Chunfeng Yuan

SALR is proposed, a schema-anchored latent reasoning method for LF construction that performs multi-step reasoning by generating continuous thoughts in the model's hidden states, thereby delaying the explicit commitment to LF decisions.

GMM-EVA slashes visual token budgets in long video understanding by intelligently prioritizing keyframes based on event-level structures.

Visual reranking and active rejection in MMAgent-R$^2$ significantly boost retrieval accuracy in challenging KB-VQA tasks, outperforming traditional methods.

MLLMs can be tricked into missing 90% of harmful content simply by encoding it in images that humans can easily read.