机器学习学术速递[3.3]

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382篇

大模型相关(40篇)

【1】Expanding LLM Agent Boundaries with Strategy-Guided Exploration
标题：通过战略引导的探索扩大LLM代理边界
链接：https://arxiv.org/abs/2603.02045

作者：Andrew Szot,Michael Kirchhof,Omar Attia,Alexander Toshev
摘要：Reinforcement learning (RL) has demonstrated notable success in post-training large language models (LLMs) as agents for tasks such as computer use, tool calling, and coding. However, exploration remains a central challenge in RL for LLM agents, especially as they operate in language-action spaces with complex observations and sparse outcome rewards. In this work, we address exploration for LLM agents by leveraging the ability of LLMs to plan and reason in language about the environment to shift exploration from low-level actions to higher-level language strategies. We thus propose Strategy-Guided Exploration (SGE), which first generates a concise natural-language strategy that describes what to do to make progress toward the goal, and then generates environment actions conditioned on that strategy. By exploring in the space of strategies rather than the space of actions, SGE induces structured and diverse exploration that targets different environment outcomes. To increase strategy diversity during RL, SGE introduces mixed-temperature sampling, which explores diverse strategies in parallel, along with a strategy reflection process that grounds strategy generation on the outcomes of previous strategies in the environment. Across UI interaction, tool-calling, coding, and embodied agent environments, SGE consistently outperforms exploration-focused RL baselines, improving both learning efficiency and final performance. We show that SGE enables the agent to learn to solve tasks too difficult for the base model.

【2】KDFlow: A User-Friendly and Efficient Knowledge Distillation Framework for Large Language Models
标题：KDFlow：一个用户友好且高效的大型语言模型知识提炼框架
链接：https://arxiv.org/abs/2603.01875

作者：Songming Zhang,Xue Zhang,Tong Zhang,Bojie Hu,Yufeng Chen,Jinan Xu
备注：8 pages, 4 figures, 3 tables, code is available at: https://github.com/songmzhang/KDFlow
摘要：Knowledge distillation (KD) is an essential technique to compress large language models (LLMs) into smaller ones. However, despite the distinct roles of the student model and the teacher model in KD, most existing frameworks still use a homogeneous training backend (e.g., FSDP and DeepSpeed) for both models, leading to suboptimal training efficiency. In this paper, we present a novel framework for LLM distillation, termed \textbf{KDFlow}, which features a decoupled architecture and employs SGLang for teacher inference. By bridging the training efficiency of FSDP2 and the inference efficiency of SGLang, KDFlow achieves full utilization of both advantages in a unified system. Moreover, instead of transferring full logits across different processes, our framework only transmits the teacher's hidden states using zero-copy data transfer and recomputes the logits on the student side, effectively balancing the communication cost and KD performance. Furthermore, our framework supports both off-policy and on-policy distillation and incorporates KD algorithms for cross-tokenizer KD through highly extensible and user-friendly APIs. Experiments show that KDFlow can achieve \textbf{1.44$\times$ to 6.36$\times$} speedup compared to current KD frameworks, enabling researchers to rapidly prototype and scale LLM distillation with minimal engineering overhead. Code is available at: https://github.com/songmzhang/KDFlow

【3】D3LM: A Discrete DNA Diffusion Language Model for Bidirectional DNA Understanding and Generation
标题：D3 LM：用于双向DNA理解和生成的离散DNA扩散语言模型
链接：https://arxiv.org/abs/2603.01780

作者：Zhao Yang,Hengchang Liu,Chuan Cao,Bing Su
备注：Accepted as a workshop paper at MLGenX 2026
摘要：Early DNA foundation models adopted BERT-style training, achieving good performance on DNA understanding tasks but lacking generative capabilities. Recent autoregressive models enable DNA generation, but employ left-to-right causal modeling that is suboptimal for DNA where regulatory relationships are inherently bidirectional. We present D3LM (\textbf{D}iscrete \textbf{D}NA \textbf{D}iffusion \textbf{L}anguage \textbf{M}odel), which unifies bidirectional representation learning and DNA generation through masked diffusion. D3LM directly adopts the Nucleotide Transformer (NT) v2 architecture but reformulates the training objective as masked diffusion in discrete DNA space, enabling both bidirectional understanding and generation capabilities within a single model. Compared to NT v2 of the same size, D3LM achieves improved performance on understanding tasks. Notably, on regulatory element generation, D3LM achieves an SFID of 10.92, closely approaching real DNA sequences (7.85) and substantially outperforming the previous best result of 29.16 from autoregressive models. Our work suggests diffusion language models as a promising paradigm for unified DNA foundation models. We further present the first systematic study of masked diffusion models in the DNA domain, investigating practical design choices such as tokenization schemes and sampling strategies, thereby providing empirical insights and a solid foundation for future research. D3LM has been released at https://huggingface.co/collections/Hengchang-Liu/d3lm.

【4】FT-Dojo: Towards Autonomous LLM Fine-Tuning with Language Agents
标题：FT-Dojo：利用语言代理实现自主LLM微调
链接：https://arxiv.org/abs/2603.01712

作者：Qizheng Li,Yifei Zhang,Xiao Yang,Xu Yang,Zhuo Wang,Weiqing Liu,Jiang Bian
备注：24 pages, 6 figures, 9 tables
摘要：Fine-tuning large language models for vertical domains remains a labor-intensive and expensive process, requiring domain experts to curate data, configure training, and iteratively diagnose model behavior. Despite growing interest in autonomous machine learning, no prior work has tackled end-to-end LLM fine-tuning with agents. Can LLM-based agents automate this complete process? We frame this as a substantially open problem: agents must navigate an open-ended search space spanning data curation from diverse data sources, processing with complex tools, building a training pipeline, and iteratively refining their approach based on evaluation outcomes in rapidly growing logs--an overall scenario far more intricate than existing benchmarks. To study this question, we introduce FT-Dojo, an interactive environment comprising 13 tasks across 5 domains. We further develop FT-Agent, an autonomous system that mirrors human experts by leveraging evaluation-driven feedback to iteratively diagnose failures and refine fine-tuning strategies. Experiments on FT-Dojo demonstrate that purpose-built fine-tuning agents significantly outperform general-purpose alternatives, with FT-Agent achieving the best performance on 10 out of 13 tasks across all five domains. Ablations show that the approach generalizes effectively to 3B models, with additional insights on data scaling trade-offs and backbone sensitivity. Case analyses reveal that agents can recover from failures through cumulative learning from historical experience, while also exposing fundamental limitations in causal reasoning--highlighting both the promise and current boundaries of autonomous LLM fine-tuning.

【5】Building a Strong Instruction Language Model for a Less-Resourced Language
标题：为资源较少的语言构建强大的教学语言模型
链接：https://arxiv.org/abs/2603.01691

作者：Domen Vreš,Tjaša Arčon,Timotej Petrič,Dario Vajda,Marko Robnik-Šikonja,Iztok Lebar Bajec
备注：Currently under review at Natural Language Processing Special Issue on Language Models for Low-Resource Languages
摘要：Large language models (LLMs) have become an essential tool for natural language processing and artificial intelligence in general. Current open-source models are primarily trained on English texts, resulting in poorer performance on less-resourced languages and cultures. We present a set of methodological approaches necessary for the successful adaptation of an LLM to a less-resourced language, and demonstrate them using the Slovene language. We present GaMS3-12B, a generative model for Slovene with 12 billion parameters, and demonstrate that it is the best-performing open-source model for Slovene within its parameter range. We adapted the model to the Slovene language using three-stage continual pre-training of the Gemma 3 model, followed by two-stage supervised fine-tuning (SFT). We trained the model on a combination of 140B Slovene, English, Bosnian, Serbian, and Croatian pretraining tokens, and over 200 thousand English and Slovene SFT examples. We evaluate GaMS3-12B on the Slovenian-LLM-Eval datasets, English-to-Slovene translation, and the Slovene LLM arena. We show that the described model outperforms 12B Gemma 3 across all three scenarios and performs comparably to much larger commercial GPT-4o in the Slovene LLM arena, achieving a win rate of over 60 %.

【6】IDProxy: Cold-Start CTR Prediction for Ads and Recommendation at Xiaohongshu with Multimodal LLMs
标题：IDProxy：通过多模式LLM对小红书广告和推荐进行冷启动TLR预测
链接：https://arxiv.org/abs/2603.01590

作者：Yubin Zhang,Haiming Xu,Guillaume Salha-Galvan,Ruiyan Han,Feiyang Xiao,Yanhua Huang,Li Lin,Yang Luo,Yao Hu
摘要：Click-through rate (CTR) models in advertising and recommendation systems rely heavily on item ID embeddings, which struggle in item cold-start settings. We present IDProxy, a solution that leverages multimodal large language models (MLLMs) to generate proxy embeddings from rich content signals, enabling effective CTR prediction for new items without usage data. These proxies are explicitly aligned with the existing ID embedding space and are optimized end-to-end under CTR objectives together with the ranking model, allowing seamless integration into existing large-scale ranking pipelines. Offline experiments and online A/B tests demonstrate the effectiveness of IDProxy, which has been successfully deployed in both Content Feed and Display Ads features of Xiaohongshu's Explore Feed, serving hundreds of millions of users daily.

【7】SafeSci: Safety Evaluation of Large Language Models in Science Domains and Beyond
标题：SafeSci：科学领域及其他领域大型语言模型的安全评估
链接：https://arxiv.org/abs/2603.01589

作者：Xiangyang Zhu,Yuan Tian,Qi Jia,Kaiwei Zhang,Zicheng Zhang,Chunyi Li,Kaiyuan Ji,Dongrui Liu,Zijian Chen,Lu Sun,Renrui Zhang,Yan Teng,Jing Shao,Wei Sun,Xia Hu,Yu Qiao,Guangtao Zhai
摘要：The success of large language models (LLMs) in scientific domains has heightened safety concerns, prompting numerous benchmarks to evaluate their scientific safety. Existing benchmarks often suffer from limited risk coverage and a reliance on subjective evaluation. To address these problems, we introduce SafeSci, a comprehensive framework for safety evaluation and enhancement in scientific contexts. SafeSci comprises SafeSciBench, a multi-disciplinary benchmark with 0.25M samples, and SafeSciTrain, a large-scale dataset containing 1.5M samples for safety enhancement. SafeSciBench distinguishes between safety knowledge and risk to cover extensive scopes and employs objective metrics such as deterministically answerable questions to mitigate evaluation bias. We evaluate 24 advanced LLMs, revealing critical vulnerabilities in current models. We also observe that LLMs exhibit varying degrees of excessive refusal behaviors on safety-related issues. For safety enhancement, we demonstrate that fine-tuning on SafeSciTrain significantly enhances the safety alignment of models. Finally, we argue that knowledge is a double-edged sword, and determining the safety of a scientific question should depend on specific context, rather than universally categorizing it as safe or unsafe. Our work provides both a diagnostic tool and a practical resource for building safer scientific AI systems.

【8】GAC: Stabilizing Asynchronous RL Training for LLMs via Gradient Alignment Control
标题：PAC：通过梯度对齐控制稳定LLM的同步RL训练
链接：https://arxiv.org/abs/2603.01501

作者：Haofeng Xu,Junwei Su,Yukun Tian,Lansong Diao,Zhengping Qian,Chuan Wu
摘要：Asynchronous execution is essential for scaling reinforcement learning (RL) to modern large model workloads, including large language models and AI agents, but it can fundamentally alter RL optimization behavior. While prior work on asynchronous RL focuses on training throughput and distributional correction, we show that naively applying asynchrony to policy-gradient updates can induce qualitatively different training dynamics and lead to severe training instability. Through systematic empirical and theoretical analysis, we identify a key signature of this instability: asynchronous training exhibits persistently high cosine similarity between consecutive policy gradients, in contrast to the near-orthogonal updates observed under synchronized training. This stale-aligned gradient effect amplifies correlated updates and increases the risk of overshooting and divergence. Motivated by this observation, we propose GRADIENT ALIGNMENT CONTROL(GAC), a simple dynamics-aware stabilization method that regulates asynchronous RL progress along stale-aligned directions via gradient projection. We establish convergence guarantees under bounded staleness and demonstrate empirically that GAC recovers stable, on-policy training dynamics and matches synchronized baselines even at high staleness.

【9】Inference-Time Safety For Code LLMs Via Retrieval-Augmented Revision
标题：通过检索增强修订实现代码LLM的推理时安全性
链接：https://arxiv.org/abs/2603.01494

作者：Manisha Mukherjee,Vincent J. Hellendoorn
备注：Accepted at the ICLR 2026 Workshop on Principled Design for Trustworthy AI: Interpretability, Robustness, and Safety Across Modalities
摘要：Large Language Models (LLMs) are increasingly deployed for code generation in high-stakes software development, yet their limited transparency in security reasoning and brittleness to evolving vulnerability patterns raise critical trustworthiness concerns. Models trained on static datasets cannot readily adapt to newly discovered vulnerabilities or changing security standards without retraining, leading to the repeated generation of unsafe code. We present a principled approach to trustworthy code generation by design that operates as an inference-time safety mechanism. Our approach employs retrieval-augmented generation to surface relevant security risks in generated code and retrieve related security discussions from a curated Stack Overflow knowledge base, which are then used to guide an LLM during code revision. This design emphasizes three aspects relevant to trustworthiness: (1) interpretability, through transparent safety interventions grounded in expert community explanations; (2) robustness, by allowing adaptation to evolving security practices without model retraining; and (3) safety alignment, through real-time intervention before unsafe code reaches deployment. Across real-world and benchmark datasets, our approach improves the security of LLM-generated code compared to prompting alone, while introducing no new vulnerabilities as measured by static analysis. These results suggest that principled, retrieval-augmented inference-time interventions can serve as a complementary mechanism for improving the safety of LLM-based code generation, and highlight the ongoing value of community knowledge in supporting trustworthy AI deployment.

【10】3BASiL: An Algorithmic Framework for Sparse plus Low-Rank Compression of LLMs
标题：3BASiL：LLM稀疏加低秩压缩的数学框架
链接：https://arxiv.org/abs/2603.01376

作者：Mehdi Makni,Xiang Meng,Rahul Mazumder
备注：The Thirty-ninth Annual Conference on Neural Information Processing Systems
摘要：Sparse plus Low-Rank $(\mathbf{S} + \mathbf{LR})$ decomposition of Large Language Models (LLMs) has emerged as a promising direction in model compression, aiming to decompose pre-trained model weights into a sum of sparse and low-rank matrices $(\mathbf{W} \approx \mathbf{S} + \mathbf{LR})$. Despite recent progress, existing methods often suffer from substantial performance degradation compared to dense models. In this work, we introduce 3BASiL-TM, an efficient one-shot post-training method for $(\mathbf{S} + \mathbf{LR})$ decomposition of LLMs that addresses this gap. Our approach first introduces a novel 3-Block Alternating Direction Method of Multipliers (ADMM) method, termed 3BASiL, to minimize the layer-wise reconstruction error with convergence guarantees. We then design an efficient transformer-matching (TM) refinement step that jointly optimizes the sparse and low-rank components across transformer layers. This step minimizes a novel memory-efficient loss that aligns outputs at the transformer level. Notably, the TM procedure is universal as it can enhance any $(\mathbf{S} + \mathbf{LR})$ decomposition, including pure sparsity. Our numerical experiments show that 3BASiL-TM reduces the WikiText2 perplexity gap relative to dense LLaMA-8B model by over 30% under a (2:4 Sparse + 64 LR) configuration, compared to prior methods. Moreover, our method achieves over 2.5x faster compression runtime on an A100 GPU compared to SOTA $(\mathbf{S} + \mathbf{LR})$ method. Our code is available at https://github.com/mazumder-lab/3BASiL.

【11】Constructing Synthetic Instruction Datasets for Improving Reasoning in Domain-Specific LLMs: A Case Study in the Japanese Financial Domain
标题：构建综合指令数据集以改善特定领域LLM的推理：日本金融领域的案例研究
链接：https://arxiv.org/abs/2603.01353

作者：Yuma Okochi,Fabio Milentiansen Sim,Tomoyasu Okada
备注：8 pages, 2 figures. Japanese version published in NLP2026
摘要：In adapting LLMs to specific domains, achieving both domain expertise and reasoning ability remains an urgent challenge. This study proposes a general method for constructing high-quality synthetic instruction data for any domain, starting from domain-specific vocabulary. As a demonstration, we applied this method to the financial domain and constructed a large-scale instruction dataset totaling approximately 9.5 billion tokens with Chain-of-Thought reasoning traces. Evaluation results confirmed performance improvements over baseline models on financial benchmarks, demonstrating the effectiveness of our approach. We also report findings on the impact of reasoning trace length on performance and its limitations. Lastly, we open-source our models and datasets on https://huggingface.co/nri-ai .

【12】MetaState: Persistent Working Memory for Discrete Diffusion Language Models
标题：MetaState：离散扩散语言模型的持久工作记忆
链接：https://arxiv.org/abs/2603.01331

作者：Kejing Xia,Mingzhe Li,Lixuan Wei,Zhenbang Du,Xiangchi Yuan,Qirui Jin,Wenke Lee
摘要：Discrete diffusion language models (dLLMs) generate text by iteratively denoising a masked sequence. Compared with autoregressive models, this paradigm naturally supports parallel decoding, bidirectional context, and flexible generation patterns. However, standard dLLMs condition each denoising step only on the current hard-masked sequence, while intermediate continuous representations are discarded after sampling and remasking. We refer to this bottleneck as the \textbf{Information Island} problem. It leads to redundant recomputation across steps and can degrade cross-step consistency. We address this limitation with \textbf{MetaState}, a lightweight recurrent augmentation that equips a frozen dLLM backbone with a persistent, fixed-size working memory that remains independent of sequence length. \textbf{MetaState} consists of three trainable modules: a cross-attention Mixer that reads backbone activations into memory slots, a GRU-style Updater that integrates information across denoising steps, and a cross-attention Injector that feeds the updated memory back into backbone activations. We train these modules with $K$-step unrolling to expose them to multi-step denoising dynamics during fine-tuning. On LLaDA-8B and Dream-7B, \textbf{MetaState} introduces negligible trainable parameters while keeping the backbone frozen, and it consistently improves accuracy over frozen baselines. These results demonstrate that persistent cross-step memory is an effective mechanism for bridging denoising steps and improving generation quality in discrete diffusion language models.

【13】SWE-Adept: An LLM-Based Agentic Framework for Deep Codebase Analysis and Structured Issue Resolution
标题：SWE-Adept：基于LLM的深度代码库分析和结构化问题解决的统计框架
链接：https://arxiv.org/abs/2603.01327

作者：Kang He,Kaushik Roy
摘要：Large language models (LLMs) exhibit strong performance on self-contained programming tasks. However, they still struggle with repository-level software engineering (SWE), which demands (1) deep codebase navigation with effective context management for accurate localization, and (2) systematic approaches for iterative, test-driven code modification to resolve issues. To address these challenges, we propose SWE-Adept, an LLM-based two-agent framework where a localization agent identifies issue-relevant code locations and a resolution agent implements the corresponding fixes. For issue localization, we introduce agent-directed depth-first search that selectively traverses code dependencies. This minimizes issue-irrelevant content in the agent's context window and improves localization accuracy. For issue resolution, we employ adaptive planning and structured problem solving. We equip the agent with specialized tools for progress tracking and Git-based version control. These tools interface with a shared working memory that stores code-state checkpoints indexed by execution steps, facilitating precise checkpoint retrieval. This design enables reliable agent-driven version-control operations for systematic issue resolution, including branching to explore alternative solutions and reverting failed edits. Experiments on SWE-Bench Lite and SWE-Bench Pro demonstrate that SWE-Adept consistently outperforms prior approaches in both issue localization and resolution, improving the end-to-end resolve rate by up to 4.7%.

