New research paper posted to arXiv proposing the use of MLLMs as training-free similarity estimators for image-to-image retrieval.

Researchers develop a novel framework using reflective reinforcement learning to enhance emotional reasoning in multimodal AI systems.