Image Classification Using Low-Rank Regularized Extreme Learning Machine

Abstract

Extreme learning machine (ELM), a least-square-based learning algorithm, is a competitive machine learning method and provides efficient unified learning solutions for the applications of classification and regression. However, most existing models do not consider the relationship between features. In this paper, a low-rank regularized extreme learning machine is proposed by imposing low-rank constraints with the extracted features that are related on ELM. By this method, our model preserves the global geometric structure and simultaneously encodes discriminant information of data. Meanwhile, it can attain the solution with minimum norm, which is very important for ELM model. Extensive experiments on four widely used face datasets illustrate that the proposed model achieves better performance than ELM algorithms with other regularization terms for image classification.