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

TL;DR

This paper introduces a CNN compression technique using CP-decomposition and Tensor Power Method, achieving significant memory and computational savings without accuracy loss.

Contribution

It presents a novel CNN compression method combining CP-decomposition with an iterative fine-tuning process for improved efficiency.

Findings

Reduces memory and computation costs significantly

Maintains accuracy after compression

Outperforms previous state-of-the-art methods

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.