【14】Truth as a Trajectory: What Internal Representations Reveal About Large Language Model Reasoning
标题：作为轨迹的真理：内部表示揭示了大型语言模型推理的内容
链接：https://arxiv.org/abs/2603.01326

作者：Hamed Damirchi,Ignacio Meza De la Jara,Ehsan Abbasnejad,Afshar Shamsi,Zhen Zhang,Javen Shi
摘要：Existing explainability methods for Large Language Models (LLMs) typically treat hidden states as static points in activation space, assuming that correct and incorrect inferences can be separated using representations from an individual layer. However, these activations are saturated with polysemantic features, leading to linear probes learning surface-level lexical patterns rather than underlying reasoning structures. We introduce Truth as a Trajectory (TaT), which models the transformer inference as an unfolded trajectory of iterative refinements, shifting analysis from static activations to layer-wise geometric displacement. By analyzing displacement of representations across layers, TaT uncovers geometric invariants that distinguish valid reasoning from spurious behavior. We evaluate TaT across dense and Mixture-of-Experts (MoE) architectures on benchmarks spanning commonsense reasoning, question answering, and toxicity detection. Without access to the activations themselves and using only changes in activations across layers, we show that TaT effectively mitigates reliance on static lexical confounds, outperforming conventional probing, and establishes trajectory analysis as a complementary perspective on LLM explainability.

【15】MOSAIC: A Unified Platform for Cross-Paradigm Comparison and Evaluation of Homogeneous and Heterogeneous Multi-Agent RL, LLM, VLM, and Human Decision-Makers
标题：马赛克：跨范式比较和评估同质和异类多代理RL、LLM、VLM和人类决策者的统一平台
链接：https://arxiv.org/abs/2603.01260

作者：Abdulhamid M. Mousa,Yu Fu,Rakhmonberdi Khajiev,Jalaledin M. Azzabi,Abdulkarim M. Mousa,Peng Yang,Yunusa Haruna,Ming Liu
备注：13 pages, 2 figures
摘要：Reinforcement learning (RL), large language models (LLMs), and vision-language models (VLMs) have been widely studied in isolation. However, existing infrastructure lacks the ability to deploy agents from different decision-making paradigms within the same environment, making it difficult to study them in hybrid multi-agent settings or to compare their behaviour fairly under identical conditions. We present MOSAIC, an open-source platform that bridges this gap by incorporating a diverse set of existing reinforcement learning environments and enabling heterogeneous agents (RL policies, LLMs, VLMs, and human players) to operate within them in ad-hoc team settings with reproducible results. MOSAIC introduces three contributions. (i) An IPC-based worker protocol that wraps both native and third-party frameworks as isolated subprocess workers, each executing its native training and inference logic unmodified, communicating through a versioned inter-process protocol. (ii) An operator abstraction that forms an agent-level interface by mapping workers to agents: each operator, regardless of whether it is backed by an RL policy, an LLM, or a human, conforms to a minimal unified interface. (iii) A deterministic cross-paradigm evaluation framework offering two complementary modes: a manual mode that advances up to N concurrent operators in lock-step under shared seeds for fine-grained visual inspection of behavioural differences, and a script mode that drives automated, long-running evaluation through declarative Python scripts, for reproducible experiments. We release MOSAIC as an open, visual-first platform to facilitate reproducible cross-paradigm research across the RL, LLM, and human-in-the-loop communities.

【16】AgilePruner: An Empirical Study of Attention and Diversity for Adaptive Visual Token Pruning in Large Vision-Language Models
标题：AgilePruner：大型视觉语言模型中自适应视觉标记修剪的注意力和多样性的实证研究
链接：https://arxiv.org/abs/2603.01236

作者：Changwoo Baek,Jouwon Song,Sohyeon Kim,Kyeongbo Kong
备注：Accepted to ICLR 2026
摘要：Large Vision-Language Models (LVLMs) have adopted visual token pruning strategies to mitigate substantial computational overhead incurred by extensive visual token sequences. While prior works primarily focus on either attention-based or diversity-based pruning methods, in-depth analysis of these approaches' characteristics and limitations remains largely unexplored. In this work, we conduct thorough empirical analysis using effective rank (erank) as a measure of feature diversity and attention score entropy to investigate visual token processing mechanisms and analyze the strengths and weaknesses of each approach. Our analysis reveals two insights: (1) Our erank-based quantitative analysis shows that many diversity-oriented pruning methods preserve substantially less feature diversity than intended; moreover, analysis using the CHAIR dataset reveals that the diversity they do retain is closely tied to increased hallucination frequency compared to attention-based pruning. (2) We further observe that attention-based approaches are more effective on simple images where visual evidence is concentrated, while diversity-based methods better handle complex images with distributed features. Building on these empirical insights, we show that incorporating image-aware adjustments into existing hybrid pruning strategies consistently improves their performance. We also provide a minimal instantiation of our empirical findings through a simple adaptive pruning mechanism, which achieves strong and reliable performance across standard benchmarks as well as hallucination-specific evaluations. Our project page available at https://cvsp-lab.github.io/AgilePruner.

【17】Reasoning Boosts Opinion Alignment in LLMs
标题：推理促进LLM中的意见对齐
链接：https://arxiv.org/abs/2603.01214

作者：Frédéric Berdoz,Yann Billeter,Yann Vonlanthen,Roger Wattenhofer
备注：Accepted at ICLR 2026
摘要：Opinion modeling aims to capture individual or group political preferences, enabling applications such as digital democracies, where models could help shape fairer and more popular policies. Given their versatility, strong generalization capabilities, and demonstrated success across diverse text-to-text applications, large language models (LLMs) are natural candidates for this task. However, due to their statistical nature and limited causal understanding, they tend to produce biased opinions when prompted naively. In this work, we study whether reasoning can improve opinion alignment. Motivated by the recent advancement in mathematical reasoning enabled by reinforcement learning (RL), we train models to produce profile-consistent answers through structured reasoning. We evaluate our approach on three datasets covering U.S., European, and Swiss politics. Results indicate that reasoning enhances opinion modeling and is competitive with strong baselines, but does not fully remove bias, highlighting the need for additional mechanisms to build faithful political digital twins using LLMs. By releasing both our method and datasets, we establish a solid baseline to support future research on LLM opinion alignment.

【18】Thoth: Mid-Training Bridges LLMs to Time Series Understanding
标题：Thoth：中期训练桥梁LLM到时间序列理解
链接：https://arxiv.org/abs/2603.01042

作者：Jiafeng Lin,Yuxuan Wang,Jialong Wu,Huakun Luo,Zhongyi Pei,Jianmin Wang
摘要：Large Language Models (LLMs) have demonstrated remarkable success in general-purpose reasoning. However, they still struggle to understand and reason about time series data, which limits their effectiveness in decision-making scenarios that depend on temporal dynamics. In this paper, we propose Thoth, the first family of mid-trained LLMs with general-purpose time series understanding capabilities. As a pivotal intermediate stage, mid-training achieves task- and domain-agnostic alignment between time series and natural language, for which we construct Book-of-Thoth, a high-quality, time-series-centric mid-training corpus. Book-of-Thoth enables both time-series-to-text and text-to-time-series generation, equipping LLMs with a foundational grasp of temporal patterns. To better evaluate advanced reasoning capabilities, we further present KnoTS, a novel benchmark of knowledge-intensive time series understanding, designed for joint reasoning over temporal patterns and domain knowledge. Extensive experiments demonstrate that mid-training with Book-of-Thoth enables Thoth to significantly outperform its base model and advanced LLMs across a range of time series question answering benchmarks. Moreover, Thoth exhibits superior capabilities when fine-tuned under data scarcity, underscoring the effectiveness of mid-training for time series understanding. Code is available at: https://github.com/thuml/Thoth.

【19】HiMAC: Hierarchical Macro-Micro Learning for Long-Horizon LLM Agents
标题：HiMAC：面向长期LLM代理的分层宏微学习
链接：https://arxiv.org/abs/2603.00977

作者：Hongbo Jin,Rongpeng Zhu,Jiayu Ding,Wenhao Zhang,Ge Li
摘要：Large language model (LLM) agents have recently demonstrated strong capabilities in interactive decision-making, yet they remain fundamentally limited in long-horizon tasks that require structured planning and reliable execution. Existing approaches predominantly rely on flat autoregressive policies, where high-level reasoning and low-level actions are generated within a single token sequence, leading to inefficient exploration and severe error propagation over extended trajectories. In this work, we propose HiMAC, a hierarchical agentic RL framework that explicitly decomposes long-horizon decision-making into macro-level planning and micro-level execution. HiMAC models reasoning as a structured blueprint generation process followed by goal-conditioned action execution, enabling robust long-horizon planning within LLM-based agents. To train this hierarchy efficiently, we introduce a critic-free hierarchical policy optimization paradigm that extends group-based reinforcement learning to bi-level structures through hierarchical relative advantage estimation. Furthermore, we propose an iterative co-evolution training strategy that alternates between planner exploration and executor adaptation, mitigating the non-stationarity inherent in hierarchical learning. Extensive experiments on ALFWorld, WebShop, and Sokoban demonstrate that HiMAC consistently outperforms strong prompting and reinforcement learning baselines, achieving state-of-the-art performance and substantially improved sample efficiency across both text-based and visually grounded environments. Our results show that introducing structured hierarchy, rather than increasing model scale alone, is a key factor for enabling robust long-horizon agentic intelligence.

【20】Curvature-Weighted Capacity Allocation: A Minimum Description Length Framework for Layer-Adaptive Large Language Model Optimization
标题：曲线加权容量分配：用于层自适应大型语言模型优化的最小描述长度框架
链接：https://arxiv.org/abs/2603.00910

作者：Theophilus Amaefuna,Hitesh Vaidya,Anshuman Chhabra,Ankur Mali
备注：20 pages, 3 figures, 5 tables

【21】Knowledge without Wisdom: Measuring Misalignment between LLMs and Intended Impact
标题：知识无智慧：衡量法学硕士与预期影响之间的不一致
链接：https://arxiv.org/abs/2603.00883

作者：Michael Hardy,Yunsung Kim

【22】Tiny-Critic RAG: Empowering Agentic Fallback with Parameter-Efficient Small Language Models
标题：微小评论家RAG：利用参数高效的小语言模型增强抽象后备能力
链接：https://arxiv.org/abs/2603.00846

作者：Yichao Wu,Penghao Liang,Yafei Xiang,Mengwei Yuan,Jianan Liu,Jing Yang,Xianyou Li,Weiran Yan

【23】Constitutional Black-Box Monitoring for Scheming in LLM Agents
标题：用于LLM代理中计划的宪法黑箱监控
链接：https://arxiv.org/abs/2603.00829

作者：Simon Storf,Rich Barton-Cooper,James Peters-Gill,Marius Hobbhahn

【24】RAIE: Region-Aware Incremental Preference Editing with LoRA for LLM-based Recommendation
标题：RAIE：使用LoRA对基于LLM的推荐进行区域感知增量偏好编辑
链接：https://arxiv.org/abs/2603.00638

作者：Jin Zeng,Yupeng Qi,Hui Li,Chengming Li,Ziyu Lyu,Lixin Cui,Lu Bai

【25】LangGap: Diagnosing and Closing the Language Gap in Vision-Language-Action Models
标题：LangGap：诊断和缩小视觉-语言-行动模型中的语言差距
链接：https://arxiv.org/abs/2603.00592

作者：Yuchen Hou,Lin Zhao
备注：7 pages, 3 figures. Code and benchmark will be available at https://github.com/YC11Hou/langgap

【26】Antibody: Strengthening Defense Against Harmful Fine-Tuning for Large Language Models via Attenuating Harmful Gradient Influence
标题：抗体：通过减弱有害梯度影响来加强对大型语言模型有害微调的防御
链接：https://arxiv.org/abs/2603.00498

作者：Quoc Minh Nguyen,Trung Le,Jing Wu,Anh Tuan Bui,Mehrtash Harandi
备注：Published at ICLR 2026

【27】Wireless Power Control Based on Large Language Models
标题：基于大型语言模型的无线电源控制
链接：https://arxiv.org/abs/2603.00474

作者：Jiacheng Wang,Yucheng Sheng,Le Liang,Hao Ye,Shi Jin
备注：13 pages, 7 figures

【28】How Large Language Models Get Stuck: Early structure with persistent errors
标题：大型语言模型如何陷入困境：具有持续错误的早期结构
链接：https://arxiv.org/abs/2603.00359

作者：Alokesh Manna,William Snyder,Whitney Tabor

【29】StethoLM: Audio Language Model for Cardiopulmonary Analysis Across Clinical Tasks
标题：StethoLM：跨临床任务心肺分析的音频语言模型
链接：https://arxiv.org/abs/2603.00355

作者：Yishan Wang,Tsai-Ning Wang,Mathias Funk,Aaqib Saeed
备注：To be published in TMLR

【30】CoPeP: Benchmarking Continual Pretraining for Protein Language Models
标题：CoPeP：蛋白质语言模型的连续预训练基准
链接：https://arxiv.org/abs/2603.00253

作者：Darshan Patil,Pranshu Malviya,Mathieu Reymond,Quentin Fournier,Sarath Chandar
备注：29 pages, 25 figures

【31】A medical coding language model trained on clinical narratives from a population-wide cohort of 1.8 million patients
标题：医学编码语言模型基于180万患者人群的临床叙述进行训练
链接：https://arxiv.org/abs/2603.00221

作者：Joakim Edin,Sedrah Butt Balaganeshan,Annike Kjølby Kristensen,Lars Maaløe,Ioannis Louloudis,Søren Brunak

【32】Bridging Policy and Real-World Dynamics: LLM-Augmented Rebalancing for Shared Micromobility Systems
标题：弥合政策与现实世界动态：LLM增强的共享微移动系统再平衡
链接：https://arxiv.org/abs/2603.00176

作者：Heng Tan,Hua Yan,Yu Yang
备注：8 pages, 7 figures, accepted by ICRA 2026

【33】LIDS: LLM Summary Inference Under the Layered Lens
标题：LIDS：分层镜头下的LLM总结推理
链接：https://arxiv.org/abs/2603.00105

作者：Dylan Park,Yingying Fan,Jinchi Lv
备注：48 pages, 15 figures

【34】The Value Sensitivity Gap: How Clinical Large Language Models Respond to Patient Preference Statements in Shared Decision-Making
标题：价值敏感性差距：临床大型语言模型如何在共同决策中响应患者偏好陈述
链接：https://arxiv.org/abs/2603.00076

作者：Sanjay Basu
备注：38 pages, 4 figures, supplementary appendix included

【35】Expert Divergence Learning for MoE-based Language Models
标题：基于MoE的语言模型的专家分歧学习
链接：https://arxiv.org/abs/2603.00054

作者：Jiaang Li,Haibin Chen,Langming Liu,Yujin Yuan,Yadao Wang,Yizhen Zhang,Chengting Yu,Xin Tong,Weidong Zhang,Shilei Liu,Wenbo Su,Bo Zheng
备注：ICLR 2026

【36】LitBench: A Graph-Centric Large Language Model Benchmarking Tool For Literature Tasks
标题：LitBench：用于文学任务的以图形为中心的大型语言模型基准工具
链接：https://arxiv.org/abs/2603.00051

作者：Andreas Varvarigos,Ali Maatouk,Jiasheng Zhang,Ngoc Bui,Jialin Chen,Leandros Tassiulas,Rex Ying

【37】Breaking the Factorization Barrier in Diffusion Language Models
标题：打破扩散语言模型中的因子化障碍
链接：https://arxiv.org/abs/2603.00045

作者：Ian Li,Zilei Shao,Benjie Wang,Rose Yu,Guy Van den Broeck,Anji Liu

【38】Maximizing the Spectral Energy Gain in Sub-1-Bit LLMs via Latent Geometry Alignment
标题：通过潜在几何对齐最大化亚1位LLM中的光谱能量收益
链接：https://arxiv.org/abs/2603.00042

作者：Banseok Lee,Youngmin Kim

【39】CARE: Confounder-Aware Aggregation for Reliable LLM Evaluation
标题：CARE：Confounder-Aware聚合以实现可靠的LLM评估
链接：https://arxiv.org/abs/2603.00039

作者：Jitian Zhao,Changho Shin,Tzu-Heng Huang,Satya Sai Srinath Namburi GNVV,Frederic Sala

【40】Probing Materials Knowledge in LLMs: From Latent Embeddings to Reliable Predictions
标题：LLM中的探测材料知识：从潜在嵌入到可靠预测
链接：https://arxiv.org/abs/2603.01834

作者：Vineeth Venugopal,Soroush Mahjoubi,Elsa Olivetti
备注：Under Review

Graph相关(图学习|图神经网络|图优化等)(20篇)

【1】Revealing Combinatorial Reasoning of GNNs via Graph Concept Bottleneck Layer
标题：通过图概念瓶颈层揭示GNN的组合推理
链接：https://arxiv.org/abs/2603.02025

作者：Yue Niu,Zhaokai Sun,Jiayi Yang,Xiaofeng Cao,Rui Fan,Xin Sun,Hanli Wang,Wei Ye
备注：20 pages

【2】Mitigating topology biases in Graph Diffusion via Counterfactual Intervention
标题：通过反事实干预减轻图扩散中的拓扑偏差
链接：https://arxiv.org/abs/2603.02005

作者：Wendi Wang,Jiaxi Yang,Yongkang Du,Lu Lin

【3】BAED: a New Paradigm for Few-shot Graph Learning with Explanation in the Loop
标题：BAED：循环解释的Few-Shot图形学习新范式
链接：https://arxiv.org/abs/2603.01941

作者：Chao Chen,Xujia Li,Dongsheng Hong,Shanshan Lin,Xiangwen Liao,Chuanyi Liu,Lei Chen
备注：Accepted to Neural Networks 2026

【4】From Variance to Invariance: Qualitative Content Analysis for Narrative Graph Annotation
标题：从方差到不变性：叙事图注释的定性内容分析
链接：https://arxiv.org/abs/2603.01930

作者：Junbo Huang,Max Weinig,Ulrich Fritsche,Ricardo Usbeck
备注：LREC 2026 Accepted Paper

【5】Trivial Graph Features and Classical Learning are Enough to Detect Random Anomalies
标题：琐碎的图特征和经典学习足以检测随机异常
链接：https://arxiv.org/abs/2603.01841

作者：Matthieu Latapy,Stephany Rajeh

【6】GCTAM: Global and Contextual Truncated Affinity Combined Maximization Model For Unsupervised Graph Anomaly Detection
标题：GCTAM：用于无监督图异常检测的全局和上下文截断亲和力组合最大化模型
链接：https://arxiv.org/abs/2603.01806

作者：Xiong Zhang,Hong Peng,Zhenli He,Cheng Xie,Xin Jin,Hua Jiang
备注：Accepted by IJCAI 2025

【7】FreeGNN: Continual Source-Free Graph Neural Network Adaptation for Renewable Energy Forecasting
标题：FreeGNN：可再生能源预测的连续无源图神经网络适应
链接：https://arxiv.org/abs/2603.01657

作者：Abderaouf Bahi,Amel Ourici,Ibtissem Gasmi,Aida Derrablia,Warda Deghmane,Mohamed Amine Ferrag
备注：16 pages, 8 figures, 8 tables

【8】Towards OOD Generalization in Dynamic Graphs via Causal Invariant Learning
标题：通过因果不变学习实现动态图中的OOD推广
链接：https://arxiv.org/abs/2603.01626

