Time: 14:00-15:30, Tuesday, August 16, 2022

Online: ZOOM ID 820 0817 5123

Host: Yizhou Zhu, Associate Professor, School of Engineering, Westlake University

Speaker：

Dr. Tieyan Liu

Distinguished Scientist & Assistant Managing Director

Microsoft Research Asia & AI4Science

Biography:

Tie-Yan Liu is a distinguished scientist of Microsoft, an assistant managing director of Microsoft Research Asia and Microsoft Research AI4Science. He is a fellow of the IEEE, the ACM, and the AAIA. He is well known for his pioneer work on learning to rank, and recently his research interests include deep learning, reinforcement learning, AI for industry, and AI for science. He has published 300+ papers at top conferences and journals, with 30K+ citations in total. He has served as general chair, program committee chair, or (senior) area chair for many top conferences, including WWW/WebConf, SIGIR, KDD, ICML, NeurIPS, ICLR, IJCAI, AAAI, ACL, and associate editor of ACM Transactions on Information Systems, ACM Transactions on the Web, and IEEE Transactions on Pattern Analysis and Machine Intelligence. He won the best (student) paper award at SIGIR (2008) and ACML (2018), the most cited paper award at Journal of Visual Communications and Image Representation (2004-2006), the most cited Chinese researcher award by Elsevier, and the most influential scholar award by AMiner (2007-2017). His team open-sourced LightGBM in 2017 (which has become one of the most popular machine learning tools in Kaggle and KDD Cup), proposed the dual learning framework in 2018 (which helped Microsoft achieve human parity in machine translation), built the world-best Mahjong AI named Suphx in 2019 (which achieved 10 DAN on Tenhou and significantly outperformed top human players in terms of stable ranking), and invented Graphormer for molecular modeling in 2021 (which won the KDD Cup on molecular property prediction and the Open Catalyst Challenge).

Abstract:

Natural Science research covers a variety of cutting-edge attempts from mankind to understand and to transform the world. In the past few years, AI as an emerging tool has been introduced into multiple subfields of natural science research, powering breakthroughs with higher accuracy, faster speed, and larger scale. In this talk, I will present our recent research progress from Microsoft Research. To reflect the multiscale nature of the physics world, we made attempts to from both microscopic and macroscopic scale.

At the microscopic scale, the configurations of molecules dictate their chemical and physical properties. We developed Graphormer, a model with better representation learning and property production, which won the KDD Cup 2021 on molecular property prediction and the direct track of Open Catalyst Challenge 2021. Based on Graphormer, we further developed highly efficient molecular dynamics simulation system, and make successful attempts in understanding the dynamic properties of coronaviruses and drug design.

At the microscopic scale, we proposed a novel data-driven training paradigm for neural operators of partial differential equations. Our approach not only improves the training speed in high-dimensional and high-order cases, but also effectively deals with the case of nonsmooth solutions. Our work expands the application scope and practical value of neural network technology in accelerating the solution of partial differential equations. In addition to the above mentioned research, I will present other cases of using AI tools to facilitate scientific research. We believe that AI tools will pave the way for more scientific research breakthroughs in the future.

Contact:

Ms. Huaiwei Shi, School of Engineering

shihuaiwei@westlake.edu.cn