DEEPO

Researchers introduced Dual-Entropy Enhanced Policy Optimization (DEEPO), a dual-stage reinforcement learning method that reduces hallucination in multimodal large language models by combining semanti…