Pruning from Scratch
This work addresses the inefficiency of traditional pruning methods for machine learning practitioners by offering a more efficient approach, though it is incremental in rethinking existing techniques.
The paper tackles the problem of reducing computational costs in neural networks by proposing a pruning pipeline that eliminates the need for pre-training an over-parameterized model, achieving similar or higher accuracy on CIFAR10 and ImageNet datasets under the same computation budgets.
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.