Yijiang (William) Li

I’m Yijiang ¹ (William) Li, a first-year PhD student in Machine Learning and Data Science, Electrical and Computer Engineering at UC San Diego.

My research focuses on the learning aspects of AI - to enable efficient (e.g. label efficiency ICCV23, CVPR23, sample effiency AAAI25, Wasserstein DD, and synthetic data) and robust learning (ICCV23, TMLR) in multi-modal (ICLR25, video-llm), interactive and 3D embodied environments. Particularly, I am interested in

• Understanding and evaluating Multi-modal Large Language models (MLLMs), e.g. core-knowledge;
• Can language (in nature semantic, abstracted and compositional) be used to scaffold vision learning (e.g. as anchor representation)?
• Intrinsic 3D world: 3D-aware architecture for MLLMs, 3D representation emergence from 2D supervision, etc.
• Embodiment theory, learning with indirect feedback (usually in social scenarios) and interactive learning (learn and inference simultaneously).

I am also broadly interested in the various applications of AI, e.g. LLM Agents (MAS), AI4Sci and AI4Health.

Bio     Contact     Github     G. Scholar     LinkedIn     Twitter