报告题目：Graph and Geometry Generative AI with Applications in Drug Discovery
报告人：徐民凯 博士生 
主持人：朱永群 教授
时   间：2023年12月27日（周三）下午2点
地   点：纳米楼521会议室
报告人简介：

Minkai is a Ph.D. student in the Computer Science Department at Stanford University, advised by Jure Leskovec and Stefano Ermon. His research is generally in machine learning, with an emphasis on Generative AI and AI for Science. His pioneering research on molecular generative models has been widely featured in both academia and industry. Previously, he received his M.S. from Mila and B.E. from Shanghai Jiao Tong University. He has spent time at AI research labs of Nvidia, Meta, Amazon, and ByteDance. His research is generously supported by Sequoia Capital Fellowship. More info: https://minkaixu.com

报告摘要：

With the recent progress in generative models and availability of large-scale datasets, Graph and Geometry Generative AI has emerged as a promising direction for scientific discovery such as drug design. These methods enable efficient chemical space exploration and hypothetical drug generation. However, fundamental challenges exist in modeling the distribution of irregular and complex relational data with physical symmetries. In this talk, we will first introduce our key developments in this field and present unified views of the representative approaches, covering our Geometric Diffusion, Latent Diffusion, Hierarchical Diffusion, Multi-modal Diffusion, and Flow Matching models for 3D molecular structure generation. We will further discuss concrete scientific problems in drug discovery, and highlight our future directions and real-world impacts.