English

CP-decomposition with Tensor Power Method for Convolutional Neural Networks Compression

Machine Learning 2017-01-26 v1

Abstract

Convolutional Neural Networks (CNNs) has shown a great success in many areas including complex image classification tasks. However, they need a lot of memory and computational cost, which hinders them from running in relatively low-end smart devices such as smart phones. We propose a CNN compression method based on CP-decomposition and Tensor Power Method. We also propose an iterative fine tuning, with which we fine-tune the whole network after decomposing each layer, but before decomposing the next layer. Significant reduction in memory and computation cost is achieved compared to state-of-the-art previous work with no more accuracy loss.

Keywords

Cite

@article{arxiv.1701.07148,
  title  = {CP-decomposition with Tensor Power Method for Convolutional Neural Networks Compression},
  author = {Marcella Astrid and Seung-Ik Lee},
  journal= {arXiv preprint arXiv:1701.07148},
  year   = {2017}
}

Comments

Accepted as a conference paper at BigComp 2017

R2 v1 2026-06-22T17:59:28.585Z