报告人：杨柳

报告题目：物理约束生成网络

报告摘要：

Part I: Physics-Informed Generative Models

We will introduce Physics-informed generative adversarial networks (PI-GANs) and Bayesian physics-informed neural networks (B-PINNs). We will discuss how to learn “data prior” from historic observations with physic-informed generative models, and infer posterior with new data in the Bayesian framework. Other work on the topic of learning optimal transport map and particle dynamics will also be introduced.

Part II: In-Context Operator Networks

Can we build a single large model for a wide range of scientific problems?

We proposed a new framework for scientific machine learning, namely “In-Context Operator Learning” and the corresponding model “In-Context Operator Networks” (ICON). A distinguishing feature of ICON is its ability to learn operators from numerical prompts during the inference phase, without weight adjustments. This is similar to how a single Large Language Model can solve a variety of natural language processing tasks specified by the language prompt. We will show how a single ICON model (without fine-tuning) manages multiple distinct problem types, encompassing forward and inverse ODE, PDE, and mean-field control problems. Through a case study on 1D conservation laws, we will show ICON’s strong generalization capability to new PDEs, as well as its advantage compared with classic operator learning methods, e.g., Fourier neural operator (FNO). We will also show the application of ICON in 2D fluid problems, where a single model can make predictions for incompressible or compressible fluids, with different viscosity.

报告时间：2025年7月24日  15:00-17：00

腾讯会议：125 792 146 

报告人简介：Dr. Yang is currently an Assistant Professor in the Department of Mathematics at National University of Singapore (NUS), awarded the NUS Presidential Young Professorship. Before joining NUS, he was an Assistant Adjunct Professor in the Department of Mathematics at UCLA. He obtained his Ph.D. in Applied Mathematics from Brown University in 2021, and B.E. in Engineering Mechanics from Tsinghua University in 2016.

He is interested in building foundation AI models for scientific challenges, including multi-physics prediction, control design, inverse problems, etc. See more details on the website: https://scaling-group.github.io/