Xuechao Zou

Single vision foundation models inevitably overfit to deepfake artifacts, but dynamically arbitrating between CLIP's semantics and DINO's structural features via prediction entropy delivers state-of-the-art generalization across unseen generators.

A multi-agent forensic reasoning framework outperforms leading closed-source models in deepfake detection by leveraging diverse analytical perspectives on forgery cues.

Hard prompt compressors can leave critical context gaps, leading to a staggering 60% of examples suffering from referential dangling, which severely impacts accuracy in multi-hop question answering.