High Speed (HS-K-WTA) Neural Network Model

1-WTA networks, called winner takes all, for selecting the largest element or the smallest one in a data set, are basic units of the self-organization neural network, and are also key parts of competitive learning.In this paper, a new Neural Network Model called HS-K-WTA is proposed. HS-K-WTA can realize k-Winners-Take-All with high computing speed. HS-K-WTA can identify the k larger elements or the k smaller ones in a data set. The algorithm complexity of HS-K-WTA has been studied carefully for the data set that are sampled from the uniform distribution. It has been proved that the convergence rate of HS-K-WTA algorithm is more quickly than that of Winstron.Based on the philosophy in Winstron and HS-K-WTA, a new Neural Network Model called HS-K-WTA-2 is proposed in this paper. HS-K-WTA-2 is with high computing speed as HS-K-WTA. The algorithm complexity of HS-K-WTA-2 has been also studied carefully for the data set that are sampled from the uniform distribution. It has been proved that the convergence rate of HS-K-WTA-2 algorithm is more quickly than that of Winstron and HS-K-WTA.The inherent properties of HS-K-WTA and HS-K-WTA-2 have been proved. The analysis results of HS-K-WTA and HS-K-WTA-2 algorithm have been discussed in detail. For the data set that are sampled from uniform distribution, standard normal distribution, negative exponential distribution (with λ =10,1,0.1), standard gamma distribution, Poisson distribution (with λ =5,10,100) respectively, and with different N values, the simulation research for HS-K-WTA and HS-K-WTA-2 algorithms to select k larger numbers between 1 to 20 has been derived; and the algorithm simulation results have been compared with Winstron algorithm ones. It is convinced that the convergent rate of HS-K-WTA and HS-K-WTA-2 algorithm is more quickly than that of WINSTRON algorithm.The array architecture realization of HS-K-WTA-2 and HS-K-WTA is presented also in this paper. Its architecture is simple and can be easily realizable. The hardware complexity of the array architecture of the HS-K-WTA-2 algorithm is same as HS-K-WTA. The hardware realization of HS-K-WTA-2 and HS-K-WTA is more quickly than that of Winstron[1] and K-Winners-Take-All. The hardware realization of HS-K-WTA-2 is more...