作者位次 #3/6ACL 2026Findings· Findings of ACL
A uto R ubric: Rubric-Based Generative Rewards for Faithful Multimodal Reasoning
Mengzhao Jia · Zhihan Zhang · Ignacio Cases · Zheyuan Liu · Meng Jiang · Peng Qi
摘要
Multimodal large language models (MLLMs) have rapidly advanced from perception tasks to complex multi-step reasoning, yet reinforcement learning with verifiable rewards (RLVR) often leads to spurious reasoning since only the final-answer correctness is rewarded. To address this limitation, we pro...