MathVis-Fine: Aligning Visual Supervision with Necessity via Progressive Dependency-Guided Training for Multimodal Mathematical Reasoning

In three linesMathVis-Fine introduces a framework for fine-grained visual dependency modeling in mathematical reasoning. A new dataset augments visual annotations with visual dependency ratings. Two-stage progressive training balances answer correctness and visual grounding rewards according to each sample's intrinsic visual necessity, reducing reward bias.

Summary generated by Claude — human-verified