作者位次 #6/6ACL 2026Poster· Natural Language Processing
AIPO : Adaptive Information Guided Token-Level Reinforcement Learning for Large Language Model Reasoning
Bin Chen · Hongfei Ye · Huiyang Wang · Wenxi Liu · Yu Zhang · Furui Liu
摘要
Reinforcement Learning with Verifiable Rewards (RLVR) improves the reasoning capability of Large Language Models (LLMs). Current RLVR trains LLMs on all generated tokens, rather than exploring which tokens actually contribute to reasoning. We propose AIPO(Adaptive–Information Policy Optimization)...