Pruning from Scratch

TL;DR

This paper introduces a novel network pruning method that allows for pruning from scratch without pre-training, reducing computational costs and achieving comparable or better accuracy than traditional methods.

Contribution

It demonstrates that effective pruning can be performed directly from randomly initialized weights, challenging the necessity of pre-training in pruning pipelines.

Findings

Pruning from scratch can match or outperform traditional pre-trained pruning methods.

The approach reduces pre-training time and computational resources.

Diverse pruned structures can be obtained directly from random initialization.

Abstract

Network pruning is an important research field aiming at reducing computational costs of neural networks. Conventional approaches follow a fixed paradigm which first trains a large and redundant network, and then determines which units (e.g., channels) are less important and thus can be removed. In this work, we find that pre-training an over-parameterized model is not necessary for obtaining the target pruned structure. In fact, a fully-trained over-parameterized model will reduce the search space for the pruned structure. We empirically show that more diverse pruned structures can be directly pruned from randomly initialized weights, including potential models with better performance. Therefore, we propose a novel network pruning pipeline which allows pruning from scratch. In the experiments for compressing classification models on CIFAR10 and ImageNet datasets, our approach not only greatly reduces the pre-training burden of traditional pruning methods, but also achieves similar or even higher accuracy under the same computation budgets. Our results facilitate the community to rethink the effectiveness of existing techniques used for network pruning.