• Complex
  • Title
  • Keyword
  • Abstract
  • Scholars
  • Journal
  • ISSN
  • Conference
搜索

Author:

Li, Y. (Li, Y..) | Shen, C. (Shen, C..) | Yang, H. (Yang, H..) (Scholars:杨宏) | Hu, H. (Hu, H..)

Indexed by:

Scopus PKU CSCD

Abstract:

To initialize convolutional neural networks better, an effective method named principal component analysis (PCA) Shuffling initialization was proposed. The method consisted of three steps. First, for the first convolutional layer, all receptive field of each feature map on training set was sampled. Then, principal component analysis of image patches separately for each feature map was conducted, and projection matrix was used to initialize filter of first convolutional layer. Finally, the first two steps on the other convolutional layers layer-wisely were performed. Experimental results on MNIST and CIFAR-10 dataset show that the proposed initialization has advantages of accuracy and speed of convergence compared to the common method such as random initialization and Xavier initialization. © 2017, Editorial Department of Journal of Beijing University of Technology. All right reserved.

Keyword:

Convolutional neural network; Initialization; Principal component analysis (PCA)

Author Community:

  • [ 1 ] [Li, Y.]College of Computer Science, Beijing University of Technology, Beijing, 100124, China
  • [ 2 ] [Shen, C.]College of Computer Science, Beijing University of Technology, Beijing, 100124, China
  • [ 3 ] [Yang, H.]College of Computer Science, Beijing University of Technology, Beijing, 100124, China
  • [ 4 ] [Hu, H.]College of Computer Science, Beijing University of Technology, Beijing, 100124, China

Reprint Author's Address:

Email:

Show more details

Related Keywords:

Related Article:

Source :

Journal of Beijing University of Technology

ISSN: 0254-0037

Year: 2017

Issue: 1

Volume: 43

Page: 22-27

Cited Count:

WoS CC Cited Count:

SCOPUS Cited Count: 2

ESI Highly Cited Papers on the List: 0 Unfold All

WanFang Cited Count:

Chinese Cited Count:

30 Days PV: 0

Online/Total:754/5299576
Address:BJUT Library(100 Pingleyuan,Chaoyang District,Beijing 100124, China Post Code:100124) Contact Us:010-67392185
Copyright:BJUT Library Technical Support:Beijing Aegean Software Co., Ltd.