Yonghan (Harry) is an undergraduate researcher in Artificial Intelligence at the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in Abu Dhabi. His work sits at the intersection of generative modeling, agentic AI, and AI for science: he likes building methods that are principled enough to prove something about, yet practical enough to run.
Most of his research asks a single question in different disguises — how do we let a model search beyond the data it was trained on without fooling itself? This led to SPADE (ICML 2026), where a calibrated conditional diffusion surrogate is kept honest by a support-proximity prior for offline black-box optimization. He is now extending these ideas to memory retrieval for autonomous agents, discrete diffusion, and biomedical foundation models, in collaboration with researchers at Mila, Harvard Medical School, and GenBio AI.
Before MBZUAI he studied at Beijing National Day School, where he led the school’s synthetic biology and iGEM teams to two consecutive gold medals, and represented China in mathematical modeling. That path — from wet-lab genetic circuits to score-based generative models — is why he cares about interdisciplinary work: the most interesting problems rarely respect departmental boundaries.