ML 日报 2026-06-03
arXiv 每日精选，覆盖 cs.LG / cs.CL / cs.AI。
精选论文
1. OpenWebRL: Demystifying Online Multi-turn Reinforcement Learning for Visual Web Agents
作者: Rui Yang, Qianhui Wu, Yuxi Chen et al. | 分类: cs.LG | arXiv
Building capable visual web agents requires long-horizon reasoning, precise grounding, and robust interaction with dynamic real-world websites. Despite rapid progress, the strongest systems remain largely proprietary, while open agents still depend heavily on supervised post-training over large collections of curated web trajectories.
2. A Local Perturbation Theory for Cross-Domain Interference and Recovery in Multi-Domain RL
作者: Lei Yang, Siyu Ding, Deyi Xiong | 分类: cs.LG | arXiv
Reinforcement learning (RL) post-training improves large language models (LLMs) on individual domains such as mathematical reasoning, code generation, question answering, and creative writing (CW), but training on one domain often degrades performance on others. Existing explanations based on catastrophic forgetting or global gradient conflict are incomplete: substantial interference can occur even when full-model gradients are nearly orthogonal.
3. ProtoAda: Prototype-Guided Adaptive Adapter Expansion and Geometric Consolidation for Multimodal Continual Instruction Tuning
作者: Yu-Cheng Shi, Zhen-Hao Xie, Jun-Tao Tang et al. | 分类: cs.CV | arXiv
Multimodal Large Language Models (MLLMs) achieve strong performance through instruction tuning, but real-world deployment requires them to continually acquire new vision-language capabilities, making Multimodal Continual Instruction Tuning (MCIT) essential. To reduce inter-task interference and promote collaboration, recent methods often employ sparse architectures like Mixture of LoRA Experts with image-text similarity routing.
4. SIRI: Self-Internalizing Reinforcement Learning with Intrinsic Skills for LLM Agent Training
作者: Zhongyu He, Yuanfan Li, Fei Huang et al. | 分类: cs.AI | arXiv
Long-horizon LLM agents can benefit from reusable skills, yet existing skill-based methods often rely on external skill generators during training or persistent skill retrieval at inference, increasing engineering complexity, context length, and deployment latency. We propose Self-Internalizing Reinforcement learning with Intrinsic skills (SIRI), a three-phase framework that enables agents to discover, validate, and internalize skills without external skill generators or inference-time skill banks.
5. Extreme Low-Bit Inference in Reasoning Models: Failure Modes and Targeted Recovery
作者: Ekaterina Alimaskina, Darya Rudas, Denis Shveykin et al. | 分类: cs.AI | arXiv
Large Reasoning Models (LRMs) rely on long reasoning traces, making inference expensive. While low-bit quantization reduces per-token decoding cost, we show that aggressive 2-bit inference can fail to deliver end-to-end speedup because instability in the generation process inflates total token count.
6. Policy and World Modeling Co-Training for Language Agents
作者: Ning Lu, Baijiong Lin, Shengcai Liu et al. | 分类: cs.LG | arXiv
Reinforcement learning (RL) improves large language model (LLM) agents by teaching them which actions lead to high rewards, but provides little supervision on what those actions do to the environment. World modeling (WM) can fill this gap, yet existing approaches often require separate simulators, extra training stages, or additional inference-time computation.
7. When Do Attention Circuits Form? Developmental Trajectories of Capability and Attention-Sink Emergence Across Three 1B-ClassArchitectures
作者: Yongzhong Xu | 分类: cs.LG | arXiv
We track the developmental trajectory of attention-head circuit formation across three 1B-class language models spanning two architecture families (dense transformer, mixture-of-experts) and two pretraining corpora (The Pile, DCLM): Pythia 1B, OLMo 1B-0724-hf, and OLMoE 1B-7B-0924. At each of 10 log-spaced revisions per model -- 30 mechanistic-interpretability runs in total -- we apply a participation-ratio (PR) spectral signal and an all-head capability-specific selectivity screen to track induction, previous-token, and BOS-attractor heads as they emerge.
8. Unveiling the Entropy Dynamics of Chain-of-Thought Reasoning
作者: Ting Xu, Xu He, Yupu Lu et al. | 分类: cs.CL | arXiv
This paper investigates the entropy dynamics of Chain-of-Thought (CoT) and uncovers a consistent two-phase structure: an Uncertainty Region of exploration transitioning sharply to a Confidence Region of convergence. We demonstrate that the Confidence Region possesses two critical properties: 1) High Reliability -- answers in the confidence region become highly accurate and stable, and 2) High Redundancy -- models generate unnecessary tokens long after reaching the correct answer.
9. Permissive Safety Through Trusted Inference: Verifiable Belief-Space Neural Safety Filters for Assured Interactive Robotics
作者: Haimin Hu | 分类: cs.RO | arXiv
Autonomous robots that interact with people must make safe and efficient decisions under human-induced uncertainty, such as their preferences, goals, competency, and willingness to cooperate. Safety filters are a popular approach for ensuring safety in interactive robotics, since their modular design separates safety from performance, allowing robots to operate safely around people with minimal impact on task efficiency.
10. Auditing Asset-Specific Preferences in Financial Large Language Models: Evidence from Bitcoin Representations and Portfolio Allocation
作者: Wenbin Wu | 分类: q-fin.GN | arXiv
Large language models now power robo-advisors and trading agents, yet whether they carry built-in biases toward specific assets is largely untested. We ask three questions: do LLMs systematically prefer certain financial instruments; can an internal representation with causal leverage over those preferences be identified; and does that representation affect downstream financial decisions?