Reflect to Inform: Boosting Multimodal Reasoning via Information-Gain-Driven Verification
This addresses a critical failure mode in multimodal reasoning for AI applications, though it is an incremental improvement on existing methods.
The paper tackles the problem of multimodal large language models drifting from image evidence in long-form generation, leading to hallucinations, and proposes a self-evolving training framework that improves reasoning accuracy and reduces hallucinations, achieving consistent gains across benchmarks.
Multimodal Large Language Models (MLLMs) achieve strong multimodal reasoning performance, yet we identify a recurring failure mode in long-form generation: as outputs grow longer, models progressively drift away from image evidence and fall back on textual priors, resulting in ungrounded reasoning and hallucinations. Interestingly, Based on attention analysis, we find that MLLMs have a latent capability for late-stage visual verification that is present but not consistently activated. Motivated by this observation, we propose Visual Re-Examination (VRE), a self-evolving training framework that enables MLLMs to autonomously perform visual introspection during reasoning without additional visual inputs. Rather than distilling visual capabilities from a stronger teacher, VRE promotes iterative self-improvement by leveraging the model itself to generate reflection traces, making visual information actionable through information gain. Extensive experiments across diverse multimodal benchmarks demonstrate that VRE consistently improves reasoning accuracy and perceptual reliability, while substantially reducing hallucinations, especially in long-chain settings. Code is available at https://github.com/Xiaobu-USTC/VRE.