Twin extreme learning machines for pattern classification

Abstract

Extreme learning machine (ELM) is an efficient and effective learning algorithm for pattern classification. For binary classification problem, traditional ELM learns only one hyperplane to separate different classes in the feature space. In this paper, we propose a novel twin extreme learning machine (TELM) to simultaneously train two ELMs with two nonparallel classification hyperplanes. Specifically, TELM first utilizes the random feature mapping mechanism to construct the feature space, and then two nonparallel separating hyperplanes are learned for the final classification. For each hyperplane, TELM jointly minimizes its distance to one class and requires it to be far away from the other class. TELM incorporates the idea of twin support vector machine (TSVM) into the basic framework of ELM, thus TELM could have the advantages of the both algorithms. Moreover, compared to TSVM, TELM has fewer optimization constraint variables but with better classification performance. We also introduce a successive over-relaxation technique to speed up the training of our algorithm. Comprehensive experimental results on a large number of datasets verify the effectiveness and efficiency of TELM.