled completions. By substituting the value function with a group-relative baseline, our approach reduces instability while maintaining the exploration benefits of online training. We find that BiasGRPO outperforms DPO and PPO across multiple benchmarks, indicating its effectiveness. To adapt GRPO, we synthetically extend a dataset spanning multiple domains and contexts. We also create and release a custom bias reward model that effectively guides generation while being highly compute-efficient and avoiding knowledge degradation, providing a valuable resource that can be seamlessly integrated into multi-objective RLHF pipelines.
【3】Cone-Compatible Monge Geometry for High-Dimensional Ordered Optimal Transport
标题：用于多维有序最优传输的圆锥兼容Monge几何
链接：https://arxiv.org/abs/2606.04695
备注：13 pages, 2 figures, including appendices
摘要：High-dimensional optimal transport is seldom available in closed form. The one-dimensional case is exceptional because the order of the real line is compatible with convex transport costs, making monotone rearrangement optimal. This paper studies when an analogous Monge structure can be recovered in higher dimensions from a partial order. We introduce a cone-compatible Monge geometry: a closed convex cone (K) induces the order (x\preceq_K y) whenever (y-x\in K), and is compatible with a cost if ordered pairs satisfy a Monge exchange inequality. For squared Mahalanobis costs (c_M(x,y)=(x-y)^\top M(x-y)), we prove a sharp characterization: compatibility holds exactly when (K) is acute under the (M)-inner product, namely (u^\top Mv\ge0) for all (u,v\in K), equivalently (K\subseteq K_M^*). Under this condition, measures supported on cone chains admit a quantile-type closed-form optimal coupling, yielding exact transport under the original ground cost rather than after projection or metric replacement. We distinguish the resulting cone-chain Wasserstein metric on canonically ordered chain distributions from an extended directed cone transport cost on general measures, and develop feasibility, duality, stability, approximation, Gaussian recovery, statistical, and computational results. The theory is complementary to sliced and tree Wasserstein distances: it is not a universal fast surrogate, but a way to obtain interpretable, direction-valid, original-space monotone transport for ordered high-dimensional data.
【4】ParetoPilot: Zero-Surrogate Offline Multi-Objective Optimization via Infer-Perturb-Guide Diffusion
标题：ParetoPilot：通过下扰动引导扩散进行零替代离线多目标优化
链接：https://arxiv.org/abs/2606.04468
摘要：
【5】An Ensembled Latent Factor Model via Differential Evolution and Gradient Descent Optimization
标题：基于差异进化和梯度下降优化的集成潜在因素模型
链接：https://arxiv.org/abs/2606.04408
摘要：
【6】TANDEM: Bi-Level Data Mixture Optimization with Twin Networks
标题：TANEM：使用Twin Networks的双层数据混合优化
链接：https://arxiv.org/abs/2606.04401
摘要：
【7】Literature-Guided Minimax Optimization of Virtual Epilepsy Neurostimulation
标题：文献引导的虚拟癫痫神经刺激的Minimax优化
链接：https://arxiv.org/abs/2606.04339
备注：9 pages, 4 figures. Code and interactive essay at this https URL
摘要：
【8】Optimal Transport Flow Matching by Design
标题：通过设计优化交通流量匹配
链接：https://arxiv.org/abs/2606.04092
备注：Project page: this https URL
摘要：
【9】Beyond Static Priors: Dynamic Neural Guidance for Large-Scale Ant Colony Optimization
标题：超越静态先验：大规模蚁群优化的动态神经指导
链接：https://arxiv.org/abs/2606.04039
备注：Accepted at KDD 2026
摘要：
【10】Inverse Critical Experiment Design via Gradient Optimization and a Multigroup Attention-Based Neural Network Architecture
标题：通过梯度优化和基于多群注意力的神经网络架构的逆临界实验设计
链接：https://arxiv.org/abs/2606.04033
摘要：
【11】Near-Optimal Decentralized Stochastic Convex Optimization over Networks
标题：网络上的近优分散随机凸优化
链接：https://arxiv.org/abs/2606.04757
备注：12 papers
摘要：
预测|估计(10篇)
【1】RIDE: An Open Dataset and Benchmark for Train Delay Prediction
标题：RIDE：火车延误预测的开放数据集和基准
链接：https://arxiv.org/abs/2606.05070
备注：58 pages, 41 figures
摘要：
【2】New Benchmarking Shows Limited Generalization Power of TCR Antigenic Epitope Prediction Models
标题：新基准显示TLR抗原表位预测模型的推广能力有限
链接：https://arxiv.org/abs/2606.04994
备注：6 pages, 1 figure. Preprint version
摘要
：Accurate computational prediction of T cell receptor (TCR) antigen specificity would transform the study of T cell biology and enable scalable immune engineering, yet existing models lack sufficient sensitivity and specificity for broad applications. A major limitation is the absence of rigorously defined, unseen benchmark datasets that allow unbiased evaluation of model performance and generalizability. Here, we describe two complementary classes of datasets that meet this criterion and argue that they provide both a robust framework for model assessment and a foundation for next-generation TCR-antigen prediction algorithm development.
【3】Prediction Under Imperfect Compression: A Theory of Approximate MDL
标题：不完全压缩下的预测：近似MDL理论
链接：https://arxiv.org/abs/2606.04834
备注：26 pages
摘要：Minimum Description Length (MDL) formalizes the principle of Occam's razor by optimizing the total description length: $L(\mathrm{model})+L(\mathrm{data} \ | \ \mathrm{model})$. For sequential prediction, the MDL method repeatedly selects a model with a minimum objective score of the observed prefix for the next step prediction. Classical MDL prediction theory shows that exact optimization of the MDL objective indeed provides a strong compression guarantee that supports reliable prediction. However, practical machine learning usually can only find models by approximately optimizing the objective function. To bridge this gap, this paper addresses the following fundamental question: Under what forms of approximation and regularization does approximate MDL still guarantee reliable sequential prediction? This work offers a principled characterization. We prove that for any approximation with additive slack $C$ of the more general form of the balanced MDL objective: $λ\cdot L(\mathrm{model})+L(\mathrm{data} \ | \ \mathrm{model})$, the cumulative expected squared prediction error is finite for all $λ\ge1$. The case $λ>1$ is proved by an affinity-telescoping argument, while the boundary case $λ=1$ is proved by a likelihood-ratio stopping argument based on exact static MDL bounds. Our results establish that classical MDL regularization remains robust to any fixed additive optimization error. Furthermore, we establish that our characterization of the approximate MDL framework is sharp: When $00$, under multiplicative approximation, and thus, additive approximation is both sufficient and essential.
