Extreme Learning Machine Based on Evolutionary Multi-objective Optimization

Abstract

Extreme learning machine (ELM), which proposed for generalized single-hidden layer feedforward neural networks, has become a popular research topic due to its fast learning speed, good generalization ability, and ease of implementation. However, ELM faces redundancy and randomness in the hidden layer which caused by random mapping of features. In ELM, although evolutionary algorithms have archived impressive improvement, they have not considered the sparsity of the hidden layers. In this paper, a hybrid learning algorithm is proposed, termed EMO-ELM, which adopts evolutionary multi-objective algorithm to optimise two conflict objectives simultaneously. Furthermore, the proposed method can be used for supervised classification and unsupervised sparse feature extraction tasks. Simulations on many UCI datasets have demonstrated that EMO-ELM generally outperforms the original ELM algorithm as well as several ELM variants in classification tasks, moreover, EMO-ELM achieves a competitive performance to PCA in sparse feature extraction tasks.