One-shot training algorithm for self-feedback neural networks

Incorporation of a specific number of stable fixed points (attractors) in a neural network is an important issue in many applications, including image processing and pattern recognition. The vast majority of model requires hundred presentation of the patterns before the learning is converged. This increases the simulation time considerably and thus limit their practical applications. In this paper, a simple and one-shot training algorithm is presented to determine the value of network parameters to control the number of fixed points and simultaneously their stability characteristics in self-feedback neural networks (SFNN). A number of explicit relationships among network parameters such as self-feedback coefficients, input weight matrix and the number of equilibrium points, are obtained. Several simulations are provided to show the effectiveness of the analytical results presented in the paper.