Clone submodules and install requirements using
git clone --recurse-submodules https://github.com/ngruver/lie-deriv.git
cd lie-deriv
pip install -r requirements.txtThe equivariance of individual models can be calculated using exps_layerwise.py, for example
python exps_layerwise.py \
--modelname=resnet50 \
--output_dir=$HOME/lee_results \
--num_imgs=20 \
--num_probes=100 \
--transform=translation Default models and transforms are available in our wandb sweep configuration
python exps_e2e.py \
--modelname=resnet50 \
--output_dir=$HOME/lee_results \
--num_datapoints=100 We also include the wandb sweep configuration for our end-to-end equivariance experiments.
We make our results and plotting code available in a google colab notebook
If you find our work helpful, please cite it with
@misc{https://doi.org/10.48550/arxiv.2210.02984,
doi = {10.48550/ARXIV.2210.02984},
url = {https://arxiv.org/abs/2210.02984},
author = {Gruver, Nate and Finzi, Marc and Goldblum, Micah and Wilson, Andrew Gordon},
title = {The Lie Derivative for Measuring Learned Equivariance},
publisher = {arXiv},
year = {2022},
copyright = {Creative Commons Attribution 4.0 International}
}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent