Hits:

Indexed by:期刊论文

Date of Publication:2009-12-01

Journal:ICIC Express Letters

Included Journals:EI、Scopus

Volume:3

Issue:4

Page Number:1093-1100

Abstract:Traditional parallel Artificial Neural Network(ANN) algorithm, though apopular method with robustness to recognize handwritten digits, is oftencomputationally inefficient as it needs lots of matrix or vector operations toget results, and the parallel computers are relatively difficult to use and maynot be accessible to most researchers. In this paper, we introduce a parallelrecognition algorithm based on LeNet,5 convolutional ANN architecture withGPU-acceleration, which maps ANN to GPU through CUD A. The analytical resultsdemonstrate that the parallel method can obtain a low error rate, speed uprecognition procedure, and provide ordinary users with a more feasible solution.ICIC International ? 2009.