A recurrent neural network with a tunable activation function for solving k-winners-take-all
Peng Miao, Yanjun Shen, Jianshu Hou, Yi Shen · 2014
In this paper, a finite time recurrent neural network with a tunable activation function is presented to solve the k-winners-take-all problem. The activation function has two tunable parameters which give more flexibility to design neural network. By Lyapunov theorem, the proposed neural network model can converge to the equilibrium point in finite time. Comparing with the existing neural networks, the faster convergence speed can be obtained. Particularly, proposed neural network has high robustness against noise. The effectiveness of our methods is validated by theoretical analysis and numerical simulations.