This repository provides a PyTorch implementation of the paper "Scalable Infomin Learning", NeurIPS 2022.
We consider learning representation with the following objective:
To optimise this objective, traditionally we need to (re-)estimate
- Python 3.5+
- Pytorch 1.12.1
- Torchvision 0.13.1
- Numpy, scipy, matplotlib
We strongly recommend to use conda to manage/update library dependence:
conda install pytorch torchvision matplotlib
Please run the following script to download the PIE dataset (contributed by https://github.com/bluer555/CR-GAN)
bash scripts/download_pie.sh
at /mi
- Pearson Correlation
- Distance Correlation
- Neural Total Correlation
- Neural Renyi Correlation
- CLUB
- Sliced independence testing (*ours)
at /tasks
- Fairness
- Disentangled representation learning
- Domain adaptation