ReflectRL: Learning from Golden Negative Trajectories via Reflective-to-Direct Reasoning

  • 类型：arxiv
  • 标识：2608.03972
  • 链接：https://arxiv.org/abs/2608.03972
  • 主分类：engineering
  • 形态：method
  • 被引：5
  • 被引来源：Semantic Scholar
  • S2被引：5
  • OpenAlex被引：0
  • 影响力被引：0
  • TLDR：ReflectRL is proposed, a lightweight plug-and-play framework that learns from Golden Negative Trajectories during on-policy training, and first uses these trajectories to elicit Reflective Reasoning, then applies Reflective-to-Direct Policy Transition to transfer the acquired reasoning behavior back to Direct Reasoning.
  • OpenAlex ID：W7172535609
  • OpenAlex DOI：10.48550/arxiv.2608.03972
  • DOI：10.48550/arxiv.2608.03972
  • DOI来源：OpenAlex
  • 开放获取：green
  • 开放获取链接：https://doi.org/10.48550/arxiv.2608.03972
  • OpenAlex更新：2026-08-29
  • 待LLM分类：否
  • 标题中文：ReflectRL: 通过反思到直接推理从 Golden Negative Trajectories 中学习
  • TLDR中文：提出 ReflectRL，一个轻量级即插即用框架，在 on-policy 训练中从 Golden Negative Trajectories 中学习：先利用这些 trajectory 引出 Reflective Reasoning，再通过 Reflective-to-Direct Policy Transition 将所学到的推理行为迁移回 Direct Reasoning。
  • 来源文件：
  • /inbox/tom/_candidates/2026-08-06-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • [OpenAlex backfill]