This video introduces PINNs, or Physics Informed Neural Networks. PINNs are a simple modification of a neural network that adds a PDE in the loss function to promote solutions that satisfy known physics. For example, if we wish to model a fluid flow field and we know it is incompressible, we can add the divergence of the field in the loss function to drive it towards zero. This approach relies on the automatic differentiability in neural networks (i.e., backpropagation) to compute partial derivatives used in the PDE loss function.
Original PINNs paper: https://www.sciencedirect.com/science/article/abs/pii/S0021999118307125
Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
M. Raissi P. Perdikaris, G.E. Karniadakis
Journal of Computational Physics
Volume 378: 686-707, 2019
This video was produced at the University of Washington, and we acknowledge funding support from the Boeing Company
%%% CHAPTERS %%%
00:00 Intro
01:54 PINNs: Central Concept
06:38 Advantages and Disadvantages
11:39 PINNs and Inference
15:23 Recommended Resources
19:33 Extending PINNs: Fractional PINNs
21:40 Extending PINNs: Delta PINNs
25:33 Failure Modes
29:40 PINNs & Pareto Fronts
31:57 Outro
source