A Simpler Approach to Coefficient Regularized Support Vector Machines Regression

Abstract EN

We consider a kind of support vector machines regression (SVMR) algorithms associated with l q ( 1 ≤ q < ∞ ) coefficient-based regularization and data-dependent hypothesis space.

Compared with former literature, we provide here a simpler convergence analysis for those algorithms.

The novelty of our analysis lies in the estimation of the hypothesis error, which is implemented by setting a stepping stone between the coefficient regularized SVMR and the classical SVMR.

An explicit learning rate is then derived under very mild conditions.