作者：Xinxun Zhang,Pengfei Jiao,Mengzhou Gao,Tianpeng Li,Xuan Guo
备注：16 pages, 9 figures, accepted by AAAI2026

【9】Invariant-Stratified Propagation for Expressive Graph Neural Networks
标题：表达图神经网络的不变分层传播
链接：https://arxiv.org/abs/2603.01388

作者：Asela Hevapathige,Ahad N. Zehmakan,Asiri Wijesinghe,Saman Halgamuge

【10】Fed-GAME: Personalized Federated Learning with Graph Attention Mixture-of-Experts For Time-Series Forecasting
标题：Fed-GAME：具有图形注意力的个性化联邦学习用于时间序列预测的专家混合
链接：https://arxiv.org/abs/2603.01363

作者：Yi Li,Han Liu,Mingfeng Fan,Guo Chen,Chaojie Li,Biplab Sikdar

【11】DWAFM: Dynamic Weighted Graph Structure Embedding Integrated with Attention and Frequency-Domain MLPs for Traffic Forecasting
标题：DWAPM：动态加权图结构嵌入注意力和频域MLP用于交通预测
链接：https://arxiv.org/abs/2603.00997

作者：Sen Shi,Zhichao Zhang,Yangfan He

【12】GeMi: A Graph-based, Multimodal Recommendation System for Narrative Scroll Paintings
标题：GeMi：基于图形的多模式叙事卷轴画推荐系统
链接：https://arxiv.org/abs/2603.00854

作者：Haimonti Dutta,Pruthvi Moluguri,Jin Dai,Saurabh Amarnath Mahindre

【13】Multi-Domain Riemannian Graph Gluing for Building Graph Foundation Models
标题：用于构建图基础模型的多域Riemann图粘合
链接：https://arxiv.org/abs/2603.00618

作者：Li Sun,Zhenhao Huang,Silei Chen,Lanxu Yang,Junda Ye,Sen Su,Philip S. Yu
备注：Accepted by ICLR'26, 41 pages

【14】Learning to Explore: Policy-Guided Outlier Synthesis for Graph Out-of-Distribution Detection
标题：学习探索：策略引导的离群值合成用于图分布外检测
链接：https://arxiv.org/abs/2603.00602

作者：Li Sun,Lanxu Yang,Jiayu Tian,Bowen Fang,Xiaoyan Yu,Junda Ye,Peng Tang,Hao Peng,Philip S. Yu
备注：Accepted by AAAI'26, 9 pages

【15】GCL-Sampler: Discovering Kernel Similarity for Sampled GPU Simulation via Graph Contrastive Learning
标题：GCL-Sampler：通过图对比学习发现采样图形处理器模拟的内核相似性
链接：https://arxiv.org/abs/2603.00551

作者：Jiaqi Wang,Jingwei Sun,Jiyu Luo,Han Li,Guangzhong Sun

【16】Dynamic Spatio-Temporal Graph Neural Network for Early Detection of Pornography Addiction in Adolescents Based on Electroencephalogram Signals
标题：基于脑电波信号的动态时空图神经网络早期检测青少年色情成瘾
链接：https://arxiv.org/abs/2603.00488

作者：Achmad Ardani Prasha,Clavino Ourizqi Rachmadi,Sabrina Laila Mutiara,Hilman Syachr Ramadhan,Chareyl Reinalyta Borneo,Saruni Dwiasnati
备注：18 pages, 24 figures, 5 tables

【17】GrapHist: Graph Self-Supervised Learning for Histopathology
标题：GrapHist：组织学的图表自我监督学习
链接：https://arxiv.org/abs/2603.00143

作者：Sevda Öğüt,Cédric Vincent-Cuaz,Natalia Dubljevic,Carlos Hurtado,Vaishnavi Subramanian,Pascal Frossard,Dorina Thanou

【18】Graph neural network force fields for adiabatic dynamics of lattice Hamiltonians
标题：格点Hamilton体隔热动力学的图神经网络力场
链接：https://arxiv.org/abs/2603.02039

作者：Yunhao Fan,Gia-Wei Chern
备注：17 pages, 7 figures

【19】Super-resolution of turbulent reacting flows on complex meshes using graph neural networks
标题：利用图神经网络实现复杂网格上湍流反应流的超分辨率
链接：https://arxiv.org/abs/2603.01080

作者：Priyabrat Dash,Konduri Aditya,Christos E. Frouzakis,Mathis Bode

【20】Exploring Drug Safety Through Knowledge Graphs: Protein Kinase Inhibitors as a Case Study
标题：通过知识图谱探索药物安全性：蛋白质Kinase抑制剂作为案例研究
链接：https://arxiv.org/abs/2603.00097

作者：David Jackson,Michael Gertz,Jürgen Hesser
备注：14 pages, 5 figures. Code and data available at https://github.com/davidjackson99/PKI_KG

Transformer(6篇)

【1】What Helps -- and What Hurts: Bidirectional Explanations for Vision Transformers
标题：什么帮助和什么伤害：视觉Transformer的双向补偿
链接：https://arxiv.org/abs/2603.01605

作者：Qin Su,Tie Luo
备注：PAKDD 2026: The 30th Pacific-Asia Conference on Knowledge Discovery and Data Mining

【2】BornoViT: A Novel Efficient Vision Transformer for Bengali Handwritten Basic Characters Classification
标题：BornoViT：一种用于孟加拉手写基本字符分类的新型高效视觉Transformer
链接：https://arxiv.org/abs/2603.00755

作者：Rafi Hassan Chowdhury,Naimul Haque,Kaniz Fatiha

【3】SpectroFusion-ViT: A Lightweight Transformer for Speech Emotion Recognition Using Harmonic Mel-Chroma Fusion
标题：SpectroFusion-ViT：使用Harmonic Mel-Chroma融合的语音情感识别轻量级Transformer
链接：https://arxiv.org/abs/2603.00746

作者：Faria Ahmed,Rafi Hassan Chowdhury,Fatema Tuz Zohora Moon,Sabbir Ahmed

【4】Detecting Transportation Mode Using Dense Smartphone GPS Trajectories and Transformer Models
标题：使用密集智能手机GPS轨迹和Transformer模型检测交通模式
链接：https://arxiv.org/abs/2603.00340

作者：Yuandong Zhang,Othmane Echchabi,Tianshu Feng,Wenyi Zhang,Hsuai-Kai Liao,Charles Chang
备注：Accepted for publication in the International Journal of Geographical Information Science, February 2026. This is the accepted manuscript. The final version of record will appear in IJGIS (Taylor and Francis)

【5】Embedding Morphology into Transformers for Cross-Robot Policy Learning
标题：将形态学嵌入Transformer中以实现跨机器人政策学习
链接：https://arxiv.org/abs/2603.00182

作者：Kei Suzuki,Jing Liu,Ye Wang,Chiori Hori,Matthew Brand,Diego Romeres,Toshiaki Koike-Akino
备注：17 pages, 8 figures (including appendix)

【6】A comparative study of transformer models and recurrent neural networks for path-dependent composite materials
标题：路径相关复合材料Transformer模型和回归神经网络的比较研究
链接：https://arxiv.org/abs/2603.00092

作者：Petter Uvdal,Mohsen Mirkhalaf

GAN|对抗|攻击|生成相关(21篇)

【1】On the Rate of Convergence of GD in Non-linear Neural Networks: An Adversarial Robustness Perspective
标题：非线性神经网络中GDJ的收敛速度：对抗鲁棒性的角度
链接：https://arxiv.org/abs/2603.02095

作者：Guy Smorodinsky,Sveta Gimpleson,Itay Safran

【2】GenDB: The Next Generation of Query Processing -- Synthesized, Not Engineered
标题：GenDB：下一代查询处理--合成的，而不是工程的
链接：https://arxiv.org/abs/2603.02081

作者：Jiale Lao,Immanuel Trummer

【3】Explanation-Guided Adversarial Training for Robust and Interpretable Models
标题：稳健且可解释模型的描述引导对抗训练
链接：https://arxiv.org/abs/2603.01938

作者：Chao Chen,Yanhui Chen,Shanshan Lin,Dongsheng Hong,Shu Wu,Xiangwen Liao,Chuanyi Liu
备注：Accepted by IEEE Transactions On Circuits and Systems For Video Technology (TCSVT 2026)

【4】Adversarial Query Synthesis via Bayesian Optimization
标题：通过Bayesian优化的对抗性查询合成
链接：https://arxiv.org/abs/2603.01570

作者：Jeffrey Tao,Yimeng Zeng,Haydn Thomas Jones,Natalie Maus,Osbert Bastani,Jacob R. Gardner,Ryan Marcus

【5】SubstratumGraphEnv: Reinforcement Learning Environment (RLE) for Modeling System Attack Paths
标题：SubstratumGraphEnv：用于建模系统攻击路径的强化学习环境（RLE）
链接：https://arxiv.org/abs/2603.01340

作者：Bahirah Adewunmi,Edward Raff,Sanjay Purushotham
备注：Presented at the AI for Cyber Security Workshop at AAAI-26

【6】JailNewsBench: Multi-Lingual and Regional Benchmark for Fake News Generation under Jailbreak Attacks
标题：JailNewsBench：越狱攻击下假新闻生成的多语言和地区基准
链接：https://arxiv.org/abs/2603.01291

作者：Masahiro Kaneko,Ayana Niwa,Timothy Baldwin
备注：ICLR 2026

【7】S2O: Enhancing Adversarial Training with Second-Order Statistics of Weights
标题：S2 O：通过权重的二阶统计增强对抗训练
链接：https://arxiv.org/abs/2603.01264

作者：Gaojie Jin,Xinping Yi,Wei Huang,Sven Schewe,Xiaowei Huang
备注：Accepted to TPAMI 2025

【8】BadRSSD: Backdoor Attacks on Regularized Self-Supervised Diffusion Models
标题：BadRSSD：对正规化自我监督扩散模型的后门攻击
链接：https://arxiv.org/abs/2603.01019

作者：Jiayao Wang,Yiping Zhang,Mohammad Maruf Hasan,Xiaoying Lei,Jiale Zhang,Junwu Zhu,Qilin Wu,Dongfang Zhao

【9】Probabilistic Learning and Generation in Deep Sequence Models
标题：深度序列模型中的概率学习和生成
链接：https://arxiv.org/abs/2603.00888

作者：Wenlong Chen
备注：PhD thesis

【10】Curation Leaks: Membership Inference Attacks against Data Curation for Machine Learning
标题：Curation Leaks：Membership Inference Attacks against Data Curation for Machine Learning
链接：https://arxiv.org/abs/2603.00811

作者：Dariush Wahdany,Matthew Jagielski,Adam Dziedzic,Franziska Boenisch
备注：Accepted at ICLR26

【11】Lookahead identification in adversarial bandits: accuracy and memory bounds
标题：对抗土匪中的前瞻识别：准确性和记忆界限
链接：https://arxiv.org/abs/2603.00803

作者：Nataly Brukhim,Nicolò Cesa-Bianchi,Carlo Ciliberto

【12】IU: Imperceptible Universal Backdoor Attack
标题：IU：不可感知的通用后门攻击
链接：https://arxiv.org/abs/2603.00711

作者：Hsin Lin,Yan-Lun Chen,Ren-Hung Hwang,Chia-Mu Yu

【13】TopoEdge: Topology-Grounded Agentic Framework for Edge Networking Code Generation and Repair
标题：TopoEdge：用于边缘网络代码生成和修复的基于拓扑式框架
链接：https://arxiv.org/abs/2603.00569

作者：Haomin Qi,Bohan Liu,Zihan Dai,Yunkai Gao
备注：6 pages, 4 figures, 3 tables

【14】Learning to Attack: A Bandit Approach to Adversarial Context Poisoning
标题：学会攻击：对抗性上下文中毒的强盗方法
链接：https://arxiv.org/abs/2603.00567

作者：Ray Telikani,Amir H. Gandomi

【15】Mathematical Foundations of Poisoning Attacks on Linear Regression over Cumulative Distribution Functions
标题：累积分布函数线性回归中毒攻击的数学基础
链接：https://arxiv.org/abs/2603.00537

作者：Atsuki Sato,Martin Aumüller,Yusuke Matsui
备注：SIGMOD 2026

【16】Multimodal Adaptive Retrieval Augmented Generation through Internal Representation Learning
标题：通过内部表示学习的多模式自适应检索增强生成
链接：https://arxiv.org/abs/2603.00511

作者：Ruoshuang Du,Xin Sun,Qiang Liu,Bowen Song,Zhongqi Chen,Weiqiang Wang,Liang Wang
备注：8 pages, 6 figures

【17】Analyzing Physical Adversarial Example Threats to Machine Learning in Election Systems
标题：分析选举系统中物理对抗示例对机器学习的威胁
链接：https://arxiv.org/abs/2603.00481

作者：Khaleque Md Aashiq Kamal,Surya Eada,Aayushi Verma,Subek Acharya,Adrian Yemin,Benjamin Fuller,Kaleel Mahmood
备注：20 pages, 8 figures, 28 tables

【18】RLShield: Practical Multi-Agent RL for Financial Cyber Defense with Attack-Surface MDPs and Real-Time Response Orchestration
标题：RL Shield：用于金融网络防御的实用多智能体RL，具有攻击面MDP和实时响应预案
链接：https://arxiv.org/abs/2603.00186

作者：Srikumar Nayak
备注：6 pages, 2 fig and 2 tables

【19】NNiT: Width-Agnostic Neural Network Generation with Structurally Aligned Weight Spaces
标题：NNiT：具有结构对齐权重空间的宽度不可知神经网络生成
链接：https://arxiv.org/abs/2603.00180

作者：Jiwoo Kim,Swarajh Mehta,Hao-Lun Hsu,Hyunwoo Ryu,Yudong Liu,Miroslav Pajic

【20】Scaling Quantum Machine Learning without Tricks: High-Resolution and Diverse Image Generation
标题：无需技巧即可扩展量子机器学习：高分辨率和多样化图像生成
链接：https://arxiv.org/abs/2603.00233

作者：Jonas Jäger,Florian J. Kiwit,Carlos A. Riofrío
备注：25 pages, 16 figures. Main text: 14 pages, 7 figures. Appendix: 11 pages, 9 figures

【21】Optimisation of SOUP-GAN and CSR-GAN for High Resolution MR Images Reconstruction
标题：SOUP-GAN和CSR-GAN用于高分辨率MR图像重建的优化
链接：https://arxiv.org/abs/2603.00204

作者：Muneeba Rashid,Hina Shakir,Humaira Mehwish,Asarim Amir,Reema Qaiser Khan

半/弱/无/有监督|不确定性|主动学习(11篇)

【1】Uncertainty Quantification of Click and Conversion Estimates for the Autobidding
标题：自动竞价点击率和转化率估计的不确定性量化
链接：https://arxiv.org/abs/2603.01825

作者：Ivan Zhigalskii,Andrey Pudovikov,Aleksandr Katrutsa,Egor Samosvat
备注：17 pages (10 main text + 7 appendix), 5 figures, 2 tables

【2】Operator Learning Using Weak Supervision from Walk-on-Spheres
标题：基于Walk-on-Spheres的弱监督算子学习
链接：https://arxiv.org/abs/2603.01193

作者：Hrishikesh Viswanath,Hong Chul Nam,Xi Deng,Julius Berner,Anima Anandkumar,Aniket Bera

【3】SphUnc: Hyperspherical Uncertainty Decomposition and Causal Identification via Information Geometry
标题：SphUnc：通过信息几何进行超球不确定性分解和因果识别
链接：https://arxiv.org/abs/2603.01168

作者：Rong Fu,Chunlei Meng,Jinshuo Liu,Dianyu Zhao,Yongtai Liu,Yibo Meng,Xiaowen Ma,Wangyu Wu,Yangchen Zeng,Kangning Cui,Shuaishuai Cao,Simon Fong
备注：22 pages, 15 figures

【4】No More Maybe-Arrows: Resolving Causal Uncertainty by Breaking Symmetries
标题：不再有可能箭头：通过打破对称性来解决因果不确定性
链接：https://arxiv.org/abs/2603.01052

作者：Tingrui Huang,Devendra Singh Dhami

【5】Polynomial Mixing for Efficient Self-supervised Speech Encoders
标题：高效自我监督语音编码器的多项混合
链接：https://arxiv.org/abs/2603.00683

作者：Eva Feillet,Ryan Whetten,David Picard,Alexandre Allauzen
备注：Accepted at ICASSP 2026

【6】CIRCUS: Circuit Consensus under Uncertainty via Stability Ensembles
标题：CIRCUS：通过稳定性合奏在不确定性下的电路共识
链接：https://arxiv.org/abs/2603.00523

作者：Swapnil Parekh

【7】FastBUS: A Fast Bayesian Framework for Unified Weakly-Supervised Learning
标题：FastBus：统一弱监督学习的快速Bayesian框架
链接：https://arxiv.org/abs/2603.00517

作者：Ziquan Wang,Haobo Wang,Ke Chen,Lei Feng,Gang Chen
备注：14 pages, 5 figures

【8】USE: Uncertainty Structure Estimation for Robust Semi-Supervised Learning
标题：用途：稳健半监督学习的不确定性结构估计
链接：https://arxiv.org/abs/2603.00404

作者：Tsao-Lun Chen,Chien-Liang Liu,Tzu-Ming Harry Hsu,Tai-Hsien Wu,Chi-Cheng Fu,Han-Yi E. Chou,Shun-Feng Su
备注：Revised mathematical derivations

【9】Bug Severity Prediction in Software Projects Using Supervised Machine Learning Models
标题：使用有监督的机器学习模型预测软件项目中的错误严重性
链接：https://arxiv.org/abs/2603.00004

作者：Nafisha Tamanna Nice
备注：Master's thesis

【10】A SUPERB-Style Benchmark of Self-Supervised Speech Models for Audio Deepfake Detection
标题：用于音频深度伪造检测的自监督语音模型的SURB式基准
链接：https://arxiv.org/abs/2603.01482

作者：Hashim Ali,Nithin Sai Adupa,Surya Subramani,Hafiz Malik
备注：Accepted at ICASSP

【11】Adaptive Uncertainty-Guided Surrogates for Efficient phase field Modeling of Dendritic Solidification
标题：自适应不确定性引导的树枝状凝固有效相场建模代理
链接：https://arxiv.org/abs/2603.00093

作者：Eider Garate-Perez,Kerman López de Calle-Etxabe,Oihana Garcia,Borja Calvo,Meritxell Gómez-Omella,Jon Lambarri
备注：This manuscript is a preprint and has not yet been peer-reviewed. It has 45 pages and 14 figures

迁移|Zero/Few/One-Shot|自适应(22篇)

【1】Adaptive Confidence Regularization for Multimodal Failure Detection
标题：多峰故障检测的自适应置信正规化
链接：https://arxiv.org/abs/2603.02200

作者：Moru Liu,Hao Dong,Olga Fink,Mario Trapp
备注：Accepted by CVPR 2026

【2】Detection-Gated Glottal Segmentation with Zero-Shot Cross-Dataset Transfer and Clinical Feature Extraction
标题：采用Zero-Shot跨数据集传输和临床特征提取的检测门控阴囊分割
链接：https://arxiv.org/abs/2603.02087

作者：Harikrishnan Unnikrishnan
备注：for associated code see: https://github.com/hari-krishnan/openglottal

【3】CA-AFP: Cluster-Aware Adaptive Federated Pruning
标题：CA-AFP：搜索者意识的自适应联合修剪
链接：https://arxiv.org/abs/2603.01739

