A self-organizing map-based initialization for hybrid training of feedforward neural networks

This paper presents a novel hybrid algorithm for feedforward neural networks, called a self organizing map-based initialization for hybrid training based on a two stage learning approach. First stage, a structure learning scheme which includes adding hidden neurons is used to determine the network size. Second stage, a FN (fuzzy neighborhood)-based hybrid learning scheme which we have recently proposed is used to adjust the network parameters. In this approach the weights between input and hidden layers are firstly adjusted by Kohonen algorithm with fuzzy neighborhood, whereas the weights connecting hidden and output layers are adjusted using gradient descent method. Four simulation examples are provided to demonstrate the efficiency of the approach compared with other well-known and recently proposed learning methods.