Particle Swarm Optimization based Piecewise Linear Approximation of nonlinear functions

Abstract

Piecewise Linear Approximation (PLA) method is widely used for linearization of nonlinear functions. Various optimization algorithms can be used to find out the number of linear segments and their breakpoints. This study proposes to provide these parameters by using Particle Swarm Optimization (PSO). PLA is widely used for implementation of nonlinear activation function of Artificial Neural Networks (ANN). Thus, linearization of the tangent sigmoid function which is used in neural networks is proposed. After Linearization, Linearized activation function can be implemented on low cost processors.