Incremental-based Extreme Learning Machine algorithms for Time-Variant Neural Networks

Extreme Learning Machine (ELM) is a novel learning algorithm for Neural Networks (NN) much faster than the traditional gradient-based learning techniques, and many variants, extensions and applications in the NN field have been appeared in the recent literature. Among them, an ELM approach has been applied to training Time-Variant Neural Networks (TV-NN), with the main objective to reduce the training time. Moreover, interesting approaches have been proposed to automatically determine the number of hidden nodes, which represents one of the limitations of original ELM algorithm for NN. In this paper, we extend the Error Minimized Extreme Learning Machine (EM-ELM) algorithm along with other two incremental based ELM methods to the time-variant case study, which is actually missing in the related literature. Comparative simulation results show the the proposed EM-ELM-TV is efficient to optimally determine the basic network architecture guaranteeing good generalization performances at the same time. © 2010 Springer-Verlag Berlin Heidelberg.