A Fast Winner-Take-All Neural Networks with the Dynamic Ratio

陳奇銘(Chi-Ming Chen) ； 徐明宏(Ming-Hung Hsu) ； 王天佑(Tien-Yo Wang)

《Journal of Information Science and Engineering》 18卷2期 (2002/03) Pp. 211-222

https://doi.org/10.6688/JISE.2002.18.2.3

並列摘要

In this paper, we propose a fast winner-take-all (WTA) neural network. The fast winner-take-all neural network with the dynamic ratio in mutual-inhibition is developed from the general mean-based neural network (GEMNET), which adopts the mean of the active neurons as the threshold of mutual inhibition. Furthermore, the other winner-take-all neural network enhances the convergence speed to become a decimal system. The proposed WTA neural networks statistically achieve the large ratio of mutual inhibition. The new WTA Neural Networks converge faster than the existing WTA neural networks for a large number of competitors based on both theoretical analyses and simulation results.

並列關鍵字

winner-take-all ； neural network ； convergence speed ； decimal system ； mutual inhibition