상세정보
학술대회
CP-decomposition with Tensor Power Method for Convolutional Neural Networks Compression
- 발행일
- 201702
- 출처
- International Conference on Big Data and Smart Computing (BigComp) 2017, pp.115-118
- 협약과제
- 16MS5200, 사용자 디지털 감성 DNA에 기반한 디지털생명체 기술 개발, 이승익
- 초록
- 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.
- KSP 제안 키워드
- Accuracy loss, Complex image, Compression method, Computation cost, Convolution neural network(CNN), Image classification, Power Method, Smart Phone, Smart devices, computational cost, fine-tuning