高速神经网络HS-K-WTA-2的研究  

High-speed Neural Network HS-K-WTA-2

作　　者：朱红[1] 陈清华[2] 刘国岁[1] 

机构地区：[1]南京理工大学电子工程与光电技术学院,江苏南京210094 [2]南京理工大学计算机科学与技术学院,江苏南京210094

出　　处：《南京理工大学学报》2007年第1期89-91,共3页Journal of Nanjing University of Science and Technology

摘　　要：该文提出了一种新的K-Winners-Take-All神经网络:High-Speed-K-Winners-Take-All-2(HS-K-WTA-2)。HS-K-WTA-2以竞争学习算法为基础。HS-K-WTA-2能够从任何一个数集中识别出K个较大的数,或K个较小的数。该文给出HS-K-WTA-2算法及算法复杂度的分析结果。用专门为研究K-WTA神经网络开发的仿真程序对HS-K-WTA-2、HS-K-WTA和Winstrons进行仿真研究。结果显示:当所取的数集N较大时,HS-K-WTA-2要比Winstrons和HS-K-WTA速度更快。HS-K-WTA-2的硬件实现比Winston的硬件实现要简单,比HS-K-WTA的硬件实现复杂。A new K-Winners-Take-All neural network： High-Speed-Winners-Take-All-2 （HS-K- WTA-2） is presented. HS-K-WTA-2 can identify the larger elements （or smaller ones） in a data set. The analysis results about HS-WTA-2 algorithm and its complexity are given. HS-K-WTA-2, HS-K-WTA and Winstrons are simulated with a specific design simulation tool software for K-WTA neural network. The results show that the speed of HS-K-WTA-2 is much quicker than Winstrons and HS-K-WTA for a data set N that has a lot of atoms. Hardware implementention of HS-K-WTA- 2 is simpler than Winstrons, and more complicated than HS-K-WTA.

关 键 词：神经网络 竞争学习算法 高速算法 选择K个较大数 K—WTA HS-K-WTA HS-K-WTA-2 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]