作者：Om Govind Jha,Harsh Shukla,Haroon R. Lone

【4】DynaMoE: Dynamic Token-Level Expert Activation with Layer-Wise Adaptive Capacity for Mixture-of-Experts Neural Networks
标题：DynaMoE：具有分层自适应能力的混合专家神经网络动态令牌级专家激活
链接：https://arxiv.org/abs/2603.01697

作者：Gökdeniz Gülmez

【5】Streaming Continual Learning for Unified Adaptive Intelligence in Dynamic Environments
标题：动态环境中实现统一自适应智能的流媒体持续学习
链接：https://arxiv.org/abs/2603.01695

作者：Federico Giannini,Giacomo Ziffer,Andrea Cossu,Vincenzo Lomonaco

【6】Adaptive Spectral Feature Forecasting for Diffusion Sampling Acceleration
标题：扩散采样加速的自适应光谱特征预测
链接：https://arxiv.org/abs/2603.01623

作者：Jiaqi Han,Juntong Shi,Puheng Li,Haotian Ye,Qiushan Guo,Stefano Ermon
备注：CVPR 2026

【7】Scalable Multi-Task Low-Rank Model Adaptation
标题：可扩展的多任务低等级模型自适应
链接：https://arxiv.org/abs/2603.01526

作者：Zichen Tian,Antoine Ledent,Qianru Sun
备注：Published as a conference paper at ICLR 2026. 21 pages, 4 figures, 11 tables. Code is available

【8】Words & Weights: Streamlining Multi-Turn Interactions via Co-Adaptation
标题：词汇和权重：通过协同适应简化多轮交互
链接：https://arxiv.org/abs/2603.01375

作者：Chenxing Wei,Hong Wang,Ying He,Zhongxiang Dai,Bo Jiang,F. Richard Yu,Yao Shu

【9】TripleSumm: Adaptive Triple-Modality Fusion for Video Summarization
标题：TripleSumm：自适应三模态融合视频摘要
链接：https://arxiv.org/abs/2603.01169

作者：Sumin Kim,Hyemin Jeong,Mingu Kang,Yejin Kim,Yoori Oh,Joonseok Lee
备注：Published as a Conference Paper at ICLR 2026

【10】Adaptive-Growth Randomized Neural Networks for Level-Set Computation of Multivalued Nonlinear First-Order PDEs with Hyperbolic Characteristics
标题：自适应增长随机神经网络用于具有双曲特征的多值非线性一阶偏头痛方程的水平集计算
链接：https://arxiv.org/abs/2603.01093

作者：Haoning Dang,Shi Jin,Fei Wang

【11】LLaDA-o: An Effective and Length-Adaptive Omni Diffusion Model
标题：LLaDA-o：一种有效且长度自适应的全方位扩散模型
链接：https://arxiv.org/abs/2603.01068

作者：Zebin You,Xiaolu Zhang,Jun Zhou,Chongxuan Li,Ji-Rong Wen

【12】Fed-ADE: Adaptive Learning Rate for Federated Post-adaptation under Distribution Shift
标题：Fed-ADE：分布转移下联邦后适应的自适应学习率
链接：https://arxiv.org/abs/2603.01040

作者：Heewon Park,Mugon Joe,Miru Kim,Kyungjin Im,Minhae Kwon
备注：Accepted at CVPR 2026

【13】MARS: Harmonizing Multimodal Convergence via Adaptive Rank Search
标题：MARS：通过自适应排名搜索协调多模式收敛
链接：https://arxiv.org/abs/2603.00720

作者：Minkyoung Cho,Insu Jang,Shuowei Jin,Zesen Zhao,Adityan Jothi,Ethem F. Can,Min-Hung Chen,Z. Morley Mao
备注：17 pages; Project Page: this https URL: https://minkyoungcho.github.io/mars/

【14】Adapt Data to Model: Adaptive Transformation Optimization for Domain-shared Time Series Foundation Models
标题：使数据适应模型：域共享时间序列基础模型的自适应转换优化
链接：https://arxiv.org/abs/2603.00629

作者：Yunzhong Qiu,Zhiyao Cen,Zhongyi Pei,Chen Wang,Jianmin Wang
备注：Published as a conference paper at ICLR 2026

【15】Benchmarking Few-shot Transferability of Pre-trained Models with Improved Evaluation Protocols
标题：使用改进的评估协议对预训练模型的Few-Shot可移植性进行基准测试
链接：https://arxiv.org/abs/2603.00478

作者：Xu Luo,Ji Zhang,Lianli Gao,Heng Tao Shen,Jingkuan Song
备注：13 pages

【16】Vectorized Adaptive Histograms for Sparse Oblique Forests
标题：稀疏斜森林的载体化自适应柱状图
链接：https://arxiv.org/abs/2603.00326

作者：Ariel Lubonja,Jungsang Yoon,Haoyin Xu,Yue Wan,Yilin Xu,Richard Stotz,Mathieu Guillame-Bert,Joshua T. Vogelstein,Randal Burns

【17】From Scale to Speed: Adaptive Test-Time Scaling for Image Editing
标题：从规模到速度：图像编辑的自适应测试时缩放
链接：https://arxiv.org/abs/2603.00141

作者：Xiangyan Qu,Zhenlong Yuan,Jing Tang,Rui Chen,Datao Tang,Meng Yu,Lei Sun,Yancheng Bai,Xiangxiang Chu,Gaopeng Gou,Gang Xiong,Yujun Cai
备注：Accepted to the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026

【18】MAML-KT: Addressing Cold Start Problem in Knowledge Tracing for New Students via Few-Shot Model-Agnostic Meta Learning
标题：MAML-KT：通过Few-Shot模型-不可知Meta学习解决新生知识追踪中的冷启动问题
链接：https://arxiv.org/abs/2603.00137

作者：Indronil Bhattacharjee,Christabel Wayllace

【19】StaTS: Spectral Trajectory Schedule Learning for Adaptive Time Series Forecasting with Frequency Guided Denoiser
标题：Stats：基于频率引导去噪器的自适应时间序列预测的谱轨迹调度学习
链接：https://arxiv.org/abs/2603.00037

作者：Jintao Zhang,Zirui Liu,Mingyue Cheng,Xianquan Wang,Zhiding Liu,Qi Liu

【20】GRIP: Geometric Refinement and Adaptive Information Potential for Data Efficiency
标题：GRIP：数据效率的几何细化和自适应信息潜力
链接：https://arxiv.org/abs/2603.00031

作者：Changhao Wang,Jiaolong Yang,Xinhao Yao,Yunfei Yu,Peng Jiao,Lu Yu,Junpeng Fang,Riccardo Cantoro,Qing Cui,Jun Zhou

【21】Co-optimization for Adaptive Conformal Prediction
标题：自适应保形预测的协同优化
链接：https://arxiv.org/abs/2603.01719

作者：Xiaoyi Su,Zhixin Zhou,Rui Luo

【22】Adaptive Estimation and Inference in Conditional Moment Models via the Discrepancy Principle
标题：条件矩模型中的方差原则自适应估计和推理
链接：https://arxiv.org/abs/2603.01337

作者：Jiyuan Tan,Vasilis Syrgkanis

强化学习(13篇)

【1】SEAR: Sample Efficient Action Chunking Reinforcement Learning
标题：SEAR：高效动作分块强化学习示例
链接：https://arxiv.org/abs/2603.01891

作者：C. F. Maximilian Nagy,Onur Celik,Emiliyan Gospodinov,Florian Seligmann,Weiran Liao,Aryan Kaushik,Gerhard Neumann

【2】Rethinking Policy Diversity in Ensemble Policy Gradient in Large-Scale Reinforcement Learning
标题：大规模强化学习中包围策略梯度的策略多样性再思考
链接：https://arxiv.org/abs/2603.01741

作者：Naoki Shitanda,Motoki Omura,Tatsuya Harada,Takayuki Osa
备注：In ICLR 2026. Website at https://naoki04.github.io/paper-cpo/

【3】MVR: Multi-view Video Reward Shaping for Reinforcement Learning
标题：MVR：强化学习的多视图视频奖励整形
链接：https://arxiv.org/abs/2603.01694

作者：Lirui Luo,Guoxi Zhang,Hongming Xu,Yaodong Yang,Cong Fang,Qing Li
备注：ICLR 2026

【4】MIST-RL: Mutation-based Incremental Suite Testing via Reinforcement Learning
标题：MIST-RL：通过强化学习进行基于突变的增量套件测试
链接：https://arxiv.org/abs/2603.01409

作者：Sicheng Zhu,Jiajun Wang,Jiawei Ai,Xin Li
备注：Preprint. 17 pages

【5】PAC Guarantees for Reinforcement Learning: Sample Complexity, Coverage, and Structure
标题：PAC对强化学习的保证：样本复杂性、覆盖率和结构
链接：https://arxiv.org/abs/2603.01309

作者：Joshua Steier
备注：43 pages

【6】Integrating LTL Constraints into PPO for Safe Reinforcement Learning
标题：将LTL约束集成到安全强化学习的PPO中
链接：https://arxiv.org/abs/2603.01292

作者：Maifang Zhang,Hang Yu,Qian Zuo,Cheng Wang,Vaishak Belle,Fengxiang He

【7】Intent-Context Synergy Reinforcement Learning for Autonomous UAV Decision-Making in Air Combat
标题：基于意图-上下文协同强化学习的无人机空战自主决策
链接：https://arxiv.org/abs/2603.00974

作者：Jiahao Fu,Feng Yang

【8】Principled Fast and Meta Knowledge Learners for Continual Reinforcement Learning
标题：用于连续强化学习的原则快速和Meta知识学习者
链接：https://arxiv.org/abs/2603.00903

作者：Ke Sun,Hongming Zhang,Jun Jin,Chao Gao,Xi Chen,Wulong Liu,Linglong Kong
备注：Published in ICLR 2026

【9】MO-MIX: Multi-Objective Multi-Agent Cooperative Decision-Making With Deep Reinforcement Learning
标题：MO-MIX：采用深度强化学习的多目标多智能体协同决策
链接：https://arxiv.org/abs/2603.00730

作者：Tianmeng Hu,Biao Luo,Chunhua Yang,Tingwen Huang
备注：15 pages, 10 figures, published in IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)

【10】Steering Away from Memorization: Reachability-Constrained Reinforcement Learning for Text-to-Image Diffusion
标题：远离并行化：文本到图像扩散的可达性约束强化学习
链接：https://arxiv.org/abs/2603.00140

作者：Sathwik Karnik,Juyeop Kim,Sanmi Koyejo,Jong-Seok Lee,Somil Bansal

【11】Safe Multi-Agent Deep Reinforcement Learning for Privacy-Aware Edge-Device Collaborative DNN Inference
标题：用于隐私感知边缘设备协作DNN推理的安全多智能体深度强化学习
链接：https://arxiv.org/abs/2603.00129

作者：Hong Wang,Xuwei Fan,Zhipeng Cheng,Yachao Yuan,Minghui Min,Minghui Liwang,Xiaoyu Xia
备注：14 pages

【12】Reinforcement Learning for Control with Probabilistic Stability Guarantee: A Finite-Sample Approach
标题：具有概率稳定性保证的控制强化学习：随机样本方法
链接：https://arxiv.org/abs/2603.00043

作者：Minghao Han,Lixian Zhang,Chenliang Liu,Zhipeng Zhou,Jun Wang,Wei Pan

【13】Alpha-RF: Automated RF-Filter-Circuit Design with Neural Simulator and Reinforcement Learning
标题：Alpha-RF：采用神经模拟器和强化学习的自动RF过滤器电路设计
链接：https://arxiv.org/abs/2603.00104

作者：Nhat Tran,Chenjie Hao,Alexander Stameroff,Anh-Vu Pham,Yubei Chen

元学习(2篇)

【1】Meta-Learning Hyperparameters for Parameter Efficient Fine-Tuning
标题：元学习超参数实现参数高效微调
链接：https://arxiv.org/abs/2603.01759

作者：Zichen Tian,Yaoyao Liu,Qianru Sun
备注：Accepted by CVPR 2025 (Highlight). Code is available at: https://github.com/doem97/metalora

【2】Hereditary Geometric Meta-RL: Nonlocal Generalization via Task Symmetries
标题：遗传几何元RL：通过任务对称性的非局部概括
链接：https://arxiv.org/abs/2603.00396

作者：Paul Nitschke,Shahriar Talebi
备注：Accepted to 2026 American Control Conference

医学相关(10篇)

【1】Learning to Read Where to Look: Disease-Aware Vision-Language Pretraining for 3D CT
标题：学习阅读在哪里看：3D CT的疾病感知视觉语言预训练
链接：https://arxiv.org/abs/2603.02026

作者：Simon Ging,Philipp Arnold,Sebastian Walter,Hani Alnahas,Hannah Bast,Elmar Kotter,Jiancheng Yang,Behzad Bozorgtabar,Thomas Brox

【2】CARE: Towards Clinical Accountability in Multi-Modal Medical Reasoning with an Evidence-Grounded Agentic Framework
标题：CARE：以证据为基础的统计框架在多模式医学推理中实现临床问责
链接：https://arxiv.org/abs/2603.01607

作者：Yuexi Du,Jinglu Wang,Shujie Liu,Nicha C. Dvornek,Yan Lu
备注：Accepted by ICLR 2026

【3】Differential privacy representation geometry for medical image analysis
标题：用于医学图像分析的差异隐私表示几何
链接：https://arxiv.org/abs/2603.01098

作者：Soroosh Tayebi Arasteh,Marziyeh Mohammadi,Sven Nebelung,Daniel Truhn

【4】Identifying and Characterising Response in Clinical Trials: Development and Validation of a Machine Learning Approach in Colorectal Cancer
标题：识别和描述临床试验中的反应：结直肠癌机器学习方法的开发和验证
链接：https://arxiv.org/abs/2603.00757

作者：Adam Marcus,Paul Agapow
备注：Accepted in NewInML @ NeurIPS 2020

【5】How Well Do Multimodal Models Reason on ECG Signals?
标题：多峰模型对心电图信号的推理效果如何？
链接：https://arxiv.org/abs/2603.00312

作者：Maxwell A. Xu,Harish Haresumadram,Catherine W. Liu,Patrick Langer,Jathurshan Pradeepkumar,Wanting Mao,Sunita J. Ferns,Aradhana Verma,Jimeng Sun,Paul Schmiedmayer,Xin Liu,Daniel McDuff,Emily B. Fox,James M. Rehg

【6】Engineering FAIR Privacy-preserving Applications that Learn Histories of Disease
标题：工程FAIR了解疾病历史的隐私保护应用程序
链接：https://arxiv.org/abs/2603.00181

作者：Ines N. Duarte,Praphulla M. S. Bhawsar,Lee K. Mason,Jeya Balaji Balasubramanian,Daniel E. Russ,Arlindo L. Oliveira,Jonas S. Almeida

【7】A Representation-Consistent Gated Recurrent Framework for Robust Medical Time-Series Classification
标题：用于稳健医学时间序列分类的表示一致的门控回归框架
链接：https://arxiv.org/abs/2603.00067

作者：Maitri Krishna Sai
备注：7 pages, 1 figure. Preprint

【8】REMIND: Rethinking Medical High-Modality Learning under Missingness--A Long-Tailed Distribution Perspective
标题：REMIND：缺失下医学高模态学习的再思考--长尾分布视角
链接：https://arxiv.org/abs/2603.00046

作者：Chenwei Wu,Zitao Shuai,Liyue Shen

【9】Econometric vs. Causal Structure-Learning for Time-Series Policy Decisions: Evidence from the UK COVID-19 Policies
标题：时间序列政策决策的计量经济学与因果结构学习：来自英国COVID-19政策的证据
链接：https://arxiv.org/abs/2603.00041

作者：Bruno Petrungaro,Anthony C. Constantinou

【10】Tipping the Balance: Impact of Class Imbalance Correction on the Performance of Clinical Risk Prediction Models
标题：打破平衡：阶级失衡纠正对临床风险预测模型性能的影响
链接：https://arxiv.org/abs/2603.00208

作者：Amalie Koch Andersen,Hadi Mehdizavareh,Arijit Khan,Tobias Becher,Simone Britsch,Markward Britsch,Morten Bøttcher,Simon Winther,Palle Duun Rohde,Morten Hasselstrøm Jensen,Simon Lebech Cichosz

蒸馏|知识提取(2篇)

【1】Sketch2Colab: Sketch-Conditioned Multi-Human Animation via Controllable Flow Distillation
标题：Sketch 2Colab：通过可控流蒸馏制作草图条件化的多人动画
链接：https://arxiv.org/abs/2603.02190

作者：Divyanshu Daiya,Aniket Bera
备注：Accepted to CVPR 2026 Main Conference (11 pages, 5 figures)

【2】Turning Black Box into White Box: Dataset Distillation Leaks
标题：将黑匣子变成白盒：数据集蒸馏泄漏
链接：https://arxiv.org/abs/2603.01053

作者：Huajie Chen,Tianqing Zhu,Yuchen Zhong,Yang Zhang,Shang Wang,Feng He,Lefeng Zhang,Jialiang Shen,Minghao Wang,Wanlei Zhou

推荐(3篇)

【1】Stop Treating Collisions Equally: Qualification-Aware Semantic ID Learning for Recommendation at Industrial Scale
标题：停止平等地对待冲突：具有资格意识的语义ID学习以供工业规模推荐
链接：https://arxiv.org/abs/2603.00632

作者：Zheng Hu,Yuxin Chen,Yongsen Pan,Xu Yuan,Yuting Yin,Daoyuan Wang,Boyang Xia,Zefei Luo,Hongyang Wang,Songhao Ni,Dongxu Liang,Jun Wang,Shimin Cai,Tao Zhou,Fuji Ren,Wenwu Ou

【2】Trinity: A Scenario-Aware Recommendation Framework for Large-Scale Cold-Start Users
标题：Trinity：面向大规模冷启动用户的场景感知推荐框架
链接：https://arxiv.org/abs/2603.00502

作者：Wenhao Zheng,Wang Lu,Fangshuang Tang,Yiyang Lu,Jun Yang,Pengcheng Xiong,Yulan Yan

【3】Mag-Mamba: Modeling Coupled spatiotemporal Asymmetry for POI Recommendation
标题：Mag-Mamba：为兴趣点推荐建模耦合时空不对称
链接：https://arxiv.org/abs/2603.00053

作者：Zhuoxuan Li,Tangwei Ye,Jieyuan Pei,Haina Liang,Zhongyuan Lai,Zihan Liu,Yiming Wu,Qi Zhang,Liang Hu
备注：14 pages, 7 figures

超分辨率|去噪|去模糊|去雾(1篇)

【1】CASCADE: Cross-scale Advective Super-resolution with Climate Assimilation and Downscaling Evolution
标题：级联：跨尺度平流超分辨率与气候同化和降尺度演变
链接：https://arxiv.org/abs/2603.00109

作者：Alexander Kovalenko

自动驾驶|车辆|车道检测等(1篇)

【1】The Impact of Battery Cell Configuration on Electric Vehicle Performance: An XGBoost-Based Classification with SHAP Interpretability
标题：电池配置对电动汽车性能的影响：基于XGBoost的具有SHAP可解释性的分类
链接：https://arxiv.org/abs/2603.01275

作者：Santanam Wishal,Louis Filiepe Tio Jansel,Matthew Abednego Inkiriwang,Jason Sebastian
备注：12 pages, 7 figures, 3 tables

点云|SLAM|雷达|激光|深度RGBD相关(2篇)