【4】Signed Dual Attention: Capturing Signed Dependencies in Time Series Forecasting
标题：签署的双重注意力：捕捉时间序列预测中的签署倾向
链接：https://arxiv.org/abs/2606.04833
备注：5 pages, 3 figures, accepted at AAAI 2026 AI4TS Workshop
摘要：Initially developed for natural language processing, Transformer architectures and attention mechanisms are now central to a wide range of deep learning models, including applications in time series forecasting. A standard attention mechanism, however, implicitly assumes homophilic interactions, limiting its ability to model data with positive and negative dependencies, such as time series. In this work, we introduce the Signed Dual Attention, a novel attention formulation that captures both positive and negative relational patterns without additional parameters. By leveraging a dual message-passing scheme inspired by correlation structures, Signed Dual Attention propagates both supportive and contrastive information within a single shared block, effectively achieving the expressiveness of two head attention without additional parameters. This module can be seamlessly integrated into existing architectures and can yield performance gains in certain situations, requiring signed relational modeling. This approach opens a pathway toward more expressive and parameter-efficient transformers.
【5】Metric-Aware Hybrid Forecasting for the CTF4Science Lorenz Challenge
标题：CTF 4 Science Lorenz挑战赛的指标感知混合预测
链接：https://arxiv.org/abs/2606.04191
摘要：
【6】When Offline Selectors Cannot Beat the Best Single Model: A Diagnostic Study on edX Dropout Prediction
标题：当离线选择器无法击败最佳单一模型时：edX辍学预测的诊断研究
链接：https://arxiv.org/abs/2606.04161
摘要：
【7】Physics-Informed Machine Learning for Short-Term Flood Prediction
标题：用于短期洪水预测的物理知识机器学习
链接：https://arxiv.org/abs/2606.04143
备注：This paper has been accepted for publication in IGARSS 2026. The final authenticated version will be available through IEEE Xplore
摘要：
【8】Stationarity-Aware Retrieval-Augmented Time Series Forecasting
标题：平稳性感知检索增强时间序列预测
链接：https://arxiv.org/abs/2606.04135
备注：Accepted by KDD 2026 research track
摘要：
【9】EpiFormer: Learning Antigen-Antibody Interactions for Epitope Prediction via Geometric Deep Learning
标题：EpiFormer：通过几何深度学习学习抗原-抗体相互作用以预测表位
链接：https://arxiv.org/abs/2606.04154
摘要：
【10】SpliceBind: Isoform-Aware Prediction of Binding Pocket Druggability
标题：SpliceBind：结合口袋可药用性的同工同感知预测
链接：https://arxiv.org/abs/2606.04020
备注：10 pages, 4 figures, ACM-BCB 2026 Main Conference Short Paper
摘要：
其他神经网络|深度学习|模型|建模(51篇)
【1】Deep Embedded Multiplicative DMD for Algebra-Preserving Koopman Learning
标题：用于保代数Koopman学习的深度嵌入式乘式DMZ
链接：https://arxiv.org/abs/2606.05131
备注：26 pages, 11 figures
摘要：
【2】Preserving Data Privacy in Learning Causal Structure with Fully Homomorphic Encryption
标题：用全同形加密学习因果结构时保护数据隐私
链接：https://arxiv.org/abs/2606.05129
摘要：
【3】FoeGlass: Simple In-Context Learning Is Enough for Red Teaming Audio Deepfake Detectors
标题：FoeGlass：简单的上下文学习足以满足Red团队音频Deepfake检测器的需求
链接：https://arxiv.org/abs/2606.05101
备注：Accepted at ICML 2026
摘要：
【4】AutoLab: Can Frontier Models Solve Long-Horizon Auto Research and Engineering Tasks?
标题：AutoLab：前沿模型能否解决长期汽车研究和工程任务？
链接：https://arxiv.org/abs/2606.05080
备注：Code: this https URL ; Website: this https URL
摘要：
【5】Validity Threats for Foundation Model Research
标题：基础模型研究的有效性威胁
链接：https://arxiv.org/abs/2606.05029
摘要：Controlled experiments are the backbone of machine learning research, but at the scale of modern foundation models, they have become prohibitively expensive. Instead, the community increasingly relies on research strategies that approximate the ideal experiment at a fraction of the cost: proxy experiments and scaling laws, observational studies with publicly available models, and single-run designs that leverage variation within individual training runs. In this work, we argue that there is no free lunch when approximating large-scale experiments on a compute budget. Specifically, savings in compute come at the cost of validity threats -- hidden and sometimes untestable assumptions that, when violated, can invalidate research claims. To help navigate such threats, we propose an evaluation framework that casts foundation model research as a causal inference problem. Within this framework, we evaluate different research strategies through four types of validity adapted from the empirical social sciences -- statistical, internal, external, and construct validity. We find that each strategy comes with a characteristic validity profile: proxy experiments trade external and construct validity for statistical and internal validity; observational studies face confounding and effect heterogeneity; and single-run designs are strained by interference between treated units. This analysis reveals several validity threats that have received insufficient attention in the literature. Overall, our evaluation framework provides researchers with a practical toolkit for scrutinizing validity threats in foundation model research~designs.
【6】Be Fair! Can Machine Learning Engineering Agents Adhere to Fairness Constraints?
标题：公平点！机器学习工程代理能否遵守公平约束？
链接：https://arxiv.org/abs/2606.04971
摘要：Machine learning engineering (MLE) agents promise to automate end-to-end ML pipeline development from raw data and natural language instructions, potentially making ML accessible to non-technical domain experts. However, in sensitive and regulated domains, this abstraction creates a responsibility gap: end-users may lack visibility into design choices that affect correctness, robustness, fairness, and regulatory compliance. We argue that existing benchmarks are insufficient to assess whether MLE agents can be safely applied in such settings. We propose desiderata for a responsibility-centered evaluation framework and conduct an exploratory study on melanoma classification, focusing on fairness across skin tones as a responsibility constraint. When evaluating two recent MLE agents, we find that agent-generated pipelines show high variance and consistently underperform manually designed baselines in both predictive quality and fairness, despite fairness-oriented prompts. These preliminary results suggest that further research is needed towards redesigning MLE agents to allow humans to guide the search process and reliably assess the compliance and quality of the generated ML pipelines.
【7】AdaKoop: Efficient Modeling of Nonlinear Dynamics from Nonstationary Data Streams with Koopman Operator Regression
标题：AdaKoop：基于Koopman算子回归的非平稳数据流非线性动力学高效建模
链接：https://arxiv.org/abs/2606.04930
备注：Accepted by KDD'26
摘要：Real-time data analysis requires the ability to accurately and adaptively address nonlinear dynamics in a nonstationary data stream while preserving computational efficiency. However, nonlinear dynamics are so complex that capturing dynamically changing nonlinear patterns and utilizing them for downstream tasks under strict time constraints is nontrivial. To bridge the gap between nonlinear complexity and computational tractability, this study applies Koopman operator theory, which states that nonlinear dynamics can be represented as linear transitions in an infinite-dimensional space. Building upon finite-dimensional approximations of this operator, we present AdaKoop, an efficient streaming algorithm for modeling nonlinear dynamics over nonstationary data streams. Our approach utilizes a probabilistic framework grounded in Koopman operator theory, treating both raw observations and reproducing kernel Hilbert space (RKHS) features as emissions from latent vectors. This dual-view formulation allows nonlinear dynamics to be expressed as a tractable linear system. Therefore, AdaKoop enables the efficient and stable modeling of nonlinear dynamics in a streaming fashion, avoiding the prohibitive computational costs of iterative nonlinear optimization. Furthermore, to address nonstationarity in data streams, AdaKoop adaptively detects the switching of patterns via statistical hypothesis testing for abrupt pattern shifts and incrementally updates model parameters to handle continuous changes. Extensive experiments on a total of 71 practical benchmark datasets across various domains demonstrate that AdaKoop outperforms state-of-the-art methods in terms of real-time forecasting accuracy and computational efficiency.
