A Review of Network Compression based on Deep Network Pruning

DOI

10.2991/icmeit-19.2019.53How to use a DOI?

Keywords

Network pruning, Deep learning, Convolutional neural network.

Abstract

In recent years, the deep network has made considerable achievements in the field of computer vision and gradually becomes a hot research topic. The performance of the deep network is very good, however, due to its large size of parameters, high storage, and computational cost, it is hard to deploy the deep network on limited hardware platforms (such as mobile devices). The parameters of the model can express its complexity to some extent, but related studies have shown that not all parameters work in the model. Some parameters are useless, redundancy, and even degrade the performance of the model. Firstly, this paper sorts the results achieved by the scholars domestic and overseas in the field of deep network pruning, and sums up the pruning methods based on single weight granularity, kernel weight granularity and channel granularity; Then, summarizes the effect of the relevant pruning methods on a variety of public deep network models; Finally, it combs the achievements of the current researches and thoughts of network pruning, summarizes the important progress and discusses the future directions.

Copyright

© 2019, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Jie Yu
AU  - Sheng Tian
PY  - 2019/04
DA  - 2019/04
TI  - A Review of Network Compression based on Deep Network Pruning
BT  - Proceedings of the 3rd International Conference on Mechatronics Engineering and Information Technology (ICMEIT 2019)
PB  - Atlantis Press
SP  - 308
EP  - 319
SN  - 2352-538X
UR  - https://doi.org/10.2991/icmeit-19.2019.53
DO  - 10.2991/icmeit-19.2019.53
ID  - Yu2019/04
ER  -