【1】Learning Vision-Based Omnidirectional Navigation: A Teacher-Student Approach Using Monocular Depth Estimation
标题：学习基于视觉的全方位导航：使用单目深度估计的师生方法
链接：https://arxiv.org/abs/2603.01999

作者：Jan Finke,Wayne Paul Martis,Adrian Schmelter,Lars Erbach,Christian Jestel,Marvin Wiedemann

【2】Spectral Condition for $μ$P under Width-Depth Scaling
标题：宽深标度下$μ$P的光谱条件
链接：https://arxiv.org/abs/2603.00541

作者：Chenyu Zheng,Rongzhen Wang,Xinyu Zhang,Chongxuan Li
备注：51 pages, 6 figures, 13 tables

联邦学习|隐私保护|加密(2篇)

【1】Decentralized Federated Learning by Partial Message Exchange
标题：通过部分消息交换实现去中心化联邦学习
链接：https://arxiv.org/abs/2603.01730

作者：Shan Sha,Shenglong Zhou,Xin Wang,Lingchen Kong,Geoffrey Ye Li

【2】DeepAFL: Deep Analytic Federated Learning
标题：DeepAFL：深度分析联邦学习
链接：https://arxiv.org/abs/2603.00579

作者：Jianheng Tang,Yajiang Huang,Kejia Fan,Feijiang Han,Jiaxu Li,Jinfeng Xu,Run He,Anfeng Liu,Houbing Herbert Song,Huiping Zhuang,Yunhuai Liu
备注：Accepted in the Fourteenth International Conference on Learning Representations (ICLR 2026)

推理|分析|理解|解释(23篇)

【1】Partial Causal Structure Learning for Valid Selective Conformal Inference under Interventions
标题：干预下有效选择性保形推理的部分因果结构学习
链接：https://arxiv.org/abs/2603.02204

作者：Amir Asiaee,Kavey Aryan,James P. Long

【2】Symbol-Equivariant Recurrent Reasoning Models
标题：符号等变循环推理模型
链接：https://arxiv.org/abs/2603.02193

作者：Richard Freinschlag,Timo Bertram,Erich Kobler,Andreas Mayr,Günter Klambauer

【3】Is Bigger Always Better? Efficiency Analysis in Resource-Constrained Small Object Detection
标题：越大越好吗？资源受限小目标检测的效率分析
链接：https://arxiv.org/abs/2603.02142

作者：Kwame Mbobda-Kuate,Gabriel Kasmi
备注：13 pages, 9 figures, 8 tables

【4】Pencil Puzzle Bench: A Benchmark for Multi-Step Verifiable Reasoning
标题：收件箱益智长凳：多步骤可验证推理的基准
链接：https://arxiv.org/abs/2603.02119

作者：Justin Waugh

【5】Recursive Models for Long-Horizon Reasoning
标题：长期推理的回归模型
链接：https://arxiv.org/abs/2603.02112

作者：Chenxiao Yang,Nathan Srebro,Zhiyuan Li

【6】Learning from Synthetic Data Improves Multi-hop Reasoning
标题：从合成数据中学习改进多跳推理
链接：https://arxiv.org/abs/2603.02091

作者：Anmol Kabra,Yilun Yin,Albert Gong,Kamilė Stankevičiūtė,Dongyoung Go,Johann Lee,Katie Z. Luo,Carla P. Gomes,Kilian Q. Weinberger
备注：Accepted to ICLR 2026

【7】Noise-Calibrated Inference from Differentially Private Sufficient Statistics in Exponential Families
标题：指数族中差异私人充分统计量的噪音校准推断
链接：https://arxiv.org/abs/2603.02010

作者：Amir Asiaee,Samhita Pal

【8】Semantic Similarity is a Spurious Measure of Comic Understanding: Lessons Learned from Hallucinations in a Benchmarking Experiment
标题：语义相似性是喜剧理解的虚假衡量标准：从基准实验中的幻觉中学到的教训
链接：https://arxiv.org/abs/2603.01950

作者：Christopher Driggers-Ellis,Nachiketh Tibrewal,Rohit Bogulla,Harsh Khanna,Sangpil Youm,Christan Grant,Bonnie Dorr
备注：8 pages, 2 figures, 3 tables. Includes link to code

【9】Generalizing Logic-based Explanations for Machine Learning Classifiers via Optimization
标题：通过优化概括机器学习分类器的基于逻辑的解释
链接：https://arxiv.org/abs/2603.01870

作者：Francisco Mateus Rocha Filho,Ajalmar Rêgo da Rocha Neto,Thiago Alves Rocha
备注：Preprint version. For the final published version, see the DOI below

【10】Hyperparameter Trajectory Inference with Conditional Lagrangian Optimal Transport
标题：条件拉格朗日最优输运的超参数轨迹推断
链接：https://arxiv.org/abs/2603.01771

作者：Harry Amad,Mihaela van der Schaar

【11】Reasoning as Gradient: Scaling MLE Agents Beyond Tree Search
标题：推理作为梯度：扩展MLE代理超越树搜索
链接：https://arxiv.org/abs/2603.01692

作者：Yifei Zhang,Xu Yang,Xiao Yang,Bowen Xian,Qizheng Li,Shikai Fang,Jingyuan Li,Jian Wang,Mingrui Xu,Weiqing Liu,Jiang Bian
备注：24 pages, 7 figures

【12】Jump Like A Squirrel: Optimized Execution Step Order for Anytime Random Forest Inference
标题：像松鼠一样跳跃：随时随机森林推理的优化执行步骤顺序
链接：https://arxiv.org/abs/2603.01588

作者：Daniel Biebert,Christian Hakert,Kay Heider,Daniel Kuhse,Sebastian Buschjäger,Jian-Jia Chen

【13】Quasar: Quantized Self-Speculative Acceleration for Rapid Inference via Memory-Efficient Verification
标题：类人猿：通过记忆高效验证实现快速推理的量化自我推测加速
链接：https://arxiv.org/abs/2603.01399

作者：Guang Huang,Zeyi Wen
备注：10 pages

【14】Theoretical Perspectives on Data Quality and Synergistic Effects in Pre- and Post-Training Reasoning Models
标题：训练前后推理模型中数据质量和协同效应的理论观点
链接：https://arxiv.org/abs/2603.01293

作者：Adel Javanmard,Baharan Mirzasoleiman,Vahab Mirrokni
备注：35 pages, 5 figures

【15】Opponent State Inference Under Partial Observability: An HMM-POMDP Framework for 2026 Formula 1 Energy Strategy
标题：部分可观察性下的对手状态推断：2026年一级方程式能源战略的HMM-POMDP框架
链接：https://arxiv.org/abs/2603.01290

作者：Kalliopi Kleisarchaki
备注：17 pages. Pre-registered theoretical framework; empirical calibration on 2026 race telemetry begins Australian Grand Prix, 8 March 2026. Paper 1 of 3. ResearchGate preprint: DOI 10.13140/RG.2.2.16034.08644

【16】Understanding LoRA as Knowledge Memory: An Empirical Analysis
标题：将LoRA理解为知识记忆：一个实证分析
链接：https://arxiv.org/abs/2603.01097

作者：Seungju Back,Dongwoo Lee,Naun Kang,Taehee Lee,S. K. Hong,Youngjune Gwon,Sungjin Ahn

【17】One-Token Verification for Reasoning Correctness Estimation
标题：推理正确性估计的单令牌验证
链接：https://arxiv.org/abs/2603.01025

作者：Zhan Zhuang,Xiequn Wang,Zebin Chen,Feiyang Ye,Ying Wei,Kede Ma,Yu Zhang

【18】Navigating Time's Possibilities: Plausible Counterfactual Explanations for Multivariate Time-Series Forecast through Genetic Algorithms
标题：探索时间的可能性：通过遗传算法进行多元时间序列预测的合理反事实解释
链接：https://arxiv.org/abs/2603.00855

作者：Gianlucca Zuin,Adriano Veloso
备注：Published on IEEE TrustCom 2024

【19】Token Management in Multi-Tenant AI Inference Platforms
标题：多租户人工智能推理平台中的代币管理
链接：https://arxiv.org/abs/2603.00356

作者：William J. Cunningham
备注：10 pages, 6 figures

【20】Stepwise Penalization for Length-Efficient Chain-of-Thought Reasoning
标题：对高效思维链推理的逐步惩罚
链接：https://arxiv.org/abs/2603.00296

作者：Xintong Li,Sha Li,Rongmei Lin,Hongye Jin,Linwei Li,Hejie Cui,Sarah Zhang,Chia-Yuan Chang,Kewei Cheng,Besnik Fetahu,Priyanka Nigam,Jingbo Shang,Bing Yin
备注：Preprint

【21】Instrumental and Proximal Causal Inference with Gaussian Processes
标题：高斯过程的工具性和近因推理
链接：https://arxiv.org/abs/2603.02159

作者：Yuqi Zhang,Krikamol Muandet,Dino Sejdinovic,Edwin Fong,Siu Lun Chau

【22】Dual-space posterior sampling for Bayesian inference in constrained inverse problems
标题：约束反问题中Bayesian推理的双空间后验抽样
链接：https://arxiv.org/abs/2603.00393

作者：Ali Siahkoohi,Kamal Aghazade,Ali Gholami

【23】Profiling vs. Case-specific Evidence: A Probabilistic Analysis
标题：特征分析与特定案例证据：概率分析
链接：https://arxiv.org/abs/2603.00098

作者：Marcello Di Bello,Nicolò Cangiotti,Michele Loi
备注：16 pages

检测相关(7篇)

【1】Neurosymbolic Learning for Advanced Persistent Threat Detection under Extreme Class Imbalance
标题：极端阶级失衡下高级持续威胁检测的神经符号学习
链接：https://arxiv.org/abs/2603.00453

作者：Quhura Fathima,Neda Moghim,Mostafa Taghizade Firouzjaee,Christo K. Thomas,Ross Gore,Walid Saad
备注：6 pages, 4 figures, accepted at IEEE ICC 2026

【2】Deep Learning-Based Meat Freshness Detection with Segmentation and OOD-Aware Classification
标题：基于深度学习的肉类新鲜度检测与分割和OOD感知分类
链接：https://arxiv.org/abs/2603.00368

作者：Hutama Arif Bramantyo,Mukarram Ali Faridi,Rui Chen,Clarissa Harris,Yin Sun

【3】Quantifying Catastrophic Forgetting in IoT Intrusion Detection Systems
标题：量化物联网入侵检测系统中的灾难性遗忘
链接：https://arxiv.org/abs/2603.00363

作者：Sourasekhar Banerjee,David Bergqvist,Salman Toor,Christian Rohner,Andreas Johnsson
备注：6 pages, 4 figures

【4】Detecting Cognitive Signatures in Typing Behavior for Non-Intrusive Authorship Verification
标题：检测打字行为中的认知签名以进行非侵入性作者身份验证
链接：https://arxiv.org/abs/2603.00177

作者：David Condrey
备注：6 pages

【5】Learning Under Extreme Data Scarcity: Subject-Level Evaluation of Lightweight CNNs for fMRI-Based Prodromal Parkinsons Detection
标题：极端数据稀缺下的学习：用于基于fMRI的先兆帕金森病检测的轻量级CNN的主题级评估
链接：https://arxiv.org/abs/2603.00060

作者：Naimur Rahman
备注：Methodological case study cs.LG on subject-level evaluation and model capacity under extreme data scarcity; 9 pages, 1 figure. Experiments use 40-subject PPMI fMRI cohort; no external validation

【6】M3-AD: Reflection-aware Multi-modal, Multi-category, and Multi-dimensional Benchmark and Framework for Industrial Anomaly Detection
标题：M3-AD：反射感知的多模式、多类别和多维工业异常检测基准和框架
链接：https://arxiv.org/abs/2603.00055

作者：Chao Huang,Yanhui Li,Yunkang Cao,Wei Wang,Hongxi Huang,Jie Wen,Wenqi Ren,Xiaochun Cao

【7】TRAKNN: Efficient Trajectory Aware Spatiotemporal kNN for Rare Meteorological Trajectory Detection
标题：TRAKNN：用于罕见气象轨迹检测的高效轨迹感知时空kNN
链接：https://arxiv.org/abs/2603.02059

作者：Guillaume Coulaud,Davide Faranda

分类|识别(6篇)

【1】Leveraging Model Soups to Classify Intangible Cultural Heritage Images from the Mekong Delta
标题：利用模型汤对湄公河三角洲非物质文化遗产图像进行分类
链接：https://arxiv.org/abs/2603.02181

作者：Quoc-Khang Tran,Minh-Thien Nguyen,Nguyen-Khang Pham
备注：Early accept of Vol 2025 No 3, November : Journal on Information Technologies & Communications

【2】Leave-One-Out Prediction for General Hypothesis Classes
标题：一般假设类的留一预测
链接：https://arxiv.org/abs/2603.02043

作者：Jian Qian,Jiachen Xu

【3】UTICA: Multi-Objective Self-Distllation Foundation Model Pretraining for Time Series Classification
标题：UTICA：时间序列分类的多目标自分散基础模型预训练
链接：https://arxiv.org/abs/2603.01348

作者：Yessin Moakher,Youssef Attia El Hili,Vasilii Feofanov

【4】I Can't Believe It's Not Robust: Catastrophic Collapse of Safety Classifiers under Embedding Drift
标题：我不敢相信它不健壮：嵌入漂移下安全分类器的灾难性崩溃
链接：https://arxiv.org/abs/2603.01297

作者：Subramanyam Sahoo,Vinija Jain,Divya Chaudhary,Aman Chadha
备注：Accepted at the ICBINB: Where LLMs Need to Improve workshop at ICLR 2026. 12 pages and 3 Figures

【5】Energy-Efficient Information Representation in MNIST Classification Using Biologically Inspired Learning
标题：使用生物启发学习的MNIST分类中的节能信息表示
链接：https://arxiv.org/abs/2603.00588

作者：Patrick Stricker,Florian Röhrbein,Andreas Knoblauch
备注：14 pages, accepted for publication in proceedings of the 10th BWHPC Symposium

【6】High-Resolution Range Profile Classifiers Require Aspect-Angle Awareness
标题：高分辨率距离轮廓分类器需要轴角感知
链接：https://arxiv.org/abs/2603.00087

作者：Edwyn Brient,Santiago Velasco-Forero,Rami Kassab

表征(7篇)

【1】Temporal Representations for Exploration: Learning Complex Exploratory Behavior without Extrinsic Rewards
标题：探索的时间表征：在没有外在奖励的情况下学习复杂的探索行为
链接：https://arxiv.org/abs/2603.02008

作者：Faisal Mohamed,Catherine Ji,Benjamin Eysenbach,Glen Berseth

【2】A Deep Learning Framework for Heat Demand Forecasting using Time-Frequency Representations of Decomposed Features
标题：使用分解特征的时频表示进行热需求预测的深度学习框架
链接：https://arxiv.org/abs/2603.01137

作者：Adithya Ramachandran,Satyaki Chatterjee,Thorkil Flensmark B. Neergaard,Maximilian Oberndoerfer,Andreas Maier,Siming Bayer

【3】AG-REPA: Causal Layer Selection for Representation Alignment in Audio Flow Matching
标题：AG-REPA：音频流匹配中用于表示对齐的因果层选择
链接：https://arxiv.org/abs/2603.01006

作者：Pengfei Zhang,Tianxin Xie,Minghao Yang,Li Liu
备注：13 pages, 4 figures, 4 tables

【4】Forgetting is Competition: Rethinking Unlearning as Representation Interference in Diffusion Models
标题：遗忘就是竞争：重新思考遗忘作为扩散模型中的代表干扰
链接：https://arxiv.org/abs/2603.00975

作者：Ashutosh Ranjan,Vivek Srivastava,Shirish Karande,Murari Mandal

【5】A Gauge Theory of Superposition: Toward a Sheaf-Theoretic Atlas of Neural Representations
标题：叠加规范理论：走向神经表示的层理论地图集
链接：https://arxiv.org/abs/2603.00824

作者：Hossein Javidnia
备注：16 pages, 4 figures

【6】Interpretable Cross-Network Attention for Resting-State fMRI Representation Learning
标题：静息状态fMRI表示学习的可解释跨网络注意力
链接：https://arxiv.org/abs/2603.00786

作者：Karanpartap Singh,Adam Turnbull,Mohammad Abbasi,Kilian Pohl,Feng Vankee Lin,Ehsan Adeli

【7】BiJEPA: Bi-directional Joint Embedding Predictive Architecture for Symmetric Representation Learning
标题：BiJEPA：用于对称表示学习的双向联合嵌入预测架构
链接：https://arxiv.org/abs/2603.00049

作者：Yongchao Huang
备注：12 pages

3D|3D重建等相关(3篇)

【1】Monocular 3D Object Position Estimation with VLMs for Human-Robot Interaction
标题：基于VLM的人机交互单目3D物体位置估计
链接：https://arxiv.org/abs/2603.01224

作者：Ari Wahl,Dorian Gawlinski,David Przewozny,Paul Chojecki,Felix Bießmann,Sebastian Bosse
备注：Accepted at Workshop on Integrating Image Processing with Large-Scale Language/Vision Models for Advanced Visual Understanding (LVLM) at IEEE International Conference on Image Processing (ICIP) 2025

【2】Exploring 3D Dataset Pruning
标题：探索3D数据集修剪
链接：https://arxiv.org/abs/2603.00651

作者：Xiaohan Zhao,Xinyi Shang,Jiacheng Liu,Zhiqiang Shen
备注：Code: https://github.com/XiaohanZhao123/3D-Dataset-Pruning

【3】ArtiFixer: Enhancing and Extending 3D Reconstruction with Auto-Regressive Diffusion Models
标题：ArtiFixer：使用自回归扩散模型增强和扩展3D重建
链接：https://arxiv.org/abs/2603.00492

作者：Riccardo de Lutio,Tobias Fischer,Yen-Yu Chang,Yuxuan Zhang,Jay Zhangjie Wu,Xuanchi Ren,Tianchang Shen,Katarina Tothova,Zan Gojcic,Haithem Turki
备注：Video results: https://artifixer2026.github.io/

编码器(2篇)

【1】Efficient Decoder Scaling Strategy for Neural Routing Solvers
标题：神经路由求解器的高效解码器缩放策略
链接：https://arxiv.org/abs/2603.00430

作者：Qing Luo,Fu Luo,Ke Li,Zhenkun Wang

【2】Data-driven Synthesis of Magnetic Resonance Spectroscopy Data using a Variational Autoencoder
标题：使用变分自动编码器的磁共振波谱数据的数据驱动合成
链接：https://arxiv.org/abs/2603.00736

作者：Dennis M. J. van de Sande,Julian P. Merkofer,Sina Amirrajab,Mitko Veta,Gerhard S. Drenthen,Jacobus F. A. Jansen,Marcel Breeuwer

优化|敛散性(21篇)

【1】Near-Optimal Regret for KL-Regularized Multi-Armed Bandits
标题：KL正规化多臂盗贼的近最佳遗憾
链接：https://arxiv.org/abs/2603.02155

作者：Kaixuan Ji,Qingyue Zhao,Heyang Zhao,Qiwei Di,Quanquan Gu

【2】Accelerating PDE Surrogates via RL-Guided Mesh Optimization
标题：通过RL引导网格优化加速PDL代理
链接：https://arxiv.org/abs/2603.02066

作者：Yang Meng,Ruoxi Jiang,Zhuokai Zhao,Chong Liu,Rebecca Willett,Yuxin Chen
备注：Accepted at AISTATS 2026

