A Novel Recurrent Neural Network with a Continuous Activation Function for Winner-Take-All

Qingshan Liu,Yan Zhao

DOI: https://doi.org/10.1109/ISMS.2013.14

2013-01-01

Abstract:In this paper, a novel recurrent neural network with a continuous activation function is proposed for solving the winner-take-all (WTA) problem. Compared with the existing WTA networks, the proposed network has a continuous activation function and lower model complexity. Moreover, global convergence of the proposed neural network is proved using the Lyapunov method. The WTA problem is first converted equivalently into a linear programming problem. Then a recurrent neural network with a single state variable is proposed to get the largest input of the WTA problem. In addition, simulation results on a numerical example show the effectiveness and performance of the proposed WTA network.