dense-rewards

#dense-rewards

Expert-Guided Rubrics as Dense Reward Signals for Reinforcement Learning in Non-Verifiable Domains

arXiv cs.LG ↗ · 2026-10-21 Cached

该论文提出将少量专家(SME)评判转化为密集奖励信号，并与检索指标结合成适用于GRPO的混合奖励，以解决非可验证领域（如金融服务检索）中稀疏奖励导致的梯度消失问题，实验显示Recall@3提升最多6%、MRR@3提升6.7%。

0 favorites 0 likes

#dense-rewards

LongR: Unleashing Long-Context Reasoning via Reinforcement Learning with Dense Utility Rewards

Hugging Face Daily Papers ↗ · 2026-10-12 Cached

LongR 是一个统一框架，通过将“Think-and-Read”动态机制与基于相对信息增益的上下文密度奖励相结合，提升 LLM 的长上下文推理能力，在 LongBench v2 上取得超过 9% 的提升，并在 RULER 和 InfiniteBench 上保持一致改进。

0 favorites 0 likes

#dense-rewards

From Connectivity to Rewards: Dense Reward Learning with Directed State Graphs

arXiv cs.LG ↗ · 2026-09-11 Cached

This paper introduces G2QDR, a framework that leverages directed state graphs to generate dense rewards for goal-conditioned hierarchical reinforcement learning, enhancing performance in sparse reward environments.

0 favorites 0 likes

dense-rewards

Expert-Guided Rubrics as Dense Reward Signals for Reinforcement Learning in Non-Verifiable Domains

LongR: Unleashing Long-Context Reasoning via Reinforcement Learning with Dense Utility Rewards

From Connectivity to Rewards: Dense Reward Learning with Directed State Graphs

Submit Feedback