【3】Causal Circuit Tracing Reveals Distinct Computational Architectures in Single-Cell Foundation Models: Inhibitory Dominance, Biological Coherence, and Cross-Model Convergence
标题：因果电路追踪揭示单细胞基础模型中独特的计算架构：抑制性优势、生物一致性和跨模型收敛
链接：https://arxiv.org/abs/2603.01752

作者：Ihor Kendiukhov

【4】LFPO: Likelihood-Free Policy Optimization for Masked Diffusion Models
标题：LFPO：掩蔽扩散模型的无可能政策优化
链接：https://arxiv.org/abs/2603.01563

作者：Chenxing Wei,Jiazhen Kang,Hong Wang,Jianqing Zhang,Hao Jiang,Xiaolong Xu,Ningyuan Sun,Ying He,F. Richard Yu,Yao Shu,Bo Jiang

【5】Training Dynamics of Softmax Self-Attention: Fast Global Convergence via Preconditioning
标题：Softmax自我注意力的训练动力学：通过预处理实现快速全球收敛
链接：https://arxiv.org/abs/2603.01514

作者：Gautam Goel,Mahdi Soltanolkotabi,Peter Bartlett

【6】Randomized Kiring Believer for Parallel Bayesian Optimization with Regret Bounds
标题：具有遗憾界的并行Bayesian优化的随机激励相信者
链接：https://arxiv.org/abs/2603.01470

作者：Shuhei Sugiura,Ichiro Takeuchi,Shion Takeno

【7】Relatively Smart: A New Approach for Instance-Optimal Learning
标题：相对智能：实例最优学习的新方法
链接：https://arxiv.org/abs/2603.01346

作者：Shaddin Dughmi,Alireza F. Pour

【8】Provable and Practical In-Context Policy Optimization for Self-Improvement
标题：可证明且实用的上下文政策优化以实现自我改进
链接：https://arxiv.org/abs/2603.01335

作者：Tianrun Yu,Yuxiao Yang,Zhaoyang Wang,Kaixiang Zhao,Porter Jenkins,Xuchao Zhang,Chetan Bansal,Huaxiu Yao,Weitong Zhang
备注：34 pages, 8 tables, 4 figures, Accepted by ICLR 2026

【9】Demystifying Group Relative Policy Optimization: Its Policy Gradient is a U-Statistic
标题：揭开群体相对政策优化的神秘面纱：其政策梯度是U型统计数据
链接：https://arxiv.org/abs/2603.01162

作者：Hongyi Zhou,Kai Ye,Erhan Xu,Jin Zhu,Shijin Gong,Chengchun Shi
备注：32 pages, 4 figures

【10】A Decomposition Framework for Certifiably Optimal Orthogonal Sparse PCA
标题：可证明最优的垂直稀疏PCA的分解框架
链接：https://arxiv.org/abs/2603.01144

作者：Difei Cheng,Qiao Hu
备注：14 pages; 12 figures

【11】Stabilizing Policy Optimization via Logits Convexity
标题：通过Logits凸性稳定政策优化
链接：https://arxiv.org/abs/2603.00963

作者：Hongzhan Chen,Tao Yang,Yuhua Zhu,Shiping Gao,Xiaojun Quan,Ting Yao

【12】A short tour of operator learning theory: Convergence rates, statistical limits, and open questions
标题：操作员学习理论的简短游览：收敛率、统计限制和开放性问题
链接：https://arxiv.org/abs/2603.00819

作者：Simone Brugiapaglia,Nicola Rares Franco,Nicholas H. Nelsen
备注：12 pages

【13】Exact and Asymptotically Complete Robust Verifications of Neural Networks via Quantum Optimization
标题：基于量子优化的神经网络的精确渐近完备鲁棒性验证
链接：https://arxiv.org/abs/2603.00408

作者：Wenxin Li,Wenchao Liu,Chuan Wang,Qi Gao,Yin Ma,Hai Wei,Kai Wen

【14】Knowledge-guided generative surrogate modeling for high-dimensional design optimization under scarce data
标题：知识引导的生成式代理建模，用于稀缺数据下的多维设计优化
链接：https://arxiv.org/abs/2603.00052

作者：Bingran Wang,Seongha Jeong,Sebastiaan P. C. van Schie,Dongyeon Han,Jaeho Min,John T. Hwang

【15】Transit Network Design with Two-Level Demand Uncertainties: A Machine Learning and Contextual Stochastic Optimization Framework
标题：基于机器学习和上下文随机优化的两级需求不确定公交网络设计
链接：https://arxiv.org/abs/2603.00010

作者：Hongzhao Guan,Beste Basciftci,Pascal Van Hentenryck

【16】Quantitative Convergence of Wasserstein Gradient Flows of Kernel Mean Discrepancies
标题：核平均偏差Wasserstein梯度流的定量收敛
链接：https://arxiv.org/abs/2603.01977

作者：Lénaïc Chizat,Maria Colombo,Roberto Colombo,Xavier Fernández-Real

【17】GPU-friendly and Linearly Convergent First-order Methods for Certifying Optimal $k$-sparse GLMs
标题：用于认证最优$k$-稀疏GLM的GOP友好且线性收敛的一阶方法
链接：https://arxiv.org/abs/2603.01306

作者：Jiachang Liu,Andrea Lodi,Soroosh Shafiee
备注：Extended version of the ICML 2025 conference paper

【18】Non-Rectangular Average-Reward Robust MDPs: Non-Rectangular Average-Reward Robust MDPs:Optimal Policies and Their Transient Values
标题：非矩形平均报酬鲁棒MDP：最优策略及其瞬时值
链接：https://arxiv.org/abs/2603.00945

作者：Shengbo wang,Nian Si

【19】Time-Aware Latent Space Bayesian Optimization
标题：时间感知潜在空间Bayesian优化
链接：https://arxiv.org/abs/2603.00935

作者：Tuan A. Vu,Julien Martinelli,Harri Lähdesmäki

【20】Bilevel Optimization with Lower-Level Uniform Convexity: Theory and Algorithm
标题：具有低级一致凸度的二层优化：理论与算法
链接：https://arxiv.org/abs/2603.00027

作者：Yuman Wu,Xiaochuan Gong,Jie Hao,Mingrui Liu

【21】Riemannian Dueling Optimization
标题：雷曼决斗优化
链接：https://arxiv.org/abs/2603.00023

作者：Yuxuan Ren,Abhishek Roy,Shiqian Ma

预测|估计(12篇)

【1】MAC: A Conversion Rate Prediction Benchmark Featuring Labels Under Multiple Attribution Mechanisms
标题：MAC：一个基于多归因机制标签的转化率预测基准
链接：https://arxiv.org/abs/2603.02184

作者：Jinqi Wu,Sishuo Chen,Zhangming Chan,Yong Bai,Lei Zhang,Sheng Chen,Chenghuan Hou,Xiang-Rong Sheng,Han Zhu,Jian Xu,Bo Zheng,Chaoyou Fu
备注：Code and data available at https://github.com/alimama-tech/PyMAL

【2】Accelerating Single-Pass SGD for Generalized Linear Prediction
标题：广义线性预测的加速单程SGD算法
链接：https://arxiv.org/abs/2603.01951

作者：Qian Chen,Shihong Ding,Cong Fang
备注：50 pages

【3】GlassMol: Interpretable Molecular Property Prediction with Concept Bottleneck Models
标题：GlassMol：使用概念瓶颈模型的可解释分子性质预测
链接：https://arxiv.org/abs/2603.01274

作者：Oscar Rivera,Ziqing Wang,Matthieu Dagommer,Abhishek Pandey,Kaize Ding

【4】MultiPUFFIN: A Multimodal Domain-Constrained Foundation Model for Molecular Property Prediction of Small Molecules
标题：MultiPUFFIN：用于小分子分子性质预测的多峰域约束基础模型
链接：https://arxiv.org/abs/2603.00857

作者：Idelfonso B. R. Nogueira,Carine M. Rebelloa,Mumin Enis Leblebici,Erick Giovani Sperandio Nascimento

【5】Bi-cLSTM: Residual-Corrected Bidirectional LSTM for Aero-Engine RUL Estimation
标题：Bi-cLSTM：用于航空发动机RUL估计的剩余修正双向LSTM
链接：https://arxiv.org/abs/2603.00745

作者：Rafi Hassan Chowdhury,Nabil Daiyan,Faria Ahmed,Md Redwan Iqbal,Morsalin Sheikh

【6】ResGene-T: A Tensor-Based Residual Network Approach for Genomic Prediction
标题：ResGene-T：一种基于张量的剩余网络方法用于基因组预测
链接：https://arxiv.org/abs/2603.00744

作者：Kuldeep Pathak,Kapil Ahuja,Eric de Sturler
备注：Double column 11 Pages, 6 Figure, and 8 Tables

【7】Retrodictive Forecasting: A Proof-of-Concept for Exploiting Temporal Asymmetry in Time Series Prediction
标题：回溯预测：在时间序列预测中利用时间不对称性的概念证明
链接：https://arxiv.org/abs/2603.00636

作者：Cedric Damour
备注：27 pages, 13 figures, 5 tables, Code available at https://github.com/cdamour/retrodictive-forecasting (Zenodo: https://doi.org/10.5281/zenodo.18803446)

【8】Phys-Diff: A Physics-Inspired Latent Diffusion Model for Tropical Cyclone Forecasting
标题：Phys-Diff：用于热带气旋预测的物理启发潜在扩散模型
链接：https://arxiv.org/abs/2603.00521

作者：Lei Liu,Xiaoning Yu,Kang Chen,Jiahui Huang,Tengyuan Liu,Hongwei Zhao,Bin Li
备注：5 pages, 4 figures. Accepted to IEEE ICASSP 2026

【9】Diagnostics for Individual-Level Prediction Instability in Machine Learning for Healthcare
标题：医疗保健机器学习中个体水平预测不稳定性的诊断
链接：https://arxiv.org/abs/2603.00192

作者：Elizabeth W. Miller,Jeffrey D. Blume

【10】From GEV to ResLogit: Spatially Correlated Discrete Choice Models for Pedestrian Movement Prediction
标题：从GEV到ResLogit：用于行人运动预测的空间相关离散选择模型
链接：https://arxiv.org/abs/2603.01325

作者：Rulla Al-Haideri,Bilal Farooq

【11】A Monte Carlo estimator of flow fields for sampling and noise problems
标题：采样和噪音问题流场的蒙特卡罗估计
链接：https://arxiv.org/abs/2603.00252

作者：Michael S. Albergo,Gurtej Kanwar
备注：10 pages, 5 figures. Proceedings of the 42nd International Symposium on Lattice Field Theory (LATTICE2025)

【12】RSS map-assisted MIMO channel estimation in the upper mid-band under pilot constraints
标题：导频约束下上中带的RSS地图辅助的MMO信道估计
链接：https://arxiv.org/abs/2603.00112

作者：Alireza Javid,Nuria González-Prelcic
备注：Submitted to TMLCN

其他神经网络|深度学习|模型|建模(57篇)

【1】Frontier Models Can Take Actions at Low Probabilities
标题：前沿模型可以在低概率下采取行动
链接：https://arxiv.org/abs/2603.02202

作者：Alex Serrano,Wen Xing,David Lindner,Erik Jenner

【2】Machine Learning (ML) library in Linux kernel
标题：Linux内核中的机器学习（ML）库
链接：https://arxiv.org/abs/2603.02145

作者：Viacheslav Dubeyko

【3】Robometer: Scaling General-Purpose Robotic Reward Models via Trajectory Comparisons
标题：Robmeter：通过轨迹比较扩展通用机器人奖励模型
链接：https://arxiv.org/abs/2603.02115

作者：Anthony Liang,Yigit Korkmaz,Jiahui Zhang,Minyoung Hwang,Abrar Anwar,Sidhant Kaushik,Aditya Shah,Alex S. Huang,Luke Zettlemoyer,Dieter Fox,Yu Xiang,Anqi Li,Andreea Bobu,Abhishek Gupta,Stephen Tu,Erdem Biyik,Jesse Zhang
备注：33 pages, 17 figures

【4】CausalWrap: Model-Agnostic Causal Constraint Wrappers for Tabular Synthetic Data
标题：CASYS Wrap：表格合成数据的模型不可知因果约束Wrap
链接：https://arxiv.org/abs/2603.02015

作者：Amir Asiaee,Zhuohui J. Liang,Chao Yan

【5】MatRIS: Toward Reliable and Efficient Pretrained Machine Learning Interaction Potentials
标题：MatRIS：迈向可靠有效的预训练机器学习交互潜力
链接：https://arxiv.org/abs/2603.02002

作者：Yuanchang Zhou,Siyu Hu,Xiangyu Zhang,Hongyu Wang,Guangming Tan,Weile Jia
备注：28 pages, 9 figures, 12 tables

【6】CoVAE: correlated multimodal generative modeling
标题：CoVAE：相关多模式生成建模
链接：https://arxiv.org/abs/2603.01965

作者：Federico Caretti,Guido Sanguinetti

【7】When Numbers Tell Half the Story: Human-Metric Alignment in Topic Model Evaluation
标题：当数字说明了一半：主题模型评估中的人与指标一致
链接：https://arxiv.org/abs/2603.01945

作者：Thibault Prouteau,Francis Lareau,Nicolas Dugué,Jean-Charles Lamirel,Christophe Malaterre

【8】Dream2Learn: Structured Generative Dreaming for Continual Learning
标题：Dream2Learn：持续学习的结构化生成性梦想
链接：https://arxiv.org/abs/2603.01935

作者：Salvatore Calcagno,Matteo Pennisi,Federica Proietto Salanitri,Amelia Sorrenti,Simone Palazzo,Concetto Spampinato,Giovanni Bellitto

【9】Bound Propagation meets Constraint Simplification: Improving Logic-based XAI for Neural Networks
标题：界限传播满足约束简化：改进神经网络基于逻辑的XAI
链接：https://arxiv.org/abs/2603.01923

作者：Ronaldo Gomes,Jairo Ribeiro,Luiz Queiroz,Thiago Alves Rocha
备注：Preprint version. For the final published version, see the DOI below

【10】Learning Shortest Paths with Generative Flow Networks
标题：基于生成流网络的最短路径学习
链接：https://arxiv.org/abs/2603.01786

作者：Nikita Morozov,Ian Maksimov,Daniil Tiapkin,Sergey Samsonov

【11】CHLU: The Causal Hamiltonian Learning Unit as a Symplectic Primitive for Deep Learning
标题：CHLU：因果汉密尔顿学习单元作为深度学习的辛基元
链接：https://arxiv.org/abs/2603.01768

作者：Pratik Jawahar,Maurizio Pierini
备注：Accepted as a short paper at ICLR 2026 (AI & PDE)

【12】DGNet: Discrete Green Networks for Data-Efficient Learning of Spatiotemporal PDEs
标题：DGNet：用于时空PDEs数据高效学习的离散绿色网络
链接：https://arxiv.org/abs/2603.01762

作者：Yingjie Tan,Quanming Yao,Yaqing Wang
备注：Accepted as a conference paper at ICLR 2026

【13】Modular Memory is the Key to Continual Learning Agents
标题：模块化存储器是持续学习代理的关键
链接：https://arxiv.org/abs/2603.01761

作者：Vaggelis Dorovatas,Malte Schwerin,Andrew D. Bagdanov,Lucas Caccia,Antonio Carta,Laurent Charlin,Barbara Hammer,Tyler L. Hayes,Timm Hess,Christopher Kanan,Dhireesha Kudithipudi,Xialei Liu,Vincenzo Lomonaco,Jorge Mendez-Mendez,Darshan Patil,Ameya Prabhu,Elisa Ricci,Tinne Tuytelaars,Gido M. van de Ven,Liyuan Wang,Joost van de Weijer,Jonghyun Choi,Martin Mundt,Rahaf Aljundi
备注：This work stems from discussions held at the Dagstuhl seminar on Continual Learning in the Era of Foundation Models (October 2025)

【14】Shape-Interpretable Visual Self-Modeling Enables Geometry-Aware Continuum Robot Control
标题：形状可解释的视觉自建模实现几何感知连续机器人控制
链接：https://arxiv.org/abs/2603.01751

作者：Peng Yu,Xin Wang,Ning Tan

【15】Discrete World Models via Regularization
标题：通过正规化的离散世界模型
链接：https://arxiv.org/abs/2603.01748

作者：Davide Bizzaro,Luciano Serafini

【16】TopoCurate:Modeling Interaction Topology for Tool-Use Agent Training
标题：TopoCurate：为工具使用代理训练建模交互布局
链接：https://arxiv.org/abs/2603.01714

作者：Jinluan Yang,Yuxin Liu,Zhengyu Chen,Chengcheng Han,Yueqing Sun,Qi Gu,Hui Su,Xunliang Cai,Fei Wu,Kun Kuang
备注：Under Review

【17】Security Risks in Machining Process Monitoring: Sequence-to-Sequence Learning for Reconstruction of CNC Axis Positions
标题：加工过程监控中的安全风险：用于重建NC轴位置的序列到序列学习
链接：https://arxiv.org/abs/2603.01702

作者：Lukas Krupp,Rickmar Stahlschmidt,Norbert Wehn
备注：Accepted for presentation at the 2026 IEEE Symposium on Artificial Intelligence for Instrumentation and Measurement (AI4IM 2026). Proceedings to be included in IEEE Xplore

【18】Randomized Neural Networks for Partial Differential Equation on Static and Evolving Surfaces
标题：静态和演变表面上偏方程的随机神经网络
链接：https://arxiv.org/abs/2603.01689

作者：Jingbo Sun,Fei Wang

【19】A Practical Guide to Streaming Continual Learning
标题：流媒体持续学习实用指南
链接：https://arxiv.org/abs/2603.01677

作者：Andrea Cossu,Federico Giannini,Giacomo Ziffer,Alessio Bernardo,Alexander Gepperth,Emanuele Della Valle,Barbara Hammer,Davide Bacciu

【20】Transform-Invariant Generative Ray Path Sampling for Efficient Radio Propagation Modeling
标题：用于高效无线电传播建模的变换不变生成射线路径采样
链接：https://arxiv.org/abs/2603.01655

作者：Jérome Eertmans,Enrico M. Vitucci,Vittorio Degli-Esposti,Nicola Di Cicco,Laurent Jacques,Claude Oestges
备注：submitted to npj Wireless Technology, 26 pages, 14 figures

【21】DeLo: Dual Decomposed Low-Rank Experts Collaboration for Continual Missing Modality Learning
标题：DeLo：双分解低级别专家合作，以实现连续缺失的情态学习
链接：https://arxiv.org/abs/2603.01632

作者：Xiwei Liu,Yulong Li,Feilong Tang,Imran Razzak

【22】KERV: Kinematic-Rectified Speculative Decoding for Embodied VLA Models
标题：KERV：针对预定VLA模型的运动学纠正推测解码
链接：https://arxiv.org/abs/2603.01581

作者：Zihao Zheng,Zhihao Mao,Maoliang Li,Jiayu Chen,Xinhao Sun,Zhaobo Zhang,Donggang Cao,Hong Mei,Xiang Chen
备注：This paper has been accepted by DAC 2026

【23】SEAnet: A Deep Learning Architecture for Data Series Similarity Search
标题：SEAnet：用于数据系列相似性搜索的深度学习架构
链接：https://arxiv.org/abs/2603.01448

作者：Qitong Wang,Themis Palpanas
备注：This paper was published in IEEE Transactions on Knowledge and Data Engineering (Volume: 35, Issue: 12, Page(s): 12972 - 12986, 01 December 2023). Date of Publication: 25 April 2023

【24】Tackling multiphysics problems via finite element-guided physics-informed operator learning
标题：通过有限元素引导的物理知识操作员学习解决多物理场问题
链接：https://arxiv.org/abs/2603.01420

