Abstract Extreme Learning Machine (ELM) has attracted significant concern in recent years due to its good generalization performance and fast running speed. However, it is very sensitive to modeling data in presence of non-Gaussian noise and outliers, resulting in poor learning performance. In this paper, we incorporate correntropy loss and the distribution of modeling errors in ELM training, and propose a robust ELM that minimizes the mean and variance of modeling errors with maximum correntropy criterion (MRELM-MCC), aimed at improving modeling performance in a noisy environment. Cor rentropy, as a robust generalized nonlinear similarity measure, is developed to reflect the variance of errors in the modeling process. The powerful half quadratic optimization technique is employed to achieve the optimal solution of MRELM-MCC. The robust analysis of the iterative algorithm ensures the efficiency of model training. Experimental results on benchmark datasets with varying outliers ratios, along with parameter sensitivity analysis, demonstrate that the proposed method can derive better generalization performance compared with the competing algorithms.
Lin et al. (Thu,) studied this question.