牛哥精选 · 所有

1

📝 深度技术 arXiv 机器学习 2026-07-24

The Dark Room in the Reward Channel: Dense Prediction Rewards Collapse GRPO-Trained LLM Agents -- and What Actually Works

给LLM智能体喂"预判未来"的奖励，反而把策略喂崩了——这项研究戳破了密集奖励的幻觉，揭示了GRPO训练中真正有效的方法。

arXiv:2607.21273v1 Announce Type: new Abstract: Dense per-step supervision is an appealing remedy for sparse-reward, long-horizon LLM agents: reward t…

llm智能体 grpo强化学习密集预测奖励策略崩溃奖励设计

2

📝 深度技术 arXiv AI 2026-07-07

ACPO: Adaptive Credit Policy Optimization via Fine-Grained Surrogate Entropy

用细粒度替代熵突破轨迹奖励瓶颈，为可验证推理任务打造更精准的信用分配新方案。

arXiv:2607.03126v1 Announce Type: cross Abstract: Reinforcement Learning (RL) has substantially improved the reasoning ability of large language model…

acpo 强化学习熵奖励设计推理任务

3

📝 深度技术 arXiv 机器学习 2026-07-03

Composite Reward Design in PPO-Driven Adaptive Filtering

复合奖励设计如何提升PPO驱动的自适应滤波降噪性能？这篇来自计算科学国际会议的研究给出了具体方案与实验验证。

arXiv:2506.06323v2 Announce Type: replace-cross Abstract: Model-free and reinforcement learning-based adaptive filtering methods are gaining traction …

强化学习 ppo 自适应滤波降噪奖励设计

4

🤖 AI·大模型 arXiv AI 2026-06-23

ARCO: Adaptive Rubric with Co-Evolution for Multi-Step LLM-Based Agents

传统奖励只看成败，这套方法用自适应评分标准给多步智能体每一步“打分讲理”，可解释强化学习的新思路。

arXiv:2606.21262v1 Announce Type: new Abstract: Reinforcement learning for multi-step LLM agents often relies on scalar rewards that indicate success …

arco 强化学习 llm agent 可解释奖励自适应评分

5

🤖 AI·大模型 arXiv 计算机视觉 2026-06-17

Disentangling Perception and Reasoning in Multimodal LLMs via Reward Design

最新研究通过奖励设计巧妙解耦多模态大模型中的感知与推理模块，显著提升模型可解释性与泛化能力，值得AI研究者关注。

arXiv:2601.00215v2 Announce Type: replace Abstract: Reinforcement learning with verifiable rewards has driven major gains in LLM reasoning, and it is …

多模态llm 感知与推理奖励设计可解释性 ai研究

6

📝 深度技术 arXiv AI 2026-06-09

ComplexConstraints and Beyond: Expert Rubrics for RLVR

提出专家评分标准解决RLVR中复杂约束问题，为强化学习奖励设计提供新范式

arXiv:2606.09118v1 Announce Type: new Abstract: As LLM capabilities advance rapidly, the evaluation methods used to assess them increasingly lag behin…

rlvr 专家评分标准复杂约束强化学习奖励设计

7

🤖 AI·大模型 arXiv 机器学习 2026-06-02

Large Language Model Guided Incentive Aware Reward Design for Cooperative Multi-Agent Reinforcement Learning

大模型指导多智能体合作强化学习，激励感知奖励设计新突破，提升协同效率与泛化能力。

arXiv:2603.24324v4 Announce Type: replace Abstract: Designing effective auxiliary rewards for cooperative multi-agent systems remains challenging, as …

large lang 多智能体强化学习激励感知奖励设计合作智能体

8

📝 深度技术 arXiv 机器学习 2026-05-29

When LLM Reward Design Fails: Diagnostic-Driven Refinement for Sparse Structured RL

当大模型奖励设计失效时，这篇论文提出诊断驱动方案，用稀疏结构化RL来迭代优化，给AI训练提供新思路。

arXiv:2605.28918v1 Announce Type: new Abstract: For sparse, structured reinforcement-learning tasks with semantic reward-function interfaces, LLM-gene…

llm 奖励设计强化学习稀疏rl 诊断驱动

9

📝 深度技术 arXiv 机器学习 2026-05-27

Focal Reward: Balanced Reinforcement Learning under Rubric-Based Rewards

平衡强化学习新方法，利用Focal Reward聚焦关键学习信号，提升训练效率与效果。

arXiv:2605.26579v1 Announce Type: new Abstract: The open-ended generation in LLMs usually requires multi-dimensional rubrics to adequately assess qual…

强化学习奖励设计 focal rewa 深度学习训练优化

10

📝 深度技术 arXiv AI 2026-05-21

Attention-Guided Reward for Reinforcement Learning-based Jailbreak against Large Reasoning Models

用注意力机制优化强化学习奖励，成功攻破大型推理模型的安全防线，揭示AI安全新挑战。

arXiv:2605.19485v1 Announce Type: new Abstract: Large Reasoning Models (LRMs) have demonstrated remarkable capabilities in solving complex problems by…

强化学习越狱攻击大型推理模型注意力机制安全对齐

11

📝 深度技术 arXiv AI 2026-05-20

Beyond GRPO and On-Policy Distillation: An Empirical Sparse-to-Dense Reward Principle for Language-Model Post-Training

提出稀疏到稠密奖励原则，四阶段后训练流程更高效利用稀缺标注数据，为LLM推理优化提供新范式。

arXiv:2605.12483v2 Announce Type: replace-cross Abstract: When labeled verifiable training data is scarce, each checked example should be used where i…

llm后训练强化学习奖励设计 grpo 知识蒸馏

🐂 牛哥精选