Learning Conditional Invariances through Non-Commutativity

Official implementation of "Learning Conditional Invariances through Non-Commutativity", ICLR 2024.

The current version provides a non-commutative version of Domain Adversarial Neural Networks. In theory, any invariance learning algorithm that has an associated commutative operator (Definition 2 in the main paper), can be adapted to have an NCI variant.

Environment Setup

1. Clone the project repository:

git clone https://github.com/abhrac/nci.git

2. Install dependences:

pip install -r requirements.txt

3. Run

python main.py --algorithm=NCI --data_dir=path/to/dataset/root --dataset=PACS --uda_holdout_fraction=0.2 --task=domain_adaptation --batch_size=64

Citation

@inproceedings{
  chaudhuri2024nci,
  title={Learning Conditional Invariances through Non-Commutativity},
  author={Abhra Chaudhuri, Serban Georgescu, Anjan Dutta},
  booktitle={The Twelfth International Conference on Learning Representations},
  year={2024},
  url={https://openreview.net/forum?id=tUVG9nGzgE}
}

Acknowledgements

Experimentation framework adapted from DomainBed.