This code is the official implementation of

Hiroshi Morioka and Aapo Hyvärinen, Connectivity-Contrastive Learning: Combining Causal Discovery and Representation Learning for Multimodal Data. Proceedings of The 26th International Conference on Artificial Intelligence and Statistics (AISTATS2023).

If you are using pieces of the posted code, please cite the above paper.

Python3

Pytorch

To train the models in the paper, run this command:

python ccl_training.py

Set 'method' in the code either 'ccl' or 'cclalt'.

'ccl': Train by CCL

'cclalt': Train by CCLalt. Require 'pair' parameter

To evaluate the trained model, run:

python ccl_evaluation.py