Convergence of online learning algorithm with a parameterized loss

Abstract

The research on the learning performance of machine learning algorithms is one of the important contents of machine learning theory, and the selection of loss function is one of the important factors affecting the learning performance. In this paper, we introduce a parameterized loss function into the online learning algorithm and investigate the performance. By applying convex analysis techniques, the convergence of the learning sequence is proved and the convergence rate is provided in the expectation sense. The analysis results show that the convergence rate can be greatly improved by adjusting the parameter in the loss function.

Keywords

• online learning
• parameterized loss
• convergence rate
• reproducing hilbert space
• convex analysis