This repo contains the implementation and the experiments for the paper

Residual Pathway Priors for Soft Equivariance Constraints

To run the scripts you will instead need to clone the repo and install it locally which you can do with

git clone https://github.com/mfinzi/residual-pathway-priors.git
cd residual-pathway-priors
pip install -e .

• To reproduce the reinforcment learning results from the paper, see the RL/ directory.
• To reproduce the results in cases with exact symmetries (Figure 2a), see experiments/perfect-symmetry/
• To reproduce the results in cases with approximate symmetries (Figures 2b & 7), see experiments/prior-var-ablation/
• To reproduce the results in cases with mis-specified symmetries (Figure 2c), see experiments/misspec-symmetry/
• To reproduce the UCI results in Table 1, see experiments/UCI/
• To reproduce the CIFAR-10 resultsin Table 1, see experiments/cifar/

If you find our work helpful, cite it with

@inproceedings{finzi2021residual,
  title={Residual Pathway Priors for Soft Equivariance Constraints},
  author={Finzi, Marc and Benton, Gregory and Wilson, Andrew G},
  booktitle={Advances in Neural Information Processing Systems},
  volume={34},
  year={2021}
}