Semi-supervised extreme learning machine with manifold and pairwise constraints regularization

Traditional kernel-based semi-supervised learning (SSL) algorithms usually have high computational complexity. Moreover, few SSL methods have been proposed to utilize both the manifold of unlabeled data and pairwise constraints effectively. In this paper, we first construct a unified SSL framework to combine the manifold regularization and the terms based on the pairwise constraints for semi-supervised classification tasks. Motivated by the effectiveness of extreme learning machine (ELM), we further utilize ELM to approximate the established kernel-based SSL framework. Finally, we present a fast semi-supervised extreme learning machine with manifold regularization and pairwise constraints. Experimental results on a variety of real-world data sets demonstrate the effectiveness of the proposed fast SSL algorithm.