Researches

• Geometric Learning

One of my research direction focuses on geometric learning, particularly on incorporating symmetry and structure into machine learning models. I work on group equivariance and symmetry and design networks that respect geometric representations. I also work on 3D Graph Representation Learning, developing efficient and trustworthy models that capture both spatial and relational aspects of 3D graph data.

Application: Molecules, Point Clouds, etc.

• Probabilistic Generation Model

Another direction of my research focuses on probabilistic generative models centers on improving both efficiency and expressivity. I work on methods to accelerate the training and sampling processes of generative models, making them more practical for large-scale or real-time applications. At the same time, I aim to preserve essential structural properties, such as equivariance, symmetry, and other domain-specific constraints within the generation process, ensuring that the outputs remain faithful to the underlying data geometry.

Application: Diffusion Models, Flow-matching Models, etc.