报告题目：Machine learning of structured representations for scientific discovery and simulation
报告人：吴泰霖（西湖大学）
时间地点：2023年6月14日周三上午10:30，理科楼C109
报告摘要：In this talk, I will present my work on developing machine learning methods for accelerating scientific discovery and simulation. In scientific discovery, discovering universal, simple laws from multiple dynamical systems is critical across many scientific disciplines. I introduce a “theory learning” paradigm and method incorporating four inductive biases from physicists, and show that it can discover universal laws across multiple environments with significantly better accuracy, interpretability, and sample efficiency. In scientific simulation, a critical challenge is that many systems are multi-resolution. I introduce LAMP, which uses a graph neural network (GNN) to learn the dynamics, and another GNN to learn the policy of spatial coarsening or refinement of the mesh representation, achieving a controllable tradeoff between prediction accuracy and computation efficiency at inference time.
报告人简介：Tailin Wu is an incoming assistant professor of AI in Westlake University (西湖大学). He did his postdoc in Stanford Computer Science, working with Prof. Jure Leskovec. He received his Ph.D. from MIT, and B.S. from Peking University. His research interests include developing machine learning methods for large-scale scientific simulations, design, and scientific discovery, with graph neural networks, information theory and representation learning. His work has been published in top machine learning conferences and leading physics journals, and featured in MIT Technology Review. He also serves as a reviewer for high-impact journals such as PNAS, Nature Communications, Nature Machine Intelligence, and Science Advances.
