Accelerating Convolutional Neural Networks for Mobile Applications

Peisong Wang, Jian Cheng
2016 Proceedings of the 2016 ACM on Multimedia Conference - MM '16  
Convolutional neural networks (CNNs) have achieved remarkable performance in a wide range of computer vision tasks, typically at the cost of massive computational complexity. The low speed of these networks may hinder realtime applications especially when computational resources are limited. In this paper, an efficient and effective approach is proposed to accelerate the test-phase computation of CNNs based on low-rank and group sparse tensor decomposition. Specifically, for each convolutional
more » ... ayer, the kernel tensor is decomposed into the sum of a small number of low multilinear rank tensors. Then we replace the original kernel tensors in all layers with the approximate tensors and fine-tune the whole net with respect to the final classification task using standard backpropagation. Comprehensive experiments on ILSVRC-12 demonstrate significant reduction in computational complexity, at the cost of negligible loss in accuracy. For the widely used VGG-16 model, our approach obtains a 6.6× speed-up on PC and 5.91× speed-up on mobile device of the whole network with less than 1% increase on top-5 error.
doi:10.1145/2964284.2967280 dblp:conf/mm/WangC16 fatcat:pgox4okm6ndrtaidwjgsvisb4q