A Functional Extension of Multi-Output Learning

Anonymous

Keywords: multi-output regression, operator-valued kernels, quantile regression, density level set estimation

TL;DR: We propose an extension of multi-output learning to a continuum of tasks using operator-valued kernels.

Abstract: When considering simultaneously a finite number of tasks, multi-output learning enables one to account for the similarities of the tasks via appropriate regularizers. We propose a generalization of the classical setting to a continuum of tasks by using vector-valued RKHSs.