ML4PHY: Physics-Informed Machine Learning
Vikiran Talk, Lecture Hall, NISER, Bhubaneswar, Odisha, India
The talk was aimed at introducing audience to a class of neural network achitecture called PINN. The talk can mainly be divided into three parts. First and foremost, introduction to ML including definition of some standard terms. An idea of bias-variance trade-off was also provided. This takes us to next part of talk, what are PINN exactly, and how does it hep reduce variance (or generalization error). The talk concludes with showing application of this architecture and its close relatives- Hamiltonian Neural Network & Lagrangian Neural Network to show how these techniques can be leveraged to learn dynamics of system from data itself.