An Improved Indefinite Kernel Machine Regression Algorithm with Norm-r Loss Function

Abstract

Indefinite kernel machine regression algorithm (IKMRA), in which only constrains the minimum total regression error, but each sample point regression error is ignored. Thus the accuracy and the generalization performance of the IKMRA can not be satisfied. In order to improve the precision and the generalization performance of the IKMRA, we proposed that each sample regression error be constrained besides the total regression error. We introduced the norm-r loss function and the slack variables in order to constrain each sample regression error, derived the iterative formula of corresponding gradient decent method and devised the corresponding algorithm. Experimental results show that our improved indefinite kernel machine regression algorithm (IIKMRA) is effective and feasible.