Haochen Liang

MAVIS redefines video retrieval by enabling agents to collaboratively reason and refine candidate selections, outperforming traditional methods without task-specific tuning.

Multi-agent collaboration and retrieval augmentation can overcome the limitations of static parametric memory in LLMs, enabling more nuanced and accurate multimodal emotion recognition.

LLMs can surgically remove semantic noise from smart contract constraints, dramatically accelerating hybrid fuzzing without sacrificing soundness.

By explicitly modeling multi-view visual-semantic relationships, MViR achieves state-of-the-art fake news detection, suggesting that subtle image-text alignments are critical for identifying misinformation.