作者位次 #3/8ACL 2026Poster· Natural Language Processing
Visually-Guided Policy Optimization for Multimodal Reasoning
Zengbin Wang · Feng Xiong · Liang Lin · Xuecai Hu · Yong Wang · Yanlin Wang · Man Zhang · Xiangxiang Chu
摘要
Reinforcement learning with verifiable rewards (RLVR) has significantly advanced the reasoning ability of vision-language models (VLMs). However, the inherent text-dominated nature of VLMs often leads to insufficient visual faithfulness, characterized by sparse attention activation to visual toke...