An improved hybrid regularization approach for extreme learning machine

Abstract

Extreme learning machine (ELM) is a network model that arbitrarily initializes the first hidden layer and can be computed speedily. In order to improve the classification performance of ELM, a ${\ell }_2$ and ${\ell }_{0.5}$ regularization ELM model (${\ell }_2$-${\ell }_{0.5}$-ELM) is proposed in this paper. An iterative optimization algorithm of the fixed point contraction mapping is applied to solve the ${\ell }_2$-${\ell }_{0.5}$-ELM model. The convergence and sparsity of the proposed method are discussed and analyzed under reasonable assumptions. The performance of the proposed ${\ell }_2$-${\ell }_{0.5}$-ELM method is compared with BP, SVM, ELM, ${\ell }_{0.5}$-ELM, ${\ell }_1$-ELM, ${\ell }_2$-ELM and ${\ell }_2$-${\ell }_1$ELM, the results show that the prediction accuracy, sparsity, and stability of the ${\ell }_2$-${\ell }_{0.5}$-ELM are better than the other $7$ models.