Discovering PDEs from Multiple Experiments

In this repo, we share the code, data and results of the paper (https://arxiv.org/abs/2109.11939)

(1) To promote grouped sparsity:

we propose and implement a randomized adaptive group Lasso with stability selection and error control

see /pdeX/sparsity_estimators.py

(2) Deep learning based model discovery:

we implement the latter sparsity estimator in DeepMod (that we extend to handle multiple experiments)

we leverage JAX to perform backward and forward autodiffs

see /pdeX/DeepModx.py

(3) We share the code to reproduce the numerical experiments:

varying parameters (paramsXX.ipynb), varying initial conditions (ICs_XX.ipynb) and different chaotic regimes (chaos_XX.ipynb)

where XX = {GL: randomized adaptive Group Lasso (grouped sparsity) or IL: randomized adaptive Lasso (individual sparsity)}

Requirements: conda and pip requirements are shared (see .txt files)