【8】Worker Utility as Hysteresis: A Preisach Model of Transaction Acceptance in Gig Labour Markets
标题：工人效用作为滞后：零工劳动力市场交易接受的Preisach模型
链接：https://arxiv.org/abs/2606.04916
备注：18 pages, 5 figures
摘要
：Worker utility is not observed -- only its consequence is. Each gig transaction produces a single bit: accepted or rejected. We argue this structure points directly to the Preisach hysteresis model as the natural representation of latent worker preferences. The Preisach operator models aggregate output as an integral over a population of binary threshold elements -- precisely the structure that emerges when heterogeneous workers each carry a private acceptance wage. We estimate two latent utility surfaces: acceptance utility U_1(X) and rejection utility U_0(X), via a dual-output neural network (shared layers 256->128, margin loss enforcing U_1 >= U_0). Classification reduces to the Preisach gap U_1(X) - U_0(X), passed into an XGBoost classifier alongside clip-stabilised price-to-threshold encodings. On 36,891 gig transactions, this pipeline achieves Jaccard = 0.827 and ROC AUC = 0.799. The price-to-threshold encoding accounts for +11.0 pp AUC over raw utility features. The model confirms the directional asymmetry hysteresis predicts: price decreases depress completion rates more than equivalent increases raise them. Applied to the full dataset, the model's recommendations simultaneously reduce the total wage bill by 21.3% and increase expected fill rate by 9.7 pp. For 74.2% of transactions, P(accept) already exceeds 0.80; reducing the wage keeps it above threshold (mean post-cut P = 0.972), releasing cost savings (median 31%). For the remaining 25.4%, a median 7% wage increase recovers +43 pp acceptance. A model without an explicit indifference zone cannot execute both moves simultaneously.
【9】Learning Empirically Admissible Neural Heuristics for Combinatorial Search
标题：学习经验容许的组合搜索神经启发式算法
链接：https://arxiv.org/abs/2606.04860
备注：13 pages, 3 figures, 2 tables, 1 algorithm
摘要：Finding optimal solution paths for combinatorial puzzles like the Rubik's Cube, sliding tile puzzles, and Lights Out remains a classical challenge in artificial intelligence. Heuristic search algorithms, such as A* , guarantee path optimality only when using an admissible heuristic-one that never overestimates the true remaining cost-to-go. Deep reinforcement learning (RL) methods like DeepCubeA train deep neural networks to approximate cost-to-go heuristics. However, standard mean-squared error (MSE) training regularly yields overestimations, violating admissibility and compromising solution optimality. In this paper, we introduce a generalizable framework for learning validation-calibrated admissible neural heuristics. We train a value network using an underestimating Admissible Bellman Operator combined with an Asymmetric Loss function to penalize overestimation. To account for residual neural function approximation errors, we propose a post-hoc calibration safety offset computed over validation scrambles. We demonstrate that our calibrated neural heuristics achieve no observed admissibility violations under the evaluation protocol and preserve path optimality in practice while reducing search node expansions by up to 83.0% on a 2 by 2 Rubik's Cube, 19.9% on a 3 by 3 Lights Out grid, and 1.9% on an 8-Puzzle compared to standard analytical baselines.
【10】Rethinking Incompleteness: Formalizing Protocol Divergence and Train-Once Learning for Robust IMVC
标题：重新思考不完整性：将协议分歧和一次训练学习形式化以实现稳健的IMVC
链接：https://arxiv.org/abs/2606.04857
摘要：Standard IMVC evaluation retrains separate models for different missing-data configurations. We show that this paradigm obscures a fundamental vulnerability: missing rate alone is insufficient to characterize data incompleteness. Specifically, we show that protocols with identical nominal missing rates can differ by up to $50\times$ in their proportion of fully observed samples, inducing drastically different learning regimes. We formalize this phenomenon as incompleteness divergence, providing measures that capture structural disparities across missing-data protocols. We further prove that for a broad class of reconstruction-based objectives, learning becomes structurally ill-posed when the proportion of complete samples falls below a critical threshold, leading to near-random performance. To bypass this theoretical bound, we propose CRAFT (Complete-data Robust Attention-masked Fusion Transformer). CRAFT shifts the burden of robustness from the loss function to the architecture via two key properties: (i) per-sample independence, which removes reliance on complete-sample co-occurrence, and (ii) mask-aware variable-length fusion, which aggregates only observed views through attention masking. This design allows a single model, trained once on complete data, to generalize to diverse missing patterns at inference time without retraining. Extensive experiments on seven benchmarks show that CRAFT matches or outperforms per-configuration baselines while reducing training overhead by $8.8\times$, demonstrating that robustness to missing data can be achieved as an inherent architectural property. Code (CRAFT) and our imvc-audit toolkit are available at https://anonymous.4open.science/r/CRAFT-BF80/ and https://anonymous.4open.science/r/imvc-audit-8263/.
【11】Reconciling Causality and Non-Equilibrium Thermodynamics with Hamiltonian Causal Models
标题：用汉密尔顿因果模型研究因果关系和非平衡热力学
链接：https://arxiv.org/abs/2606.04822
摘要：Causal modeling of physical temporal phenomena must handle interventions that act along trajectories, nonstationary induced laws, path-dependent effects, and feedback mediated by dynamics, all challenging in standard causal models. We introduce Hamiltonian Causal Models (HCMs), a trajectory-level framework in which observed variables interact with local environments and interventions act as controls of Hamiltonian mechanisms. HCMs separate immutable equations of motion from intervenable mechanisms and define causal effects as discrepancies between interventional path laws. A key motivation for HCMs is their natural interface with non-equilibrium thermodynamics. Entropy production quantifies the irreversibility of a process and is a central causal observable: it is estimable from data and witnesses causal effects along the system's evolution that are invisible to endpoint and cumulative versions of the standard average treatment effect. As in physics, cause and effect are not primitives of the relation between two random variables but arise from the non-invertibility of the thermodynamic arrow. With this, our paper reconciles the language of statistical causal models and non-stationary thermodynamics, offering new tools to describe causality in a wide range of physical systems.
