Material for Newton Institute Tutorial 2023 by Luca Magri with assistance from PhD students Defne Ozan and Daniel Kelshaw for the codes.

First create a new virtual environment

$ python -m venv venv;
$ source venv/bin/activate

Note: You can equally use conda.

Install the requirements:

$ pip install -r requirements.txt

Optionally create a new ipykernel:

$ python -m ipykernel install --user --name=newton_workshop

Handouts for feedforward and convolutional neural networks (no physics): https://doi.org/10.5281/zenodo.7538419

Day 1: Feedforward neural networks; hard-physics-constrained neural networks for nonlinear waves.

Day 2: Convolutional neural networks (CNNs); soft-physics-constrained CNNs for super-resolution of turbulence (Navier-Stokes PDEs)

Machine vs human modelling

Feedforward neural networks

Physics-constrained neural networks by way of example

Nonlinear wave equations in acoustics and Hard and soft constraints

Convolutional neural networks (CNNs)

Physics-constrained CNNs by way of example

2D turbulence and Super-resolution

Enjoy the drinks reception