L1-Norm Robust Regularized Extreme Learning Machine with Asymmetric C-Loss for Regression

Extreme learning machines (ELMs) have recently attracted significant attention due to their fast training speeds and good prediction effect. However, ELMs ignore the inherent distribution of original samples, they are prone overfitting, which fails at achieving generalization performance. In this paper, based on expectile penalty correntropy, an asymmetric C-loss function (called AC-loss) is proposed, non-convex, bounded, relatively insensitive noise. Further, a novel extreme machine called L1 norm robust regularized with (L1-ACELM) presented handle overfitting problem. The proposed algorithm benefits from replaces square loss AC-loss function. L1-ACELM can generate more compact network fewer hidden nodes reduce impact To evaluate effectiveness noisy datasets, different levels noise added in numerical experiments. results for types artificial benchmark datasets demonstrate that achieves better performance compared other state-of-the-art algorithms, especially when exists datasets.