Pruning from Scratch via Shared Pruning Module and Nuclear norm-based Regularization

Official implementation of Pruning from Scratch via Shared Pruning Module and Nuclear norm-based Regularization (WACV 2024).

Abstract

Most pruning methods focus on determining redundant channels from the pre-trained model. However, they overlook the cost of training large networks and the significance of selecting channels for effective reconfiguration. In this paper, we present a "pruning from scratch" framework that considers reconfiguration and expression capacity. Our Shared Pruning Module (SPM) handles a channel alignment problem in residual blocks for lossless reconfiguration after pruning. Moreover, we introduce nuclear norm-based regularization to preserve the representability of large networks during the pruning process. By combining it with MACs-based regularization, we achieve an efficient and powerful pruned network while compressing towards target MACs. The experimental results demonstrate that our method prunes redundant channels effectively to enhance representation capacity of the network. Our approach compresses ResNet50 on ImageNet without requiring additional resources, achieving a top-1 accuracy of 75.3% with only 41% of the original model's MACs.

Comparison of Top-1 Accuracy according to MACs for compressed ResNet50 models on ImageNet

Overall method

Training & Reconfiguration

cd ./CLS

Prepare ImageNet dataset

1. Download dataset
2. Extract the training and validation dataset

sh ./scripts/extract_ILSVRC.sh

Training

sh ./scripts/train.sh

Weight

Download weight

• Accuracy : 75.3%
• MACs : 1.7G
• $\alpha_{mac}$ : 0.5
• $\alpha_{nuc}$ : 0.001

Reconfiguration

Reconfiguration is required for real network acceleration of structured pruning.

python reconfigure.py -b 256 -p 50 --state_dict <state_dict path>

Acknowledgements

This repository is built using DeepLearningExamples. This work is supported by Institute of Information & communications Technology Planning & Evaluation (IITP). (No. 2020-0-01077 and No. 2021-0-02068)

Citation

If you find our work is useful in your research, please consider citing:

@InProceedings{Lee_2024_WACV,
    author    = {Lee, Donghyeon and Lee, Eunho and Hwang, Youngbae},
    title     = {Pruning From Scratch via Shared Pruning Module and Nuclear Norm-Based Regularization},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
    month     = {January},
    year      = {2024},
    pages     = {1393-1402}
}