UPDATE in 2024 Nov 11th. The code is used for a computer session of the course AMCS304, Deep learning and analysis, taught by Prof. Jinchao Xu at KAUST. Some functions are optimized in this updated version. Added a 1D Neumann problem as a demonstration for solving PDEs.
This repository contains code for the programming assignment from the summer lectures by Prof. Jinchao Xu at CIRM, Marseille, France, July 18th 2023.
The Colab environment suggeseted by the organizers, http://smai.emath.fr/cemracs/cemracs23/installation_instructions.html, is good for running the code in this repository.
Three examples are covered. The detailed programming exercises are in the pdf file ComputerSession.pdf.
- Gradient descent for a 3 by 3 system, and a 4 by 4 system. (GD4expandedSystem.ipynb)
- Use the gradient descent method to train ReLU shallow neural networks for 1D
$L^2$ fitting tasks. (L2Fitting1D_GD.ipynb) - Use an orthogonal greedy algorithm to train ReLU shallow neural networks for 1D
$L^2$ fitting tasks. (L2Fitting1D_OGA.ipynb)
The two other files code_explanation_gd.ipynb, code_explanation_oga.ipynb are the computer session materials that break down the code in L2Fitting1D_GD.ipynb, L2Fitting1D_OGA.ipynb into easily accessible parts for teaching purposes.