MIRL uses mutual information to guide trajectory selection and provide separate rewards for visual perception in RLVR for VLMs, achieving 70.22% average accuracy with 25% fewer full trajectories.
MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts , booktitle =
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
MIRL: Mutual Information-Guided Reinforcement Learning for Vision-Language Models
MIRL uses mutual information to guide trajectory selection and provide separate rewards for visual perception in RLVR for VLMs, achieving 70.22% average accuracy with 25% fewer full trajectories.