【12】Learning While Acting: A Skill-Enhanced Test-Time Co-Evolution Framework for Online Lifelong Learning Agents
标题：边学边做：一个在线终身学习代理的技能增强测试时间协同进化框架
链接：https://arxiv.org/abs/2606.04815
摘要：Lifelong learning is essential for Large Language Model (LLM) agents operating in dynamic, interactive environments. However, existing lifelong learning agents for long-horizon tasks typically depend on discrete skill or past experiences retrieval with static parameters during inference, which prevents them from continuously internalizing test-time feedback like human learners. To bridge this gap, we propose Skill-enhanced Test-Time Co-Evolution (\texttt{LifeSkill}), a two-stage reinforcement learning framework for Online Lifelong Learning Agents. Specifically, we design Verifier-Guided Skill Learning that addresses the lack of direct supervision for skill extraction by rewarding candidate skills according to the average verifier success of multiple skill-conditioned policy rollouts, encouraging the model to generate skills that are useful for solving tasks rather than merely plausible in text. Furthermore, we introduce Online Skill Internalization, which continuously improves the policy model during test-time interaction by transforming skill-conditioned trajectories into reward signals. This enables the agent to directly internalize reasoning capabilities into its parameters, avoiding the context bloat of experience retrieval. Experiments on LifelongAgentBench show that LifeSkill improves average performance by 7 absolute points by comparing with existing lifelong agent baselines.
【13】Measuring Model Robustness via Fisher Information: Spectral Bounds, Theoretical Guarantees, and Practical Algorithms
标题：通过Fisher信息测量模型稳健性：谱界限、理论保证和实际算法
链接：https://arxiv.org/abs/2606.04767
备注：35 pages, 1 figure
摘要：The robustness of deep neural networks is crucial for safety-critical deployments, yet existing evaluation methods are often attack-dependent and lack interpretability. We propose a principled, attack-agnostic robustness metric based on the spectral norm of the Fisher Information Matrix (FIM), which quantifies the worst-case sensitivity of the model's output distribution to input perturbations. Theoretically, we establish that the FIM equals the variance of the input Jacobian and derive closed-form spectral bounds for common architectures, including VGG, ResNet, DenseNet, and Transformer, providing the first theoretical robustness ranking. To enable scalable evaluation, we develop efficient algorithms, including power iteration and Hutchinson-based estimation, that support both white-box and black-box settings. Extensive experiments across multiple datasets, including CIFAR, ImageNet, and medical images, and across multiple architectures show a strong correlation between our metric and adversarial vulnerability. Our framework serves as an interpretable diagnostic tool that complements attack-based evaluations, offering insights into architectural sensitivity and guiding the design of more robust models. Code is available at: https://github.com/franz-chang/SRP/.
【14】Curvature-aware dynamic precision approach for physics-informed neural networks
标题：物理信息神经网络的曲线感知动态精度方法
链接：https://arxiv.org/abs/2606.04736
摘要：Physics-informed neural networks (PINNs) have become a promising framework for simulating partial differential equations (PDEs) by embedding physical laws directly into neural network training. However, recent studies show that PINN optimisation is sensitive to numerical precision. Existing implementations commonly use either single precision (FP32), which is computationally efficient but prone to failure modes, or double precision (FP64), which is robust but substantially expensive. This creates a trade-off between computational efficiency and numerical accuracy. To reduce the computational cost of double-precision training while retaining prediction accuracy, we propose a curvature-aware precision controller that adapts numerical precision during training rather than treating it as a fixed implementation choice. The proposed method reuses curvature information derived from the limited-memory BFGS (L-BFGS) optimiser to construct a precision controller, retaining FP32 when lower precision is sufficient and promoting computation to FP64 when the training dynamics indicate numerical sensitivity or precision-limited stagnation. We evaluate the proposed approach on four canonical PINN failure-mode benchmarks and an irradiance-driven ordinary differential equation example. We further test the proposed approach across different neural network architectures. The method consistently matches or even slightly exceeds full FP64 solution accuracy while reducing training time relative to full double-precision training on all benchmark equations. The obtained results indicate that precision sensitivity in PINN optimisation is phase-dependent, and that selectively applying higher precision only during numerically critical stages can lower computational cost without sacrificing predictive accuracy.
【15】Learning symplectic model reduction based on a approximation theorem of symplectic embeddings
标题：基于辛嵌入逼近定理学习辛模型约简
链接：https://arxiv.org/abs/2606.04623
摘要：
【16】HalfNet: Randomized Neural Networks with Learned Subspace Geometry
标题：HalfNet：具有习得子空间几何的随机神经网络
链接：https://arxiv.org/abs/2606.04583
备注：6 pages (+2 pages of appendix), 6 figures
摘要：
【17】SurvPFN: Towards Foundation Models for Survival Predictions
标题：SurvPFN：生存预测的基础模型
链接：https://arxiv.org/abs/2606.04564
备注：10 pages, 1 figure. Accepted to "Foundation Models for Structured Data" Workshop at the International Conference on Machine Learning (ICML) 2026
摘要：
【18】GENEB: Why Genomic Models Are Hard to Compare
标题：GENEB：为什么基因组模型难以比较
链接：https://arxiv.org/abs/2606.04525
摘要：
【19】LimiX-2M: Mitigating Low-Rank Collapse and Attention Bottlenecks in Tabular Foundation Models
标题：LimiX-2 M：缓解表格基础模型中的低级崩溃和注意力瓶颈
链接：https://arxiv.org/abs/2606.04485
备注：Accepted by ICML 2026
摘要：
【20】When Both Layers Learn: Training Dynamics of Representing Linear Models via ReLU Networks
标题：当两层都学习时：通过ReLU网络训练表示线性模型的动态
链接：https://arxiv.org/abs/2606.04476
备注：47 pages, 8 figures, published at the 39th Annual Conference on Learning Theory (COLT), 2026
摘要：
【21】D^2SD: Accelerating Speculative Decoding with Dual Diffusion Draft Models
标题：D ' 2SD：用双重扩散草案模型加速推测解码
链接：https://arxiv.org/abs/2606.04446
摘要：
【22】LoopMoE: Unifying Iterative Computation with Mixture-of-Experts for Language Modeling
标题：LoopMoE：将迭代计算与专家混合统一起来进行语言建模
链接：https://arxiv.org/abs/2606.04438
摘要：
【23】The price of multi-group transductive learning
标题：多群体转换学习的代价
链接：https://arxiv.org/abs/2606.04423
摘要：
【24】An Empirical Study of Data Scale, Model Complexity, and Input Modalities in Visual Generalization
标题：视觉概括中的数据规模、模型复杂性和输入方式的实证研究
链接：https://arxiv.org/abs/2606.04409
备注：12 pages, 9 figures, 4 tables
摘要：
【25】DPDL: Towards Differential Privacy Preservation in Decentralized Stochastic Learning on Non-IID Data
标题：DPDL：在非IID数据的分散随机学习中实现差分隐私保护
链接：https://arxiv.org/abs/2606.04399
摘要：
