作者位次 #10/10ACL 2026Findings· Findings of ACL
L earn A lign: Data Selection for LLM Reinforcement Learning with Improved Gradient Alignment
Shipeng Li · Zhiqin Yang · Shikun Li · Xiaobo Xia · Hengyu Liu · Xinghua Zhang · Gaode Chen · Dong Fang · Ying Tai · Zhe Peng
摘要
Reinforcement learning with verifiable rewards (RLVR) has become a key technique for enhancing LLMs’ reasoning abilities, yet its data inefficiency remains a major bottleneck. To address this critical yet challenging issue, we present a novel gradient-alignment-based method, named LearnAlign, whi...