An improved extreme learning machine tuning by flower pollination algorithm

Abstract

The second generation of algorithms intended for neural networks is named extreme learning machines (ELMs). Since the computing of output weights of ELM encounters the outliers problems, we advise a recently introduced Flower Pollination Algorithm (FPA) for accurately tuning synaptic input weights of ELM. The hybridization between ELM and FPA provides robust FPA-ELM approach which can efficiently solve outlier problems as well as it can significantly reduce the size of latent nodes. Extensive simulation results based on 16 well-known benchmark problems were conducted to reveal the effectiveness of the stated hybridization. Furthermore, it has been proved that our FPA-ELM approach is superior to other state-of-the-art algorithms from literature and that it can learn much faster weight coefficients compared to the other traditional learning methods, as well.