【26】Physics-Informed Neural Network Modeling of Biodegradable Contaminant Transport through GCL/SL Composite Liners
标题：生物可降解污染物通过GCL/SL复合衬垫传输的物理信息神经网络建模
链接：https://arxiv.org/abs/2606.04392
摘要：
【27】Shortcomings and capacities of real-constrained neural networks in complex spaces
标题：复杂空间中实约束神经网络的缺点和容量
链接：https://arxiv.org/abs/2606.04390
备注：First version
摘要：
【28】Spatially Grounded Concept Bottleneck Models via Part-Factorized Attention
标题：通过部分因素化注意力的空间立足概念瓶颈模型
链接：https://arxiv.org/abs/2606.04364
摘要：
【29】A Geometric Characterization of the Stationary Plateau for Two-Layer Neural Networks
标题：两层神经网络平稳平台的几何特征
链接：https://arxiv.org/abs/2606.04327
备注：47 pages
摘要：
【30】Measuring What Matters: Synthetic Benchmarks for Concept Bottleneck Models
标题：衡量重要的事情：概念瓶颈模型的综合基准
链接：https://arxiv.org/abs/2606.04326
备注：Benchmarks available at this https URL
摘要：
【31】OpenRFM: Dissecting Relational In-Context Learning
标题：OpenRFM：剖析关系性上下文学习
链接：https://arxiv.org/abs/2606.04320
备注：25 pages, including appendix
摘要：
【32】Testing Neural Networks via Bayesian-Guided Exploration of Decision Landscapes
标题：通过Bayesian引导的决策景观探索测试神经网络
链接：https://arxiv.org/abs/2606.04314
摘要：
【33】Folded Transport MCMC: Certifiable Quotient Posterior Computation for Symmetric Bayesian Models
标题：折叠传输MCMC：对称Bayesian模型的可证明商后验计算
链接：https://arxiv.org/abs/2606.04307
备注：48 pages (including supplementary material), 5 figures, 6 tables. Submitted to Journal of the Royal Statistical Society: Series B
摘要：
【34】Offline-to-Online Learning in Linear Bandits
标题：线性盗贼的线下到线上学习
链接：https://arxiv.org/abs/2606.04305
摘要：
【35】Efficient and Training-Free Single-Image Diffusion Models
标题：高效且免训练的单图像扩散模型
链接：https://arxiv.org/abs/2606.04299
备注：CVPR 2026; Project Page: this https URL
摘要：
【36】PE-MHL: Physics-Encoded Modular Hybrid Layers for Scalable Learning of Complex Systems
标题：PE-MHL：用于复杂系统可扩展学习的物理编码模块化混合层
链接：https://arxiv.org/abs/2606.04290
摘要：
【37】Sparse Mixture-of-Experts Reward Models Learn Interpretable and Specialized Experts for Personalized Preference Modeling
标题：稀疏专家混合奖励模型为个性化偏好建模学习可解释和专业的专家
链接：https://arxiv.org/abs/2606.04284
摘要：
【38】Derivative Informed Learning of Exchange-Correlation Functionals
标题：交换相关函数的派生知情学习
链接：https://arxiv.org/abs/2606.04279
备注：Proceedings of the 43rd International Conference on Machine Learning
摘要：
【39】Long-Term and Short-Term Transistor Aging in Deep Neural Networks: Impact and Mitigation
标题：深度神经网络中的长期和短期晶体管老化：影响和缓解
链接：https://arxiv.org/abs/2606.04266
备注：28 pages, 16 figures
摘要：
【40】Edge of Stability Selectively Shapes Learning Across the Data Distribution
标题：跨数据分布的稳定边选择性学习
链接：https://arxiv.org/abs/2606.04212
备注：27 pages, 22 figures, ICML HiLD 2026
摘要：
【41】Unlocking Feature Learning in Gated Delta Networks at Scale
标题：大规模解锁门控三角洲网络中的特征学习
链接：https://arxiv.org/abs/2606.04048
摘要：
【42】Novel Aspects of IEEE SA P3109 Arithmetic Formats for Machine Learning
标题：IEEE SA P3109机器学习算术算法的新方面
链接：https://arxiv.org/abs/2606.04028
摘要：
【43】How do machines learn? Evaluating the AIcon2abs method
标题：机器如何学习？评估AIcon 2abs方法
链接：https://arxiv.org/abs/2401.07386
备注：textual review (spelling and grammar); reorganization of the elements of some figures; New references included
摘要：
【44】Learning Control-Affine Reduced-Order Models via Autoencoders
标题：通过自动编码器学习控制仿射降阶模型
链接：https://arxiv.org/abs/2606.05045
摘要：
【45】Bayesian learning for the stochastic shortest path problem
标题：随机最短路径问题的Bayesian学习
链接：https://arxiv.org/abs/2606.04845
备注：50 pages, 19 figures
摘要：
【46】ReSGA: A Large Tail Risk Model for Learning Value-at-Risk and Expected Shortfall
标题：ReLGA：学习风险价值和预期短缺的大尾部风险模型
链接：https://arxiv.org/abs/2606.04576
摘要：
【47】Scaling Datasets for Multi-Sensor, Multi-Agent, and Multi-Domain Learning in Autonomous Systems
标题：扩展自治系统中多传感器、多代理和多域学习的数据集
链接：https://arxiv.org/abs/2606.04444
摘要：
【48】Flatness and Generalization: Learning Multi-Index Models with Homogeneous Neural Networks
标题：平坦性和泛化：使用同质神经网络学习多指标模型
链接：https://arxiv.org/abs/2606.04429
摘要：
【49】Knockoffs-based False Discovery Rate Control and Simplification for Deep Neural Networks
标题：深度神经网络基于模仿的错误发现率控制和简化
链接：https://arxiv.org/abs/2606.04404
摘要：
【50】REGAIN: REconciliation GAIN-driven Auxiliary Direction Learning
标题：REGAIN：再调解增益驱动的辅助方向学习
链接：https://arxiv.org/abs/2606.04380
摘要：
【51】SPLIT-PINN: Separable Probability Learning Technique via Physics-Informed Neural Networks for High-Dimensional Probabilistic Modeling
标题：SplIT-PINN：通过物理信息神经网络的可分离概率学习技术，用于多维概率建模
链接：https://arxiv.org/abs/2606.04000
摘要：
其他(49篇)
【1】STRIDE: Training Data Attribution via Sparse Recovery from Subset Perturbations
标题：WRIDE：通过子集扰动的稀疏恢复训练数据归因
链接：https://arxiv.org/abs/2606.05165
备注：project page: this https URL
摘要：
【2】Geometry Gaussians: Decoupling Appearance and Geometry in Gaussian Splatting
标题：几何高斯：高斯飞溅中的外观与几何脱钩
链接：https://arxiv.org/abs/2606.05124
摘要：
【3】Identifying Gems from Roman RAPIDly
标题：识别Roman RAPIDly的宝石
链接：https://arxiv.org/abs/2606.05103
备注：15 pages, 10 figures, Submitted to the Publications of the Astronomical Society of the Pacific
摘要：
【4】Fast & Faithful Function Vectors
标题：快速且忠实的功能载体
链接：https://arxiv.org/abs/2606.05079
摘要：
【5】AlphaQ: Calibration-Free Bit Allocation for Mixture-of-Experts Quantization
标题：AlphaQ：专家混合量化的免校准位分配
链接：https://arxiv.org/abs/2606.04980
备注：28 pages, 11 figures
摘要：Mixture-of-Experts (MoE) architectures scale model capacity through sparse expert activation, but their deployment remains memory-bound because all expert weights must reside in memory. Mixed-precision quantization can substantially reduce this footprint by assigning different bit-widths to different experts. Existing approaches, however, typically rely on calibration data to estimate expert importance and determine bit allocation. For frontier MoE LLMs, the original training data, and hence the true training distribution, is proprietary and inaccessible. As a result, calibration sets are inevitably imperfect surrogates, and this can misestimate expert utilization and lead to suboptimal bit allocation. Motivated by the substantial cross-expert quality variability observed in modern MoE models, and by the success of Heavy-Tailed Self-Regularization (HT-SR) theory at predicting neural network model quality without access to training or testing data, we propose AlphaQ, a calibration-free bit-allocation method for MoE quantization. AlphaQ draws on HT-SR theory and follows a simple principle: experts with more heavy-tailed weight spectra are typically better trained and hence should receive higher bit-widths, while experts with weaker heavy-tailed structure can be quantized more aggressively. AlphaQ operationalizes this principle by measuring expert-wise spectral heavy-tailedness and solving a budget-constrained optimization problem that minimizes total quantization error under a global bit-budget constraint. Across several MoE models, AlphaQ consistently outperforms calibration-based baselines under matched bit budgets. Notably, on Qwen1.5-MoE, AlphaQ achieves near full-precision accuracy with an average expert precision of only 3.5 bits, while delivering more than 4$\times$ memory compression. Our code is available at https://github.com/Superone77/AlphaQ.
