Optimization of artificial neural network architecture using neuroplasticity

Abstract

Artificial neural networks, which are inspired by the behavior of central nervous system have the capability of finding good generalized solutions for many real world problems due to their characteristics such as massively parallel, ability to learn and adapt to the environment by altering the synaptic weights. However, despite of all the advantages of artificial neural networks, determining the most appropriate architecture for the given problem still remains as an unsolved problem. This paper presents a pruning method based on the backpropagation algorithm to solve this problem. The pruning method is inspired by the concepts of neuroplasticity and experimental results show that the proposed method approaches the minimal architecture faster than the other existing methods.