Interpretable operator learning and posthoc/by-design XAI
Codes and data used in the examples presented in the paper:
Interpreting and generalizing deep learning in physics-based problems with functional linear models
https://arxiv.org/abs/2307.04569
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Matlab codes: Matlab implementation uploaded.
Python codes: To be added in the future.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Instructions:
Run the main.m file. Flag_nn_approx should be set to 1 for NN-driven results and 0 for data-driven results.
Test case 1: Flag_method = 3 for EMNIST results and Flag_method = 1 for MNIST results.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Data:
The input training and test data (based on the data as well as the probed neural network) are available here:
https://drive.google.com/drive/folders/1lUkeI_QE9GZRx5APfZ_mMla04qr4xdg-?usp=drive_link
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Using the codes on your own data:
A sample Pytorch code is placed under pt2mat folder where you can see how the *.pt files from Pytorch can be loaded and probed for generating the NN-driven data. In this code, you can also see how in this case vtu files (simulation results from FEniCS) are loaded to create the input/output training data.