【6】A General Framework for Dynamic Consistent Submodular Maximization
标题：动态一致子模极大化的一般框架
链接：https://arxiv.org/abs/2606.04946
备注：Accepted at ICML 2026
摘要：Consistency is an important property in dynamic submodular maximization and entails maintaining a near-optimal solution at all times, making only a small number of adjustments to the solution in each step. Prior work has explored this question for the insertion-only case, where the algorithm faces a stream of $n$ insertions, and has established lower and upper bounds for the cardinality-constrained version of the problem. We consider this question in the fully dynamic setting, where the stream of operations may contain both insertions and deletions. We develop a general framework for designing algorithms for this setting, and instantiate it to obtain the first constant-factor approximations with sublinear consistency. For cardinality constraints, we propose a $\frac 12 - O(\varepsilon)$ approximation that is $O\left(\frac{1}{\varepsilon^2}\right)$ consistent. For rank-$k$ matroid constraints, we construct a $\frac 14 - O(\varepsilon)$ approximation to the dynamic optimum that is $O\left(\frac{\log k}{\varepsilon^2}\right)$ consistent.
【7】Mean-based algorithms: A lower bound and regret
标题：基于均值的算法：下限和遗憾
链接：https://arxiv.org/abs/2606.04931
摘要：Mean-based algorithms are a class of online learning algorithms that assign low probability to actions with low average rewards. Recent work indicates these algorithms converge favorably to serially undominated actions, which approximate Nash equilibria in economic games. However, empirical studies also show slower convergence compared to established algorithms in bandit-feedback scenarios. We study mean-based algorithms when the time horizon is unknown and only bandit feedback is available. In this setting, we provide the first lower bound on the algorithm-defining sequence $γ_t$ that formally establishes a limit on how fast these algorithms can learn. Additionally, we propose two mean-based algorithms: one generalizes $ε$-greedy, and the other extends the mean-based Exp3 to unknown horizons. Our experiments show that mean-based algorithms, although slightly slower, can perform competitively with other bandit-feedback algorithms. We further analyze the relationship to no-regret algorithms. Depending on the choice of $γ_t$, the intersection with no-regret algorithms is non-trivial, and we show that algorithms exist that are both mean-based and no-regret. This adds context to the "exploitability" of this class of algorithms that previous contributions suggest.
【8】Toward Multi-Domain and Long-Tailed Quantization via Feature Alignment and Scaling
标题：通过特征对齐和缩放实现多域和长尾量化
链接：https://arxiv.org/abs/2606.04920
摘要
：Quantizing deep neural networks is essential for efficient inference on resource-constrained devices. However, most existing methods are designed for single-domain and class-balanced data, leaving practical settings with domain shifts or severe class imbalance underexplored. We address these challenges with Efficient Multi-Domain Alignment Quantization (EmaQ), which aligns domain distributions through a CDF-based projection and uses sensitivity-aware weight aggregation to stabilize multi-domain quantization. We further extend EmaQ to EmaQ-LT for long-tailed quantization by introducing class-conditioned variance scaling and confidence-based logit adjustment to mitigate majority-class overconfidence. Theoretical analyses establish convergence guarantees and motivate the proposed sensitivity and scaling mechanisms. Experiments on standard, multi-domain (Office-31, Digits), and long-tailed (SynDigits-LT, CIFAR-10-LT, CIFAR-100-LT) benchmarks show that EmaQ and EmaQ-LT achieve strong low-bit performance under domain shift and class imbalance.
【9】Towards Pretraining Text Encoders for TabPFN
标题：迈向TabPFN的预训练文本编码器
链接：https://arxiv.org/abs/2606.04876
摘要：Tabular foundation models, such as TabPFN, achieve strong performance on tabular datasets with numerical and categorical data, but do not natively handle high-cardinality text features. Standard pipelines, therefore, embed text with a language model and compress the resulting vectors with PCA into a small number of scalar features before inputting them into TabPFN. This creates an information bottleneck: most embedding dimensions are discarded, and the compressed representation must then be expanded again by TabPFN's feature encoder. End-to-end alternatives can avoid PCA, but they require large amounts of pretraining data containing text cells and usually perform subpar compared to tabular foundation models that were pretrained on large amounts of synthetic data. Inspired by modality-alignment approaches like LLaVA (vision-to-LLM token projection) and TableGPT-style systems (table-to-LLM token projection), we introduce the TabPFN Text Adapter (text-to-TFM token projection). We freeze both the sentence encoder and TabPFN, and train only a lightweight adapter that maps text embeddings into a short sequence of tokens in TabPFN's embedding space. This design removes the PCA bottleneck, preserves TabPFN's numerical strengths, and is more efficient to train than end-to-end text-tabular pipelines.
