Porf. Tie-yan Liu: AI for Science - The Fifth Paradigm of Scientific Discovery
On December 22, 2022, Tie-yan Liu, the Distinguished Chief Scientist of Microsoft, Vice Dean of Microsoft Research Asia, and Head of Microsoft Research AI4Science in Asia, delivered a fascinating presentation titled "AI for Science: The Fifth Paradigm of Scientific Discovery" at the "Peking University Intelligence Science Frontier Forum" through an online format. The event attracted the participation of many professors and students from the SIST, Institute for Artificial Intelligence, Wangxuan Institute of Computer Technology, and School of Computer Science.
At the beginning of the lecture, Prof. Liu outlined the emerging research field of AI for Science. He pointed out that with the development of AI, people hope to use AI as a means to discover scientific laws, rather than just as a tool for data processing. AI for Science combines AI with the four paradigms of human scientific discovery, further elevating the level of scientific discovery and becoming the fifth paradigm of scientific discovery.
Prof. Liu delved into three aspects of AI for Science, based on relevant research achievements from Microsoft Research. The first aspect is the use of AI to solve scientific equations. He introduced the Graphormer model for molecular modeling and the DeepVortexNe model for solving the Navier-Stokes equations. The second aspect is the use of AI to analyze scientific data. He introduced the SPT scientific language model for scientific literature mining and the LorentzNet model for analyzing jet data to discover new particles. The third aspect is the use of AI to automatically discover scientific laws. He introduced the NNPhD model, which learns physics formulas from non-conserved components of data.
In the conclusion of the presentation, Prof. Liu discussed the future development of AI for Science and patiently answered questions raised by students. He pointed out that the development of AI for Science requires in-depth interdisciplinary integration between the expertise of AI and other scientific fields, and designing specialized AI frameworks for AI for Science is one of the challenging technical problems. Prof. Zhouchen Lin summarized and expressed gratitude for Prof. Tie-yan Liu's excellent presentation, marking the successful conclusion of this event.
