Tag
This paper traces step-by-step how PyTorch's automatic differentiation engine computes gradients for Physics-Informed Neural Network training, including the two levels of differentiation needed for physics residuals and parameter gradients, using a simple MLP and ODE example.