【10】OA-CutMix: Correcting the Label Bias of CutMix
标题：OA-CutMix：纠正CutMix的标签偏差
链接：https://arxiv.org/abs/2606.04820
摘要：CutMix has become the de facto standard mixing augmentation, yet its label assignment rests on a flawed assumption: The area of the pasted patch faithfully reflects its semantic contribution to the mixed image. In practice, however, patches frequently land on background regions, assigning label credit to classes whose objects are not visible. The mean discrepancy of the CutMix label and the semantic object area is $21.5\%$. In $17\%$ of samples an image contributes zero visible object pixels yet receives nonzero label weight. We propose Object-Aware CutMix (OA-CutMix), which corrects this bias by replacing the area-based CutMix weight with one derived from precomputed segmentation masks, assigning labels in proportion to the visible object area each image contributes to the mix. The image mixing procedure is left entirely unchanged. We evaluate OA-CutMix against 10+ static and dynamic mixing methods across 4 architectures and 6 datasets. OA-CutMix consistently achieves the highest accuracy over all tasks, outperforming even dynamic mixing methods, but at a fraction of the training-time cost. Improvements are largest for small objects, where the label bias from CutMix is greatest. Thus, correcting the label is sufficient to match or exceed the performance of methods modifying the image mixing algorithm.
【11】Crafting Your Evolving Dreams: Concept-Incremental Versatile Customization
标题：打造您不断发展的梦想：概念增量多功能定制
链接：https://arxiv.org/abs/2606.04797
备注：Accepted to Transactions on Pattern Analysis and Machine Intelligence (TPAMI)
摘要：Custom diffusion models (CDMs) have garnered significant interest owing to their remarkable capacity for generating personalized concepts. However, the majority of CDMs unrealistically presume that the user's collection of personalized concepts is static and incapable of incremental growth over time. Furthermore, they exhibit significant catastrophic forgetting and concept neglect of previously learned concepts when incrementally learning a sequence of new ones. To resolve the above challenges, we develop a novel Continually Customizable Diffusion Model (CCDM), enabling users to perform concept-incremental versatile customization. Specifically, we design an attribute-decoupled LoRA (AD-LoRA) module and a relevance-guided AD-LoRA aggregation strategy to mitigate catastrophic forgetting. They can preserve concept-specific attributes of each task and leverage beneficial inter-task correlations to enhance the continual learning of new customization tasks. Additionally, to address the challenge of concept neglect, we propose a controllable regional context synthesis strategy that performs multi-concept composition in alignment with user-provided conditions. This strategy enhances the overall consistency in multi-concept synthesis by guaranteeing semantic independence between user-defined regions and their smooth boundary transitions. Experiments show our CCDM exhibits significant improvements over baseline methods.
【12】Beyond Structural Symmetries: Linear Mode Connectivity via Neuron Identifiability
标题：超越结构对称性：通过神经元可识别性实现线性模式连接
链接：https://arxiv.org/abs/2606.04754
备注：Accepted at ICML 2026
摘要：Many striking phenomena in deep learning, such as linear mode connectivity and the structured behavior of training dynamics, are closely tied to parameter symmetries: transformations that leave the realized function unchanged. Despite growing attention to parameter symmetries, the exact interplay between parameters, data, and representations remains underexplored. To investigate this, we develop a theoretical framework of effective function classes, i.e., the set of functions a neuron can realize on its input support, and the norm cost of realizing them. We then formalize effective symmetry breaking via neuron identifiability across independent training runs. Our analysis shows that neural networks can admit large families of approximately equivalent solutions even in structurally asymmetric models. We further show that neuron identifiability enables representation merging without prior alignment, and characterize when such merging admits a linear low-loss path. These findings highlight the role of effective function classes in affecting the loss landscape.
【13】TIDE: Proactive Multi-Problem Discovery via Template-Guided Iteration
标题：TIDE：通过模板引导迭代主动发现多问题
链接：https://arxiv.org/abs/2606.04743
摘要
：Agents are widely deployed as assistants over documents, tools, and code. However, they typically act only on explicit user requests, which surface only the problems the user has noticed, while many other important problems coexist, hidden in plain sight, within the broader user context, with their total number unknown in advance. We frame this as the task of discovering multiple hidden problems from context, in which coexisting problems should be uncovered, grounded in supporting evidence, and paired with concrete actions. To this end, we introduce TIDE, a template-guided iterative framework with two complementary mechanisms. Specifically, motivated by the observation that single-pass prediction anchors on the most salient cases and yields generic claims, we propose iterative discovery, which surfaces a small batch of candidates per round while conditioning on what has already been found, so subsequent rounds extend coverage; and thought templates, reusable schemas distilled from previously solved cases that specify what contextual signals to attend to and how to connect them, anchoring each prediction in a recognizable problem class. We validate TIDE on two realistic settings, personal workspaces and software repositories, across four model backbones, showing substantial gains over single-shot and parallel multi-agent baselines on task coverage, identification, and resolution.
【14】Fitting scattered data with optional monotonicity constraints on GPU: LipFit package
标题：在图形处理器上用可选的单调性约束来匹配分散数据：LipFit包
链接：https://arxiv.org/abs/2606.04670
摘要：This paper presents a method of multivariate scattered data interpolation and approximation that produces optimal Lipschitz-continuous approximation, subject to the desired monotonicity constraints. This method relies on tight upper and lower approximations to the data, and is similar in its spirit to the nearest-neighbour approximation but does not suffer from discontinuities. Local Lipschitz interpolation and Lipschitz smoothing are also presented. This approach falls under the umbrella of instance-based approximation with no training phase, and it is suitable for GPU-based parallelisation. A Python GPU-friendly package LipFit which implements the methods discussed is discussed.
【15】Why Muon Outperforms Adam: A Curvature Perspective
标题：为什么Muon优于Adam：弯曲角度
链接：https://arxiv.org/abs/2606.04662
摘要：
【16】CRAFT: Cost-aware Refinement And Front-aware Tuning of Prompts
标题：CRAFT：成本意识的优化和预算提前调整
链接：https://arxiv.org/abs/2606.04661
摘要：
【17】Rollout-Level Advantage-Prioritized Experience Replay for GRPO
标题：GRPO的推出级别优先级体验重演
链接：https://arxiv.org/abs/2606.04560
摘要：
【18】Cartridges at Scale: Training Modular KV Caches over Large Document Collections
标题：大规模卡片：在大型文档收藏中训练模块化KV缓存
链接：https://arxiv.org/abs/2606.04557
备注：21 pages, 5 figures, 17 tables
摘要：
【19】ANN Search: Recall What Matters
标题：NN搜索：回忆重要的事情
链接：https://arxiv.org/abs/2606.04522
摘要：
【20】CyberGym-E2E: Scalable Real-World Benchmark for AI Agents' End-to-End Cybersecurity Capabilities
标题：CyberGym-E2 E：AI代理端到端网络安全能力的可扩展现实基准
链接：https://arxiv.org/abs/2606.04460
备注：ICML 2026
摘要：
【21】On Out-of-sample Embedding in UMAP
标题：UMAP中的样本外嵌入
链接：https://arxiv.org/abs/2606.04451
备注：22 pages, 16 figures
摘要：
【22】Trivium: Temporal Regret as a First-Class Objective for Causal-Memory Controllers
标题：Trivium：作为一级目标的时间后悔-记忆控制器
链接：https://arxiv.org/abs/2606.04421
备注：62 pages, 12 tables, 12 figures
摘要：
【23】Loss-Conditional PINNs for Parametric PDE Families
标题：参数化PDL系列的损失条件PINN
链接：https://arxiv.org/abs/2606.04420
摘要：
【24】(Mis)generalization of Helpful-only Fine-tuning
标题：（不）仅有帮助的微调的概括
链接：https://arxiv.org/abs/2606.04413
备注：77 pages, 50 figures
摘要：
【25】Revisiting Privacy Amplification by Subsampling in Selective Release DPSGD
标题：再论选择性释放DPSGD中的子采样隐私放大
链接：https://arxiv.org/abs/2606.04384
摘要：
【26】When Do Fewer Coordinates Suffice in DP-SGD?
