Improved Evolutionary Extreme Learning Machines Based on Particle Swarm Optimization and Clustering Approaches

Abstract

Extreme Learning Machine (ELM) is a new learning method for single-hidden layer feedforward neural network (SLFN) training. ELM approach increases the learning speed by means of randomly generating input weights and biases for hidden nodes rather than tuning network parameters, making this approach much faster than traditional gradient-based ones. However, ELM random generation may lead to non-optimal performance. Particle Swarm Optimization (PSO) technique was introduced as a stochastic search through an n-dimensional problem space aiming the minimization (or the maximization) of the objective function of the problem. In this paper, two new hybrid approaches are proposed based on PSO to select input weights and hidden biases for ELM. Experimental results show that the proposed methods are able to achieve better generalization performance than traditional ELM in real benchmark datasets.