This repository contains the project associated to the course of numerical approximation and neural networks from Bruno Desprès at Sorbonne university. More information on the course at : https://www.ljll.math.upmc.fr/despres/ (bottom of the page).

It proposes an implementation of : "Controlling oscillations in spectral methods by local artificial viscosity governed by neural networks." Lukas Schwander, Deep Ray, Jan S. Hesthaven https://www.sciencedirect.com/science/article/pii/S002199912100036X.

We didn't succeed to reproduce the results of the papers, probably because of the data generation. With our code, we achieved ad accuracy of 85% at predicting the class of regularity of the functions.