标题：DP-Singapore中什么时候更少的坐标就足够了？
链接：https://arxiv.org/abs/2606.04375
备注：14 pages
摘要：
【27】Policy Gradient for Continuous-Time Robust Markov Decision Processes
标题：连续时间鲁棒Markov决策过程的政策梯度
链接
：https://arxiv.org/abs/2606.04335
摘要：
【28】Can Generalist Agents Automate Data Curation?
标题：多面手可以自动化数据保存吗？
链接：https://arxiv.org/abs/2606.04261
备注：Preprint
摘要：
【29】VAMPS: Visual-Assisted Mathematical Problem Solving Benchmark
标题：VAMPS：视觉辅助数学问题解决基准
链接：https://arxiv.org/abs/2606.04244
摘要：
【30】A Geometric View of Counterfactual Behavior: Interaction of Boundary Proximity and Local Support
标题：反事实行为的几何观：边界接近性和局部支持的相互作用
链接：https://arxiv.org/abs/2606.04209
摘要：
【31】Training-Free Lexical-Dense Fusion for Conversational-Memory Retrieval
标题：用于对话记忆检索的免训练词汇密集融合
链接：https://arxiv.org/abs/2606.04194
备注：9 pages, 3 figures, 10 tables. Code, data, and per-table receipts: this https URL
摘要：
【32】Dual Advantage Fields
标题：双重优势领域
链接：https://arxiv.org/abs/2606.04188
备注：Accepted by ICML 2026 Workshop on Decision-Making from Offline Datasets to Online Adaptation: Black-Box Optimization to Reinforcement Learning
摘要：
【33】Low-rank Distributional Matrix Completion
标题：低级分布矩阵完成
链接：https://arxiv.org/abs/2606.04176
摘要：
【34】When Autoregressive Consistency Hurts Safety Alignment
标题：当自回归一致性损害安全一致性时
链接：https://arxiv.org/abs/2606.04168
备注：21 pages
摘要：
【35】Expert-Aware Refusal Steering
标题：专家意识的拒绝引导
链接：https://arxiv.org/abs/2606.04160
备注：Under review for COLM 2026
摘要：
【36】EvalStop: Using World Feedback to Detect and Correct Reward Overoptimization in Multi-Tenant RLHF Platforms
标题：EvalStop：使用世界反馈来检测和纠正多租户WLHF平台中的奖励过度优化
链接：https://arxiv.org/abs/2606.04145
摘要：
【37】dMX: Differentiable Mixed-Precision Assignment for Low-Precision Floating-Point Formats
标题：dMX：低精度浮点格式的差异混合精度分配
链接：https://arxiv.org/abs/2606.04115
摘要：
【38】Variance Reduction for Heavy-Tailed Monetization Metrics in Ranking Experiments via Post-Stratification
标题：通过后分层进行排名实验中重尾货币化的方差降低
链接：https://arxiv.org/abs/2606.04110
备注：Accepted as Industry Track paper in the 2026 ACM SIGIR Conference on Research and Development in Information Retrieval
摘要：
【39】Building The Ph(ysical)AI Layer Of Machine Intelligence
标题：构建机器智能的Ph（ysical）AI层
链接：https://arxiv.org/abs/2606.04106
备注：102 pages, 11 Figures
摘要：
【40】The Differentiable Auditory Loop (DAL): An ML Framework for Hyper-Personalized Hearing Aids
标题：可区分听觉环（DAL）：超个性化助听器的ML框架
链接：https://arxiv.org/abs/2606.04103
摘要：
【41】Spectral Scaling Laws of Muon
标题：μ子的谱标度定律
链接：https://arxiv.org/abs/2606.04058
摘要：
【42】A Goal-Set Characterization of Task Composition in the Boolean Task Algebra
标题：布尔任务代数中任务合成的目标集刻画
链接：https://arxiv.org/abs/2606.04053
摘要：
【43】Toward Pre-Deployment Assurance for Enterprise AI Agents: Ontology-Grounded Simulation and Trust Certification
标题：实现企业人工智能代理的预部署保证：基于实体的模拟和信任认证
链接：https://arxiv.org/abs/2606.04037
备注：26 pages, 3 figures. Companion to arXiv:2604.00555
摘要：
【44】Self-Distilled Policy Gradient
标题：自我提炼的政策梯度
链接：https://arxiv.org/abs/2606.04036
摘要：
【45】Pseudospectral Bounds for Transient Amplification in Coupled Gradient Descent
标题：耦合梯度下降中瞬时放大的伪谱界限
链接：https://arxiv.org/abs/2606.04031
备注：11 pages, 3 tables. Accepted as poster at HiLD 2026 (4th Workshop on High-dimensional Learning Dynamics, ICML 2026)
摘要：
【46】AI from concrete to abstract: demystifying artificial intelligence to the general public
标题：人工智能从具体到抽象：向公众揭开人工智能的神秘面纱
链接：https://arxiv.org/abs/2006.04013
备注：23 pages; 2 tables; 47 figures; review comment: Included references for the final published peer-reviewed version of this pre-print: this https URL and this https URL typos corrected
摘要：
【47】Reconstructing Unobservable Temperature Fields via Simulation-Aided Intelligent Sensing
标题：通过仿真辅助智能传感重建不可观测温度场
链接：https://arxiv.org/abs/2606.04582
备注：Presented at IEEE International Instrumentation and Measurement Technology Conference (I2MTC), Nancy, France, 2026
摘要：
【48】Nonlocal Mean Field Schrödinger Bridge with Learned Interactions
标题：非局域平均场薛定谔桥与学习相互作用
链接：https://arxiv.org/abs/2606.04265
备注：31 pages, 15 figures
摘要：
【49】Finite-Iteration Local Dynamics and Warm Starts for Alternating Power Iteration in Spiked Tensor PCA
标题：尖峰张量PCA中交流功率迭代的迭代局部动力学和热启动
链接：https://arxiv.org/abs/2606.04065
备注：67 pages, 0 figures. The paper studies local dynamics and warm-start analysis for alternating power iteration in spiked tensor PCA
摘要：
机器翻译由腾讯交互翻译提供，仅供参考
点击“阅读原文”获取带摘要的学术速递
Python社区是高质量的Python/Django开发社区
本文地址：http://www.python88.com/topic/197302
本文地址：http://www.python88.com/topic/197302