Online regression with unbounded sampling

Abstract

In this paper, we consider a kernel-based online learning algorithm for regression when the sampling process is unbounded. Under a moment hypothesis on the sampling outputs, we provide a confidence-based bound for the error in the corresponding reproducing kernel Hilbert space.

Keywords:

• learning theory
• online learning
• regression
• reproducing kernel Hilbert spaces
• unbounded sampling process

2010 AMS Subject Classifications :

• 68T05
• 62J02