作者位次 #2/8ACL 2026Poster· Natural Language Processing
Rethinking Sample Polarity in Reinforcement Learning with Verifiable Rewards
Xinyu Tang · Yuliang Zhan · Zhixun Li · Wayne Xin Zhao · Zhenduo Zhang · zujie wen · Zhiqiang Zhang · JUN ZHOU
摘要
Large reasoning models (LRMs) are typically trained using reinforcement learning with verifiable reward (RLVR) to enhance their reasoning abilities. In this paradigm, policies are updated using both positive and negative self-generated rollouts, which correspond to distinct sample polarities . In...