Robust extreme learning machine in the presence of outliers by iterative reweighted algorithm

Highlights：

• A novel robust ELM is proposed and the resultant optimization can be implemented by iterative reweighted algorithm, termed as IRRELM. In each iteration, IRRELM is equivalent to solve a weighted ELM. Numerical experiments on several artificial datasets and real-world datasets demonstrate that the proposed IRRELM achieves superior generalization performance and robustness than other comparisons for modeling datasets containing outliers, especially in the cases of higher outliers level.

摘要

•This paper aim to address the robustness problem of extreme learning machine (ELM) for regression when confronting with samples in the presence of outliers. The non convex 2-norm loss function is introduced to overcome this dilemma, which sets a fixed penalty to any potential outliers for reducing their negative influences.•A novel robust ELM is proposed and the resultant optimization can be implemented by iterative reweighted algorithm, termed as IRRELM. In each iteration, IRRELM is equivalent to solve a weighted ELM. Numerical experiments on several artificial datasets and real-world datasets demonstrate that the proposed IRRELM achieves superior generalization performance and robustness than other comparisons for modeling datasets containing outliers, especially in the cases of higher outliers level.

论文评审过程：Received 4 July 2019, Revised 3 February 2020, Accepted 23 February 2020, Available online 18 March 2020, Version of Record 18 March 2020.