Physics Informed Neural Networks (PINNs) [Physics Informed Machine Learning]

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