Quadratic Residual Networks: A New Class of Neural Networks for Solving Forward and Inverse Problems in Physics Involving PDEs
The Quadratic Residual Network (QRes) architecture:
Most experiments are in Jupyter notebooks while functions and classes are defined in .py Python scripts.
For reference for PINN: repo, doc
General Requirements:
- PyTorch 1.5 (for most experiments)
- Tensorflow v1 etc (for reproducing PINN experiments, please follow the descriptions in PINN)
High frequency Reponse (learning with MSE loss function alone):
Error Map of Learning Navier-Stoke Equation:
