Physics-guided deep learning methods and applications
Deep learning (DL)-based methods have been widely applied across various science and engineering disciplines. This course focuses on deep learning techniques within the context of AI for Science, with a specific emphasis on their applications in applied geophysics. The curriculum starts with some broadly applicable methodologies drawn from the latest literature, which are general and not limited to specific fields, and then introduces some of their improvements, extensions, and applications. In addition, the lecturer will discuss some specific issues and the answers provided by AI tools and also introduce some related ideas and research work. All attendees are welcome to ask questions and communicate with each other.
Lecturer
Date
13th October ~ 29th December, 2026
Location
| Weekday | Time | Venue | Online | ID | Password |
|---|---|---|---|---|---|
| Tuesday | 13:30 - 16:55 | Shuimo | ZOOM 09 | 230 432 7880 | BIMSA |
Prerequisite
Basic knowledge on deep learning methods, geophysics, and the Python language.
Syllabus
1. Introduction to neural network (NN) architectures, physics-guided deep learning, current mainstream topics
Basic methodology:
2. Physics-informed neural networks (PINNs)
3. Operator learning (a): DeepONet, FNO, and Transformer
4. Operator learning (b): Design NN’s input, auxiliary sample approach
Latest updates and important applications:
5. Selection of training samples in operator learning
6. Some applications related to operator learning
7. Equation discovery and field reconstruction
8. Neural network surrogate modeling method
AI for Geophysics:
9. NN-based inversion methods and auxiliary sample approach
10. Seismic rock physics modeling and parameter inversion
11. Application in full wave inversion (FWI) and ambient noise tomography (ANT)
12. Some research results, communication and interaction
Basic methodology:
2. Physics-informed neural networks (PINNs)
3. Operator learning (a): DeepONet, FNO, and Transformer
4. Operator learning (b): Design NN’s input, auxiliary sample approach
Latest updates and important applications:
5. Selection of training samples in operator learning
6. Some applications related to operator learning
7. Equation discovery and field reconstruction
8. Neural network surrogate modeling method
AI for Geophysics:
9. NN-based inversion methods and auxiliary sample approach
10. Seismic rock physics modeling and parameter inversion
11. Application in full wave inversion (FWI) and ambient noise tomography (ANT)
12. Some research results, communication and interaction
Reference
1. The latest published articles related to the topic of AI for Science.
2. Some notes summarized from articles, and the discussion results provided by AI tools.
3. The ideas, practical applications, and research results of the lecturer.
2. Some notes summarized from articles, and the discussion results provided by AI tools.
3. The ideas, practical applications, and research results of the lecturer.
Audience
Graduate
, Postdoc
, Researcher
Video Public
No
Notes Public
No
Language
Chinese
Lecturer Intro
Fansheng Xiong (熊繁升) is currently an Assistant Professor of BIMSA. Before that, he received his doctoral degree in 2020 from Tsinghua University, and he was a visiting research assistant at Yale University during 2018-2019. His research interest mainly focuses on solving forward/inverse problems of PDEs based on AI methods, and their application in geophysics (especially seismic rock physics) and computational mathematics. He is PI for grant from National Natural Science Foundation of China, and he has published papers in journals like Journal of Geophysical Research-Solid Earth, Geophysical Journal International, and Geophysics.