作者：Yusuke Yamazaki,Reza Najian Asl,Markus Apel,Mayu Muramatsu,Shahed Rezaei

【25】Learn Hard Problems During RL with Reference Guided Fine-tuning
标题：通过参考引导微调在RL期间学习难问题
链接：https://arxiv.org/abs/2603.01223

作者：Yangzhen Wu,Shanda Li,Zixin Wen,Xin Zhou,Ameet Talwalkar,Yiming Yang,Wenhao Huang,Tianle Cai
备注：16 pages, 5 figures

【26】Agents Learn Their Runtime: Interpreter Persistence as Training-Time Semantics
标题：代理人学习他们的时间表：口译员持久性作为训练时的语义学
链接：https://arxiv.org/abs/2603.01209

作者：Victor May,Aaditya Salgarkar,Yishan Wang,Diganta Misra,Huu Nguyen
备注：Code: https://github.com/mrcabbage972/agents-learn-runtime

【27】Scaling of learning time for high dimensional inputs
标题：扩展多维输入的学习时间
链接：https://arxiv.org/abs/2603.01184

作者：Carlos Stein Brito
备注：14 pages, 5 figures

【28】Unified Vision-Language Modeling via Concept Space Alignment
标题：通过概念空间对齐实现统一的视觉语言建模
链接：https://arxiv.org/abs/2603.01096

作者：Yifu Qiu,Paul-Ambroise Duquenne,Holger Schwenk
备注：ICLR 2026

【29】A level-wise training scheme for learning neural multigrid smoothers with application to integral equations
标题：学习神经多重网格平滑器的分层训练方案及其应用于积分方程
链接：https://arxiv.org/abs/2603.01064

作者：Lingfeng Li,Yin King Chu,Raymond Chan,Justin Wan

【30】Compensation-free Machine Unlearning in Text-to-Image Diffusion Models by Eliminating the Mutual Information
标题：通过消除互信息实现文本到图像扩散模型中的无补偿机器去学习
链接：https://arxiv.org/abs/2603.00992

作者：Xinwen Cheng,Jingyuan Zhang,Zhehao Huang,Yingwen Wu,Xiaolin Huang

【31】SoberDSE: Sample-Efficient Design Space Exploration via Learning-Based Algorithm Selection
标题：SoberDSE：通过基于学习的算法选择进行样本高效的设计空间探索
链接：https://arxiv.org/abs/2603.00986

作者：Lei Xu,Shanshan Wang,Chenglong Xiao

【32】When Does Margin Clamping Affect Training Variance? Dataset-Dependent Effects in Contrastive Forward-Forward Learning
标题：保证金限制何时影响训练差异？对比前向学习中的数据集依赖效应
链接：https://arxiv.org/abs/2603.00951

作者：Joshua Steier
备注：17 pages, 2 figures, 15 tables, including appendices

【33】Wave-Attractor-Tree: A Hierarchical Binary Tree Reduction Architecture for Efficient Sequence Modeling
标题：波吸引器-树：一种用于高效序列建模的分层二元树约简架构
链接：https://arxiv.org/abs/2603.00812

作者：Igor Berezkin
备注：5 pages, 5 tables. Source code and benchmarks are available at [https://github.com/IgorBerezkin/WAT]

【34】General Proximal Flow Networks
标题：一般近端流网络
链接：https://arxiv.org/abs/2603.00751

作者：Alexander Strunk,Roland Assam

【35】Reward-Modulated Local Learning in Spiking Encoders: Controlled Benchmarks with STDP and Hybrid Rate Readouts
标题：峰值编码器中的奖励调节本地学习：具有STDP和混合速率读数的受控基准
链接：https://arxiv.org/abs/2603.00710

作者：Debjyoti Chakraborty
备注：10 pages, 5 figures. Submitted to IEEE Transactions on Neural Networks and Learning Systems

【36】IDER: IDempotent Experience Replay for Reliable Continual Learning
标题：IDER：等效体验重演，实现可靠的持续学习
链接：https://arxiv.org/abs/2603.00624

作者：Zhanwang Liu,Yuting Li,Haoyuan Gao,Yexin Li,Linghe Kong,Lichao Sun,Weiran Huang
备注：Accepted by ICLR 2026

【37】CMI-RewardBench: Evaluating Music Reward Models with Compositional Multimodal Instruction
标题：CMI-RewardBench：使用合成多模式教学评估音乐奖励模型
链接：https://arxiv.org/abs/2603.00610

作者：Yinghao Ma,Haiwen Xia,Hewei Gao,Weixiong Chen,Yuxin Ye,Yuchen Yang,Sungkyun Chang,Mingshuo Ding,Yizhi Li,Ruibin Yuan,Simon Dixon,Emmanouil Benetos

【38】Heterophily-Agnostic Hypergraph Neural Networks with Riemannian Local Exchanger
标题：具有Riemann局部交换器的异类不可知超图神经网络
链接：https://arxiv.org/abs/2603.00599

作者：Li Sun,Ming Zhang,Wenxin Jin,Zhongtian Sun,Zhenhao Huang,Hao Peng,Sen Su,Philip Yu
备注：Accepted by WWW'26, 12 pages

【39】Enhancing Molecular Property Predictions by Learning from Bond Modelling and Interactions
标题：通过学习键模型和相互作用增强分子性质预测
链接：https://arxiv.org/abs/2603.00568

作者：Yunqing Liu,Yi Zhou,Wenqi Fan
备注：Accepted to ICLR 2026

【40】Improving Full Waveform Inversion in Large Model Era
标题：大模型时代改善全波倒置
链接：https://arxiv.org/abs/2603.00377

作者：Yinan Feng,Peng Jin,Yuzhe Guo,Yinpeng Chen,Youzuo Lin

【41】KROM: Kernelized Reduced Order Modeling
标题：KROM：核化降阶建模
链接：https://arxiv.org/abs/2603.00360

作者：Aras Bacho,Jonghyeon Lee,Houman Owhadi

【42】Polynomial Surrogate Training for Differentiable Ternary Logic Gate Networks
标题：可微三值逻辑门网络的多项替代训练
链接：https://arxiv.org/abs/2603.00302

作者：Sai Sandeep Damera,Ryan Matheu,Aniruddh G. Puranic,John S. Baras
备注：28 pages, 13 figures. Submitted to 3rd International Conference on Neuro-Symbolic Systems (NeuS) 2026

【43】Scalable Gaussian process modeling of parametrized spatio-temporal fields
标题：参数化时空场的可扩展高斯过程建模
链接：https://arxiv.org/abs/2603.00290

作者：Srinath Dama,Prasanth B. Nair

【44】Task-Driven Subspace Decomposition for Knowledge Sharing and Isolation in LoRA-based Continual Learning
标题：基于LoRA的持续学习中知识共享和隔离的任务驱动子空间分解
链接：https://arxiv.org/abs/2603.00191

作者：Lingfeng He,De Cheng,Huaijie Wang,Xi Yang,Nannan Wang,Xinbo Gao
备注：preprint

【45】OSF: On Pre-training and Scaling of Sleep Foundation Models
标题：OSF：关于预训练和睡眠扩展基金会模型
链接：https://arxiv.org/abs/2603.00190

作者：Zitao Shuai,Zongzhe Xu,David Yang,Wei Wang,Yuzhe Yang

【46】Summer-22B: A Systematic Approach to Dataset Engineering and Training at Scale for Video Foundation Model
标题：Summer-22 B：视频基础模型大规模数据集工程和训练的系统方法
链接：https://arxiv.org/abs/2603.00173

作者：Simo Ryu,Chunghwan Han
备注：28 pages, 16 figures, 5 tables

【47】Wideband Power Amplifier Behavioral Modeling Using an Amplitude Conditioned LSTM
标题：使用幅度调节LSTM的宽带功率放大器行为建模
链接：https://arxiv.org/abs/2603.00101

作者：Abdelrahman Abdelsalam,You Fei
备注：7 Pages, 6 Figures

【48】Error Bounds for Physics-Informed Neural Networks in Fokker-Planck PDEs
标题：Fokker-Planck PDEs中物理信息神经网络的误差界
链接：https://arxiv.org/abs/2410.22371

作者：Chun-Wei Kong,Luca Laurenti,Jay McMahon,Morteza Lahijanian
备注：Accepted at Uncertainty in Artificial Intelligence (UAI) 2025

【49】Orchestrating Multimodal DNN Workloads in Wireless Neural Processing
标题：在无线神经处理中描述多峰DNN工作负载
链接：https://arxiv.org/abs/2603.02109

作者：Sai Xu,Kai-Kit Wong,Yanan Du,Hyundong Shin

【50】Deep Learning for Financial Time Series: A Large-Scale Benchmark of Risk-Adjusted Performance
标题：金融时间序列深度学习：风险调整后绩效的大规模基准
链接：https://arxiv.org/abs/2603.01820

作者：Adir Saly-Kaufmann,Kieran Wood,Jan Peter-Calliess,Stefan Zohren
备注：43 pages, 27 figures, 11 tables

【51】PhysFormer: A Physics-Embedded Generative Model for Physically Self-Consistent Spectral Synthesis
标题：PhysFormer：用于物理自相容光谱合成的物理嵌入生成模型
链接：https://arxiv.org/abs/2603.01459

作者：Siqi Wang,Mengmeng Zhang,Yude Bu,Chaozhou Mou
备注：9 pages, 5 figures

【52】VoxKnesset: A Large-Scale Longitudinal Hebrew Speech Dataset for Aging Speaker Modeling
标题：VoxKnesset：用于老化说话人建模的大规模纵向希伯来语语音数据集
链接：https://arxiv.org/abs/2603.01270

作者：Yanir Marmor,Arad Zulti,David Krongauz,Adam Gabet,Yoad Snapir,Yair Lifshitz,Eran Segal
备注：4 pages, 5 figures, 2 tables

【53】Grokking as a Phase Transition between Competing Basins: a Singular Learning Theory Approach
标题：Grokking作为竞争盆地之间的阶段过渡：奇异学习理论方法
链接：https://arxiv.org/abs/2603.01192

作者：Ben Cullen,Sergio Estan-Ruiz,Riya Danait,Jiayi Li

【54】Structure-preserving Randomized Neural Networks for Incompressible Magnetohydrodynamics Equations
标题：不可压缩磁流体动力学方程的保结构随机神经网络
链接：https://arxiv.org/abs/2603.01102

作者：Yunlong Li,Fei Wang,Lingxiao Li

【55】Learning with the Nash-Sutcliffe loss
标题：从纳什-萨克利夫的失利中学习
链接：https://arxiv.org/abs/2603.00968

作者：Hristos Tyralis,Georgia Papacharalampous
备注：77 pages, 4 figures, 6 tables

【56】Using Artificial Neural Networks to Predict Claim Duration in a Work Injury Compensation Environment
标题：使用人工神经网络预测劳资纠纷赔偿环境中的索赔持续时间
链接：https://arxiv.org/abs/2603.00100

作者：Anthony Almudevar
备注：8 pages; 9 figures; 6 tables

【57】The minimal width of universal $p$-adic ReLU neural networks
标题：泛$p$-adic ReLU神经网络的最小宽度
链接：https://arxiv.org/abs/2603.00064

作者：Sándor Z. Kiss,Ambrus Pál
备注：16 pages, comments welcome!

其他(88篇)

【1】Conformal Policy Control
标题：保形政策控制
链接：https://arxiv.org/abs/2603.02196

作者：Drew Prinster,Clara Fannjiang,Ji Won Park,Kyunghyun Cho,Anqi Liu,Suchi Saria,Samuel Stanton

【2】From Leaderboard to Deployment: Code Quality Challenges in AV Perception Repositories
标题：从排行榜到部署：AV感知存储库中的代码质量挑战
链接：https://arxiv.org/abs/2603.02194

作者：Mateus Karvat,Bram Adams,Sidney Givigi

【3】Multi-Head Low-Rank Attention
标题：多头低级别注意力
链接：https://arxiv.org/abs/2603.02188

作者：Songtao Liu,Hongwu Peng,Zhiwei Zhang,Zhengyu Chen,Yue Guo
备注：Accepted by ICLR 2026

【4】Reservoir Subspace Injection for Online ICA under Top-n Whitening
标题：Top-n白化下在线ICA的储层子空间注入
链接：https://arxiv.org/abs/2603.02178

作者：Wenjun Xiao,Yuda Bi,Vince D Calhoun

【5】De-paradox Tree: Breaking Down Simpson's Paradox via A Kernel-Based Partition Algorithm
标题：去悖论树：通过基于核的划分算法分解辛普森悖论
链接：https://arxiv.org/abs/2603.02174

作者：Xian Teng,Yu-Ru Lin

【6】SageBwd: A Trainable Low-bit Attention
标题：SageBWD：可训练的低位注意力
链接：https://arxiv.org/abs/2603.02170

作者：Jintao Zhang,Marco Chen,Haoxu Wang,Kai Jiang,Ion Stoica,Joseph E. Gonzalez,Jianfei Chen,Jun Zhu

【7】Stochastic Multi-Armed Bandits with Limited Control Variates
标题：控制变量有限的随机多臂强盗
链接：https://arxiv.org/abs/2603.02100

作者：Arun Verma,Manjesh Kumar Hanawal,Arun Rajkumar
备注：Accepted at COMSNETS 2026

【8】Adam Converges Without Any Modification On Update Rules
标题：Adam在更新规则上没有任何修改的情况下聚合
链接：https://arxiv.org/abs/2603.02092

作者：Yushun Zhang,Bingran Li,Congliang Chen,Zhi-Quan Luo,Ruoyu Sun
备注：66 pages

【9】From Pixels to Patches: Pooling Strategies for Earth Embeddings
标题：从像素到补丁：地球嵌入的池策略
链接：https://arxiv.org/abs/2603.02080

作者：Isaac Corley,Caleb Robinson,Inbal Becker-Reshef,Juan M. Lavista Ferres

【10】Scaling Laws of SignSGD in Linear Regression: When Does It Outperform SGD?
标题：线性回归中SignSingapore的比例定律：什么时候它优于Singapore？
链接：https://arxiv.org/abs/2603.02069

作者：Jihwan Kim,Dogyoon Song,Chulhee Yun
备注：Accepted at ICLR 2026, 89 pages, 25 figures

【11】Never Saddle for Reparameterized Steepest Descent as Mirror Flow
标题：永远不要为镜像流重新参数化的最陡下降马鞍
链接：https://arxiv.org/abs/2603.02064

作者：Tom Jacobs,Chao Zhou,Rebekka Burkholz

【12】Strategic Advice in the Age of Personal AI
标题：个人人工智能时代的战略建议
链接：https://arxiv.org/abs/2603.02055

作者：Yueyang Liu,Wichinpong Park Sinchaisri

【13】Rich Insights from Cheap Signals: Efficient Evaluations via Tensor Factorization
标题：来自廉价信号的丰富见解：通过张量分解进行高效评估
链接：https://arxiv.org/abs/2603.02029

作者：Felipe Maia Polo,Aida Nematzadeh,Virginia Aglietti,Adam Fisch,Isabela Albuquerque

【14】Latent attention on masked patches for flow reconstruction
标题：对血流重建的掩蔽补丁的潜在关注
链接：https://arxiv.org/abs/2603.02028

作者：Ben Eze,Luca Magri,Andrea Nóvoa
备注：8 pages, 5 figures, submitted to ICCS (International Conference on Computational Science) 2026

【15】Selection as Power: Constrained Reinforcement for Bounded Decision Authority
标题：选择作为权力：有限决策权威的约束强化
链接：https://arxiv.org/abs/2603.02019

作者：Jose Manuel de la Chica Rodriguez,Juan Manuel Vera Díaz

【16】Accurate, private, secure, federated U-statistics with higher degree
标题：准确、私密、安全、更高程度的联合U统计数据
链接：https://arxiv.org/abs/2603.01986

作者：Quentin Sinh,Jan Ramon

【17】Intrinsic Task Symmetry Drives Generalization in Algorithmic Tasks
标题：内在任务对称性推动数学任务的概括
链接：https://arxiv.org/abs/2603.01968

作者：Hyeonbin Hwang,Yeachan Park
备注：Preprint

【18】TiledAttention: a CUDA Tile SDPA Kernel for PyTorch
标题：TiledAttention：适用于PyTorch的CUDA磁贴SDPA内核
链接：https://arxiv.org/abs/2603.01960

作者：Taimur Khan

【19】The Expressive Limits of Diagonal SSMs for State-Tracking
标题：用于状态跟踪的对角RSM的表达极限
链接：https://arxiv.org/abs/2603.01959

作者：Mehran Shakerinava,Behnoush Khavari,Siamak Ravanbakhsh,Sarath Chandar
备注：18 pages, 5 figures, 4 tables. Accepted at ICLR 2026

【20】Probabilistic Retrofitting of Learned Simulators
标题：学习模拟器的可能性改造
链接：https://arxiv.org/abs/2603.01949

作者：Cristiana Diaconu,Miles Cranmer,Richard E. Turner,Tanya Marwah,Payel Mukhopadhyay
备注：Code provided at https://github.com/cddcam/lola_crps

【21】Efficient RLVR Training via Weighted Mutual Information Data Selection
标题：通过加权互信息数据选择进行高效的WLVR训练
链接：https://arxiv.org/abs/2603.01907

作者：Xinyu Zhou,Boyu Zhu,Haotian Zhang,Huiming Wang,Zhijiang Guo
备注：15 Pages

【22】Diagnosing Generalization Failures from Representational Geometry Markers
标题：从具象几何标记诊断概括失败
链接：https://arxiv.org/abs/2603.01879

作者：Chi-Ning Chou,Artem Kirsanov,Yao-Yuan Yang,SueYeon Chung
备注：Published in the International Conference on Learning Representations (ICLR), 2026

【23】Tide: A Customisable Dataset Generator for Anti-Money Laundering Research
标题：潮汐：用于反洗钱研究的可定制数据集生成器
链接：https://arxiv.org/abs/2603.01863

作者：Montijn van den Beukel,Jože Martin Rožanec,Ana-Lucia Varbanescu
备注：Synthetic AML transaction datasets (Tide, HI and LI variants) are available at https://doi.org/10.5281/zenodo.18804069

【24】Constrained Particle Seeking: Solving Diffusion Inverse Problems with Just Forward Passes
标题：约束粒子搜索：用正向传递求解扩散反问题
链接：https://arxiv.org/abs/2603.01837

作者：Hongkun Dou,Zike Chen,Zeyu Li,Hongjue Li,Lijun Yang,Yue Deng
备注：Accepted by AAAI 2026

【25】OpenAutoNLU: Open Source AutoML Library for NLU
标题：OpenAutoNLU：用于NLU的开源AutoML库
链接：https://arxiv.org/abs/2603.01824

作者：Grigory Arshinov,Aleksandr Boriskin,Sergey Senichev,Ayaz Zaripov,Daria Galimzianova,Daniil Karpov,Leonid Sanochkin

【26】Phase-Type Variational Autoencoders for Heavy-Tailed Data
标题：用于重尾数据的相型变分自动编码器
链接：https://arxiv.org/abs/2603.01800

作者：Abdelhakim Ziani,András Horváth,Paolo Ballarini

【27】Practical Deep Heteroskedastic Regression
标题：实用深度异方差回归
链接：https://arxiv.org/abs/2603.01750

作者：Mikkel Jordahn,Jonas Vestergaard Jensen,James Harrison,Michael Riis Andersen,Mikkel N. Schmidt

【28】Legal RAG Bench: an end-to-end benchmark for legal RAG
标题：Legal RAG Bench：Legal RAG的端到端基准
链接：https://arxiv.org/abs/2603.01710

