Samy Wu Fung, Department of Applied Mathematics and Statistics, Colorado School of Mines

Designing Explainable Neural Networks with Physics Constraints via Optimization

This talk explores designing neural networks that enforce physics principles as hard constraints through a class of network architectures known as implicit neural networks. These networks produce outputs satisfying specific feasibility and optimality conditions that represent domain/physics knowledge. I'll address the challenges of training these networks efficiently, illustrate some applications where they've found success, and provide a bound on how well they generalize.