Dr. Shengchao Liu | Computer Science | Research Excellence Award

The Chinese University | Hong Kong

Shengchao Liu is a tenure-track Assistant Professor in the Department of Computer Science and Engineering at The Chinese University of Hong Kong, whose research lies at the intersection of machine learning, geometry, and scientific discovery. His work focuses on developing foundation models and physics-inspired learning frameworks for molecules, proteins, and materials, with the long-term goal of accelerating discovery in chemistry, biology, and materials science. By integrating multi-modal data, symmetry principles, and domain knowledge, his research bridges theoretical advances in AI with real-world experimental impact. A central theme of Dr. Liu’s research is geometric and symmetry-informed representation learning. He has pioneered group-equivariant and manifold-constrained generative models that respect the underlying physical laws of molecular and material systems. His contributions include SE(3)-invariant pretraining methods, group-symmetric stochastic differential equation models, and rigid flow matching techniques, which have significantly improved the fidelity and interpretability of molecular generation and dynamics modeling. These methods form a unifying framework for learning across molecules, proteins, and crystalline materials, as demonstrated in his influential works at ICLR, ICML, NeurIPS, and AISTATS. Dr. Liu’s work is deeply collaborative and interdisciplinary. He has worked closely with leading researchers across academia and industry, including Mila, UC Berkeley, NVIDIA Research, and national laboratories. As a Principal Investigator, he has led NERSC-supported projects on foundation models for material discovery, leveraging large-scale GPU resources to push the frontier of generative AI for science. His research has also contributed widely used open-source resources, including geometric graph learning benchmarks and toolkits adopted by the broader AI-for-science community.

View Google Scholar Profile

Featured Publications