Local Elasticity

This is the code repository for the ICLR paper The Local Elasticity of Neural Networks. If you use this code for your work, please cite

@article{he2019local,
  title={The Local Elasticity of Neural Networks},
  author={He, Hangfeng and Su, Weijie J},
  journal={ICLR},
  year={2020}
}

Installing Dependencies

Use virtual environment tools (e.g miniconda) to install packages and run experiments
python>=3.6
pip install -r requirements.txt

Change the Dir Path

Change the /path/to/experiments/dir to your experiments dir in autoencoders_clustering.py, elastic_effect_clustering.py, elastic_effect_synthetic.py and train_models.py
You need to create corresponding directories in your experiments dir, such as models, data and figures.

Reproducing experiments

To reproduce our simulations

sh scripts/run_elastic_effect_synthetic.sh

To reproduce the results of the local elasticity based clustering on MNIST

sh scripts/run_train_models_MNIST.sh
sh scripts/run_elastic_effect_clustering_MNIST.sh

To reproduce the results of the local elasticity based clustering on CIFAR10

sh scripts/run_train_models_CIFAR10.sh
sh scripts/run_elastic_effect_clustering_CIFAR10.sh

To reproduce the results of different architectures (CNN and ResNet) based clustering on MNIST

sh scripts/run_train_models_architectures.sh
sh scripts/run_elastic_effect_clustering_architectures.sh

To reproduce the results of autoencoder based clustering on MNIST

sh scripts/run_autoencoder_clustering.sh

To reproduce the results of pre-trained ResNet152 based clustering on MNIST

Uncomment clustering_images_resnet152(train_data, index_list, train_data_label[index_list]) in sources/elastic_effect_clustering.py