A Genetic Algorithm Based Variable Structure Neural Network

This paper presents a neural network model with a variable structure, which is trained by genetic algorithm (CA). The proposed neural network consists of a Neural Network with a Node-to-Node Relationship (PR) and a Network Switch Controller (NSC). In the N'R, a modified neuron model with two activation functions in the hidden layer, and switches in its links are introduced. The NSC controls the switches in the N'R. The proposed neural network can model different input patterns with variable network structures. The proposed neural network provides better result and learning ability than traditional feed forward neural networks. Two application examples an XOR problem and hand-written pattern recognition are given to illustrate the merits of the proposed network.