On-Policy Distillation是一种强化学习技术,通过将策略模型同时作为教师和学生进行知识传递,旨在提升学习效率。2023年10月,研究人员提出了JOLT方法,让同一模型身兼师生两角进行联合训练,显著提升了强化学习的效率。
On-Policy Distillation
concept · 首次出现 2026-05-22 · 最近出现 2026-10-16 · 累计提及 44
综述
相关报道
9 条在档- BranchOPD:分支式在策略蒸馏提升数学与智能体任务表现arXiv cs.AI
- JOLT:让同一模型身兼师生两角,联合训练提升强化学习效率arXiv cs.AI
- 研究用强化学习视角解释 On-policy Distillation 崩溃成因arXiv cs.AI
- Apple 发布 On-Policy Distillation 研究:揭示其利弊与原因AK
- Towards Continual Learning: Mitigating Forgetting With Subspace TuningarXiv cs.LG
- Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy DistillationarXiv cs.LG
- NVIDIA PivotOPD Teaches Multi-Turn AI Agents to Recover From Pivotal Mistakesmarktechpost
- Composing What Each Teacher Learned: Multi-Teacher On-Policy Distillation through Teacher-Relative ShiftsarXiv cs.AI
- Sharpen Without Search: On-Policy Distillation of Sequence-Level Power DistributionarXiv cs.AI