KEPO improves RL stability and out-of-distribution performance for multimodal reasoning by selectively distilling from high-quality trajectories and using teacher hints to sample better on-policy paths.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
KEPO: Knowledge-Enhanced Preference Optimization for Multimodal Reasoning with Applications to Medical VQA
KEPO improves RL stability and out-of-distribution performance for multimodal reasoning by selectively distilling from high-quality trajectories and using teacher hints to sample better on-policy paths.