Least correntropic loss regression

Abstract

Robust linear regression is one of the well known areas in data analysis, and various methods to solve the robust regression problems are available in the literature. However, one of the key issues in these methods is the adaptability of the scale/tuning parameter to the data demographics. In this work, a correntropic loss based linear regression model is proposed. An approximation and simplification of the model reduces the model to the well known class of weighted linear regression models. Iterative solution methodology is proposed to solve the proposed formulation. Performance of the proposed approach is evaluated on simulated data. Results of the experiments highlight the usability and importance of the proposed approach.