Efficient Structured Pruning and Architecture Searching for Group Convolution

Ruizhe Zhao, Wayne Luk; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 0-0

Efficient inference of Convolutional Neural Networks is a thriving topic recently. It is desirable to achieve the maximal test accuracy under given inference budget constraints when deploying a pre-trained model. Network pruning is a commonly used technique but it may produce irregular sparse models that can hardly gain actual speed-up. Group convolution is a promising pruning target due to its regular structure; however, incorporating such structure into the pruning procedure is challenging. It is because structural constraints are hard to describe and can make pruning intractable to solve. The need for configuring group convolution architecture, i.e., the number of groups, to maximise test accuracy also increases difficulty. This paper presents an efficient method to address this challenge. We formulate group convolution pruning as finding the optimal channel permutation to impose structural constraints and solve it efficiently by heuristics. We also apply local search to exploring group configuration based on estimated pruning cost to maximise test accuracy. Compared to prior work, results show that our method produces competitive group convolution models for various tasks within a shorter pruning period and enables rapid group configuration exploration subject to inference budget constraints.