一作NeurIPS 2026Poster
KL for a KL: On-Policy Distillation with Control Variate Baseline
Minjae Oh · Sangjun Song · Gyubin Choi · Yunho Choi · Yohan Jo
强化学习与后训练OpenReview
首次获取该候选人的完整档案及 StartLux 评估共 20 积分；每位用户单独解锁，之后查看免费。主动刷新 20 积分。
实时查找电子邮箱，费用不超过20积分；未找到邮箱不扣费。
Minjae Oh · Sangjun Song · Gyubin Choi · Yunho Choi · Yohan Jo
Yunho Choi · Jongwon Lim · woojin Ahn · Minjae Oh · Jeonghoon Shim · Yohan Jo
Sangjun Song · Minjae Oh · Seungkyu Lee · Sungmin Jo · Yohan Jo
Large Reasoning Models (LRMs) allocate substantial inference-time compute to Chain-of-Thought (CoT) reasoning, improving performance on mathematics, scientific QA, and tool usage. However, this introduces overthinking: LRMs often reach a correct intermediate solution, continue reasoning, and over...