作者：Abdur-Rahman Butler,Umar Butler
备注：13 pages, 3 figures, 4 tables

【29】Search Multilayer Perceptron-Based Fusion for Efficient and Accurate Siamese Tracking
标题：搜索多层基于感知器的融合实现高效准确的连体追踪
链接：https://arxiv.org/abs/2603.01706

作者：Tianqi Shen,Huakao Lin,Ning An
备注：23 pages, 12 figures, 7 tables. This work was completed in 2024 and accepted for publication in IEEE TCDS (2026)

【30】Boosting Entropy with Bell Box Quantization
标题：用钟盒量化提高信息量
链接：https://arxiv.org/abs/2603.01599

作者：Ningfeng Yang,Tor M. Aamodt
备注：Published as a conference paper at ICLR 2026

【31】FAST-DIPS: Adjoint-Free Analytic Steps and Hard-Constrained Likelihood Correction for Diffusion-Prior Inverse Problems
标题：FAST-DIPS：扩散先验反问题的无伴随分析步骤和硬约束Likewise修正
链接：https://arxiv.org/abs/2603.01591

作者：Minwoo Kim,Seunghyeok Shin,Hongki Lim

【32】Rate-Distortion Signatures of Generalization and Information Trade-offs
标题：概括和信息权衡的速率失真签名
链接：https://arxiv.org/abs/2603.01568

作者：Leyla Roksan Caglar,Pedro A. M. Mediano,Baihan Lin

【33】State-Action Inpainting Diffuser for Continuous Control with Delay
标题：用于延迟连续控制的状态动作修复扩散器
链接：https://arxiv.org/abs/2603.01553

作者：Dongqi Han,Wei Wang,Enze Zhang,Dongsheng Li

【34】Reconstructing Content via Collaborative Attention to Improve Multimodal Embedding Quality
标题：通过协作关注重建内容以提高多模式嵌入质量
链接：https://arxiv.org/abs/2603.01471

作者：Jiahan Chen,Da Li,Hengran Zhang,Yinqiong Cai,Lixin Su,Jiafeng Guo,Daiting Shi,Dawei Yin,Keping Bi

【35】Autoregressive Synthesis of Sparse and Semi-Structured Mixed-Type Data
标题：稀疏和半结构化混合型数据的自回归合成
链接：https://arxiv.org/abs/2603.01444

作者：Thomas Rückstieß,Robin Vujanic
备注：Under Submission

【36】One Operator to Rule Them All? On Boundary-Indexed Operator Families in Neural PDE Solvers
标题：一个操作员来统治他们所有人？关于神经方程求解器中的边界索引运算符族
链接：https://arxiv.org/abs/2603.01406

作者：Lennon J. Shikhman
备注：Accepted to the ICLR 2026 Workshop on AI & PDEs. 10 pages, 5 figures

【37】Causal Neural Probabilistic Circuits
标题：因果神经概率回路
链接：https://arxiv.org/abs/2603.01372

作者：Weixin Chen,Han Zhao

【38】DUEL: Exact Likelihood for Masked Diffusion via Deterministic Unmasking
标题：决斗：通过确定性揭开掩盖扩散的确切可能性
链接：https://arxiv.org/abs/2603.01367

作者：Gilad Turok,Chris De Sa,Volodymyr Kuleshov
备注：22 pages, 5 figures 8 tables

【39】Align and Filter: Improving Performance in Asynchronous On-Policy RL
标题：对齐和过滤：提高同步按策略RL中的性能
链接：https://arxiv.org/abs/2603.01365

作者：Homayoun Honari,Roger Creus Castanyer,Michael Przystupa,Michael Noukhovitch,Pablo Samuel Castro,Glen Berseth

【40】Nonconvex Latent Optimally Partitioned Block-Sparse Recovery via Log-Sum and Minimax Concave Penalties
标题：通过Log-Sum和Minimax凹凸罚分的非凸潜在最优分区块稀疏恢复
链接：https://arxiv.org/abs/2603.01304

作者：Takanobu Furuhashi,Hiroki Kuroda,Masahiro Yukawa,Qibin Zhao,Hidekata Hontani,Tatsuya Yokota
备注：13 pages, 11 figures

【41】Attention Smoothing Is All You Need For Unlearning
标题：注意力平滑是您忘记学习所需的一切
链接：https://arxiv.org/abs/2603.01285

作者：Saleh Zare Zade,Xiangyu Zhou,Sijia Liu,Dongxiao Zhu
备注：Accepted by ICLR 2026

【42】Beyond Reward: A Bounded Measure of Agent Environment Coupling
标题：超越奖励：代理环境耦合的有限衡量标准
链接：https://arxiv.org/abs/2603.01283

作者：Wael Hafez,Cameron Reid,Amit Nazeri
备注：8 pages, 2 figures

【43】Can AI Agents Agree?
标题：人工智能代理可以同意吗？
链接：https://arxiv.org/abs/2603.01213

作者：Frédéric Berdoz,Leonardo Rugli,Roger Wattenhofer

【44】Subliminal Signals in Preference Labels
标题：偏好标签中的阈下信号
链接：https://arxiv.org/abs/2603.01204

作者：Isotta Magistrali,Frédéric Berdoz,Sam Dauncey,Roger Wattenhofer

【45】PARWiS: Winner determination under shoestring budgets using active pairwise comparisons
标题：PARWiS：使用主动成对比较在有限预算下确定获胜者
链接：https://arxiv.org/abs/2603.01171

作者：Shailendra Bhandari
备注：12 pages

【46】Alien Science: Sampling Coherent but Cognitively Unavailable Research Directions from Idea Atoms
标题：外星科学：从创意原子中抽样连贯但认知上不可用的研究方向
链接：https://arxiv.org/abs/2603.01092

作者：Alejandro H. Artiles,Martin Weiss,Levin Brinkmann,Anirudh Goyal,Nasim Rahaman
备注：Published at the ICLR 2026 Post-AGI Science and Society Workshop

【47】CARD: Towards Conditional Design of Multi-agent Topological Structures
标题：CARD：走向多智能体布局结构的条件设计
链接：https://arxiv.org/abs/2603.01089

作者：Tongtong Wu,Yanming Li,Ziye Tang,Chen Jiang,Linhao Luo,Guilin Qi,Shirui Pan,Gholamreza Haffari
备注：Accepted to ICLR 2026

【48】Evaluating GFlowNet from partial episodes for stable and flexible policy-based training
标题：从部分事件评估GFlowNet，以实现稳定和灵活的基于策略的训练
链接：https://arxiv.org/abs/2603.01047

作者：Puhua Niu,Shili Wu,Xiaoning Qian
备注：Accepted by ICLR 2026

【49】Reparameterized Tensor Ring Functional Decomposition for Multi-Dimensional Data Recovery
标题：用于多维数据恢复的重新参数化张量环函数分解
链接：https://arxiv.org/abs/2603.01034

作者：Yangyang Xu,Junbo Ke,You-Wei Wen,Chao Wang
备注：22 pages, 18 figures, 12 tables. Accepted by CVPR 2026

【50】Feature-Weighted Maximum Representative Subsampling
标题：资源加权最大代表性二次抽样
链接：https://arxiv.org/abs/2603.01013

作者：Tony Hauptmann,Stefan Kramer

【51】Evaluating AI Grading on Real-World Handwritten College Mathematics: A Large-Scale Study Toward a Benchmark
标题：评估现实世界手写大学数学的人工智能评分：迈向基准的大规模研究
链接：https://arxiv.org/abs/2603.00895

作者：Zhiqi Yu,Xingping Liu,Haobin Mao,Mingshuo Liu,Long Chen,Jack Xin,Yifeng Yu

【52】Active Flow Matching
标题：主动流量匹配
链接：https://arxiv.org/abs/2603.00877

作者：Yashvir S. Grewal,Daniel M. Steinberg,Thang D. Bui,Cheng Soon Ong,Edwin V. Bonilla

【53】Neural Latent Arbitrary Lagrangian-Eulerian Grids for Fluid-Solid Interaction
标题：流固相互作用的神经潜伏任意拉格朗日-欧拉网格
链接：https://arxiv.org/abs/2603.00792

作者：Shilong Tao,Zhe Feng,Shaohan Chen,Weichen Zhang,Zhanxing Zhu,Yunhuai Liu
备注：Proceedings of the 14th International Conference on Learning Representations

【54】Identifying the Geographic Foci of US Local News
标题：确定美国地方新闻的地理焦点
链接：https://arxiv.org/abs/2603.00787

作者：Gangani Ariyarathne,Isuru Ariyarathne,Greatness Emmanuel-King,Kate Lawal,Alexander C. Nwala
备注：This is a research paper accepted to the 18th ACM Web Science Conference 2026

【55】To Use or not to Use Muon: How Simplicity Bias in Optimizers Matters
标题：使用或不使用μ子：优化器中的简单性偏差如何重要
链接：https://arxiv.org/abs/2603.00742

作者：Sara Dragutinović,Rajesh Ranganath

【56】Frozen Policy Iteration: Computationally Efficient RL under Linear $Q^π$ Realizability for Deterministic Dynamics
标题：冻结策略迭代：线性$Q & pi $下的计算高效RL确定性动力学的可实现性
链接：https://arxiv.org/abs/2603.00716

作者：Yijing Ke,Zihan Zhang,Ruosong Wang

【57】Unlearning Evaluation through Subset Statistical Independence
标题：通过子集统计独立性放弃评估
链接：https://arxiv.org/abs/2603.00587

作者：Chenhao Zhang,Muxing Li,Feng Liu,Weitong Chen,Miao Xu
备注：21 pages, 6 figures, to appear at ICLR 2026

【58】Bridge Matching Sampler: Scalable Sampling via Generalized Fixed-Point Diffusion Matching
标题：桥匹配采样器：通过广义定点扩散匹配的可扩展采样
链接：https://arxiv.org/abs/2603.00530

作者：Denis Blessing,Lorenz Richter,Julius Berner,Egor Malitskiy,Gerhard Neumann
备注：Preprint

【59】A Polynomial-Time Axiomatic Alternative to SHAP for Feature Attribution
标题：特征归因SHAP的一种多元时间公理替代方案
链接：https://arxiv.org/abs/2603.00496

作者：Kazuhiro Hiraki,Shinichi Ishihara,Takumi Kongo,Junnosuke Shino
备注：28 pages, 4 figures, 2 tables. Code will be released

【60】Heaviside Low-Rank Support Matrix Machine
标题：Heaviside低等级支持矩阵机
链接：https://arxiv.org/abs/2603.00491

作者：Xianchao Xiu,Shenghao Sun,Xinrong Li,Jiyuan Tao

【61】Rooted Absorbed Prefix Trajectory Balance with Submodular Replay for GFlowNet Training
标题：GFlowNet训练的有根吸收的前置轨迹平衡，带有子模块重播
链接：https://arxiv.org/abs/2603.00454

作者：Xi Wang,Wenbo Lu,Shengjie Wang

【62】ROKA: Robust Knowledge Unlearning against Adversaries
标题：ROKA：强大的知识消除对手的学习
链接：https://arxiv.org/abs/2603.00436

作者：Jinmyeong Shin,Joshua Tapia,Nicholas Ferreira,Gabriel Diaz,Moayed Daneshyari,Hyeran Jeon

【63】Weight Updates as Activation Shifts: A Principled Framework for Steering
标题：随着激活变化，体重更新：指导的原则框架
链接：https://arxiv.org/abs/2603.00425

作者：Dyah Adila,John Cooper,Alexander Yun,Avi Trost,Frederic Sala

【64】Physics-Aware Learnability: From Set-Theoretic Independence to Operational Constraints
标题：物理感知的可学习性：从集合论独立性到操作约束
链接：https://arxiv.org/abs/2603.00417

作者：Jeongho Bang,Kyoungho Cho
备注：31 pages, 4 figures (Main Text + Supplementary Information) / Comment welcome

【65】TENG-BC: Unified Time-Evolving Natural Gradient for Neural PDE Solvers with General Boundary Conditions
标题：TENG-BC：具有一般边界条件的神经PDL求解器的统一时间演变自然梯度
链接：https://arxiv.org/abs/2603.00397

作者：Hongjie Jiang,Di Luo

【66】Aurchestra: Fine-Grained, Real-Time Soundscape Control on Resource-Constrained Hearables
标题：Aurchestra：对资源受限的听力进行细粒度、实时音景控制
链接：https://arxiv.org/abs/2603.00395

作者：Seunghyun Oh,Malek Itani,Aseem Gauri,Shyamnath Gollakota
备注：15 pages, 11 figures, 4 tables, submitted to ACM MobiSys 2026

【67】Acoustic Sensing for Universal Jamming Grippers
标题：通用干扰钳的声学传感
链接：https://arxiv.org/abs/2603.00351

作者：Lion Weber,Theodor Wienert,Martin Splettstößer,Alexander Koenig,Oliver Brock
备注：Accepted at ICRA 2026, supplementary material under https://rbo.gitlab-pages.tu-berlin.de/papers/acoustic-jamming-icra26/

【68】Challenges in Enabling Private Data Valuation
标题：实现私人数据估值的挑战
链接：https://arxiv.org/abs/2603.00342

作者：Yiwei Fu,Tianhao Wang,Varun Chandrasekaran

【69】When does Chain-of-Thought Help: A Markovian Perspective
标题：思想链何时有帮助：马可夫的观点
链接：https://arxiv.org/abs/2603.00306

作者：Zihan Wang,Yijun Dong,Qi Lei

【70】Efficient Long-Horizon GUI Agents via Training-Free KV Cache Compression
标题：通过免训练KV缓存压缩实现高效的长期GUI代理
链接：https://arxiv.org/abs/2603.00188

作者：Bowen Zhou,Zhou Xu,Wanli Li,Jingyu Xiao,Haoqian Wang

【71】A Boundary-Metric Evaluation Protocol for Whiteboard Stroke Segmentation Under Extreme Imbalance
标题：极端不平衡下白板笔画分割的边界度量评估协议
链接：https://arxiv.org/abs/2603.00163

作者：Nicholas Korcynski
备注：10 pages, 8 figures. Preprint

【72】SKINOPATHY AI: Smartphone-Based Ophthalmic Screening and Longitudinal Tracking Using Lightweight Computer Vision
标题：SKINOPATHY AI：使用轻量级计算机视觉的基于智能手机的眼科筛查和纵向跟踪
链接：https://arxiv.org/abs/2603.00161

作者：S. Kalaycioglu,C. Hong,M. Zhu,H. Xie
备注：25 pages , 7 figures, 5 tables

【73】Pulse-Driven Neural Architecture: Learnable Oscillatory Dynamics for Robust Continuous-Time Sequence Processing
标题：脉冲驱动神经架构：用于稳健连续时间序列处理的可学习振荡动力学
链接：https://arxiv.org/abs/2603.00153

作者：Paras Sharma
备注：16 pages, 8 figures, 8 tables

【74】SEval-NAS: A Search-Agnostic Evaluation for Neural Architecture Search
标题：SEval-NAS：神经架构搜索的搜索不可知评估
链接：https://arxiv.org/abs/2603.00099

作者：Atah Nuh Mih,Jianzhou Wang,Truong Thanh Hung Nguyen,Hung Cao
备注：To be published in the Proceedings of The 41st ACM/SIGAPP Symposium on Applied Computing (SAC26)

【75】Certainty-Validity: A Diagnostic Framework for Discrete Commitment Systems
标题：契约有效性：离散承诺系统的诊断框架
链接：https://arxiv.org/abs/2603.00070

作者：Datorien L. Anderson
备注：18 pages, 1 figure, full experiment data can be found: https://zenodo.org/records/18530003

【76】Measuring What AI Systems Might Do: Towards A Measurement Science in AI
标题：测量人工智能系统可能做什么：迈向人工智能的测量科学
链接：https://arxiv.org/abs/2603.00063

作者：Konstantinos Voudouris,Mirko Thalmann,Alex Kipnis,José Hernández-Orallo,Eric Schulz

【77】The Hidden Costs of Domain Fine-Tuning: Pii-Bearing Data Degrades Safety and Increases Leakage
标题：域微调的隐性成本：Pi承载数据降低安全性并增加泄漏
链接：https://arxiv.org/abs/2603.00061

作者：Jayesh Choudhari,Piyush Kumar Singh

【78】Property-Driven Evaluation of GNN Expressiveness at Scale: Datasets, Framework, and Study
标题：以属性驱动的GNN规模表现力评估：数据集、框架和研究
链接：https://arxiv.org/abs/2603.00044

作者：Sicong Che,Jiayi Yang,Sarfraz Khurshid,Wenxi Wang

【79】Attn-QAT: 4-Bit Attention With Quantization-Aware Training
标题：Attn-QAT：4位注意力与量化意识训练
链接：https://arxiv.org/abs/2603.00040

作者：Peiyuan Zhang,Matthew Noto,Wenxuan Tan,Chengquan Jiang,Will Lin,Wei Zhou,Hao Zhang

【80】TCG CREST System Description for the DISPLACE-M Challenge
标题：DISPLACE-M挑战赛的TCG CREST系统描述
链接：https://arxiv.org/abs/2603.02030

作者：Nikhil Raghav,Md Sahidullah
备注：Report submitted for the DISPLACE-M challenge

【81】LOCUS: A Distribution-Free Loss-Quantile Score for Risk-Aware Predictions
标题：LOCUS：风险意识预测的无分布损失分位数评分
链接：https://arxiv.org/abs/2603.01971

作者：Matheus Barreto,Mário de Castro,Thiago R. Ramos,Denis Valle,Rafael Izbicki
备注：The article contains nine pages and the appendix twelve

【82】On the Stability Connection Between Discrete-Time Algorithms and Their Resolution ODEs: Applications to Min-Max Optimisation
标题：离散时间算法及其分辨率ODE之间的稳定性联系：在Min-Max优化中的应用
链接：https://arxiv.org/abs/2603.01430

作者：Amir Ali Farzin,Yuen-Man Pun,Philipp Braun,Iman Shames

【83】Causal Effects with Unobserved Unit Types in Interacting Human-AI Systems
标题：人机交互系统中未观察到的单元类型的因果效应
链接：https://arxiv.org/abs/2603.01339

作者：William Overman,Sadegh Shirani,Mohsen Bayati

【84】Random Features for Operator-Valued Kernels: Bridging Kernel Methods and Neural Operators
标题：运算符值核的随机特征：桥核方法和神经运算符
链接：https://arxiv.org/abs/2603.00971

作者：Mike Nguyen,Nicole Mücke

【85】Efficient Conformal Volumetry for Template-Based Segmentation
标题：基于模板的高效保形体积测量
链接：https://arxiv.org/abs/2603.00798

作者：Matt Y. Cheung,Ashok Veeraraghavan,Guha Balakrishnan

【86】Initialization-Aware Score-Based Diffusion Sampling
标题：初始化感知基于分数的扩散抽样
链接：https://arxiv.org/abs/2603.00772

作者：Tiziano Fassina,Gabriel Cardoso,Sylvan Le Corff,Thomas Romary

【87】The Partition Principle Revisited: Non-Equal Volume Designs Achieve Minimal Expected Star Discrepancy
标题：重新审视分区原则：非等体积设计实现最小的预期星差
链接：https://arxiv.org/abs/2603.00202

作者：Xiaoda Xu
备注：This article serves as a clarification of the previous claims that purported to solve the open problem but encountered difficulties. My latest findings have reached a superior level

【88】What Is the Geometry of the Alignment Tax?
标题：统一税的几何结构是什么？
链接：https://arxiv.org/abs/2603.00047

作者：Robin Young

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