arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.25720cs.AIcs.CV
收藏

R-C2:循环一致性强化学习提升多模态推理

R-C2: Cycle-Consistent Reinforcement Learning Improves Multimodal Reasoning

Rutgers University(罗格斯大学);Columbia University(哥伦比亚大学);University of Chicago(芝加哥大学)

Zirui Zhang, Haoyu Dong, Kexin Pei, Chengzhi Mao

AI总结:

R-C2通过循环一致性强化学习解决多模态推理中的内部冲突,提升推理准确率7.6个百分点。

摘要:

R-C2通过循环一致性强化学习解决多模态推理中的内部冲突,提升推理准确率7.6个百分点。

英文摘要:

Robust perception and reasoning require consistency across sensory modalities. Yet current multimodal models often violate this principle, yielding contradictory predictions for visual and textual representations of the same concept. Rather than masking these failures with standard voting mechanisms, which can amplify systematic biases, we show that cross-modal inconsistency provides a rich and natural signal for learning. We introduce RC2, a reinforcement learning framework that resolves internal conflicts by enforcing cross-modal cycle consistency. By requiring a model to perform backward inference, switch modalities, and reliably reconstruct the answer through forward inference, we obtain a dense, label-free reward. This cyclic constraint encourages the model to align its internal representations autonomously. Optimizing for this structure mitigates modality-specific errors and improves reasoning accuracy by up to 7.6 points. Our results suggest that advanced reasoning emerges not only from scaling data, but also from enforcing a structurally consistent understanding of the world.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