This repo holds the codes for DeepHoyer: Learning Sparser Neural Network with Differentiable Scale-Invariant Sparsity Measures.
@inproceedings{
yang2020deephoyer,
title={DeepHoyer: Learning Sparser Neural Network with Differentiable Scale-Invariant Sparsity Measures},
author={Huanrui Yang and Wei Wen and Hai Li},
booktitle={International Conference on Learning Representations},
year={2020},
url={https://openreview.net/forum?id=rylBK34FDS}
}
The codes for MNIST, CIFAR-10 and ImageNet experiments are within mnist/, cifar/ and imagenet/ folder respectively. Please follow the README file in each folder to run the experiments. Codes are tested with Pytorch 1.2.0 and Python 3.6.8, tqdm package is preferred for better visualization of the training process.
The codes of the MNIST experiments are adapted from Deep-Compression-PyTorch.
The codes of the CIFAR-10 experiments are adapted from bearpaw/pytorch-classification.
The codes of the ImageNet experiments are adapted from pytorch/examples.