This repository contains the implementation of the paper Network Pruning that Matters: A Case Study on Retraining Variants.
Duong H. Le, Binh-Son Hua (ICLR 2021)
In this work, we study the behavior of pruned networks under different retraining settings. By leveraging the right learning rate schedule in retraining, we demonstrate a counter-intuitive phenomenon in that randomly pruned networks could even achieve better performance than methodically pruned networks (fine-tuned with the conventional approach) in many scenariors. Our results emphasize the cruciality of the learning rate schedule in pruned network retraining – a detail often overlooked by practioners during the implementation of network pruning.
If you find the paper/code helpful, please cite our paper:
@inproceedings{
le2021network,
title={Network Pruning That Matters: A Case Study on Retraining Variants},
author={Duong Hoang Le and Binh-Son Hua},
booktitle={International Conference on Learning Representations},
year={2021},
url={https://openreview.net/forum?id=Cb54AMqHQFP}
}
To run the code:
- Copy the Imagenet/CIFAR-10 dataset to
./data
folder - Run
init.sh
- Download checkpoints here then uncompress it here
- Run the desired script in each subfolder.
Our implementation is based on the official code of HRank, Taylor Pruning, Soft Filter Pruning, Rethinking.