R-C2: Cycle-Consistent Reinforcement Learning Improves Multimodal Reasoning
2026-03-26 • Artificial Intelligence
Artificial IntelligenceComputer Vision and Pattern Recognition
AI summaryⓘ
The authors explain that for better understanding, models should agree on what they see and read about the same thing. They created a method called RC2 that teaches models to check their own answers by switching between pictures and text and making sure the results match. This way, the model learns to fix its own mistakes without needing extra labeled examples. Their approach helped models improve how well they think by about 7.6 points. The study shows that making models internally consistent helps them reason better, beyond just using more data.
multimodal modelscross-modal consistencyreinforcement learningcycle consistencybackward inferenceforward inferenceinternal representationsreasoning accuracylabel-free learningsystematic biases
Authors
Zirui Zhang, Haoyu Dong, Kexin Pei, Chengzhi Mao
Abstract
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.