Solving fluid dynamics with neural networks | Vignesh Gopakumar

Presenter : Vignesh Gopakumar, Chief scientific machine learning engineer at UKAEA, United Kingdom.

💬 Contact us directly on https://www.m.me/fusioneptalks
or use the comments to ask questions!

Date: May 28th, 2020

Synopsis: Vignesh Gopakumar is a scientific machine learning engineer at the UK Atomic Energy Agency and an alumni of the European Master in Fusion Science and Engineering Physics. He uses neural networks as a surrogate for the fluid equations describing the behavior of fusion plasma in JET and MAST-U. In his FusionEPtalk, Vignesh will introduce the neural network solvers for partial differential equations. These regression models, which provide solutions while preserving most of the underlying physics, are particularly efficient in data-starved physical scenarios. Join the webinar and learn how to use his group's state-of-the-art python package for solving PDEs with artificial neural networks.

Chapters

00:00 - Introduction
03:40 - Overview of the talk
05:40 - What are Neural networks?
07:00 - Loss functions
14:36 - Loss functions with physics penalty
16:20 - Neural PDE layout
22:24 - Calculations of partial derivatives
24:00 - NPDE package - PDE_Kozhi
31:00 - Output of Neural networks
34:10 - Theory and Error Analysis for Neural PDEs
36:30 - Numerical solvers vs Neural PDEs
39:00 - Deep hidden physics model
47:40 - Q&A session
🌍 Fusion-EP Talks is a student-led seminar of the European Master In Fusion Science and Engineering program - 📚 join our international degree @ http://www.em-master-fusion.org

🕗 COMING UP NEXT on June 18th at 18:00 (CET, Prague): Steve Jepeal will talk about predicting fusion radiation damage using protons. More details about the event are available at https://fusionep-talks.egyplasma.com/events/event.php?eventID=23

Видео Solving fluid dynamics with neural networks | Vignesh Gopakumar канала FusionEPtalks