Boosts multimodal reasoning by teaching models to autonomously verify their own long-form generations against image evidence using information gain.
March 30, 2026
Original Paper
Reflect to Inform: Boosting Multimodal Reasoning via Information-Gain-Driven Verification
arXiv · 2603.26348
The Takeaway
The framework (VRE) activates a latent visual verification capability already present in MLLMs but often ignored in long-chain reasoning. It significantly reduces hallucinations in long-form tasks without requiring stronger teacher models or additional human-annotated data.
From the abstract
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