The Nystr\"om method for convex loss functions

TL;DR

This paper extends the analysis of the Nyström method in kernel learning to general convex Lipschitz losses, demonstrating scenarios where computational efficiency is improved without losing learning accuracy, especially in classification tasks.

Contribution

It generalizes previous results to non-smooth convex losses like hinge loss, enabling better understanding of Nyström method's effectiveness across various loss functions.

Findings

Computational gains are achievable without sacrificing accuracy.

Analysis applies to non-smooth convex losses like hinge loss.

Results include bounds for classification error with Nyström approximations.

Abstract

We investigate an extension of classical empirical risk minimization, where the hypothesis space consists of a random subspace within a given Hilbert space. Specifically, we examine the Nystr\"om method where the subspaces are defined by a random subset of the data. This approach recovers Nystr\"om approximations used in kernel methods as a specific case. Using random subspaces naturally leads to computational advantages, but a key question is whether it compromises the learning accuracy. Recently, the tradeoffs between statistics and computation have been explored for the square loss and self-concordant losses, such as the logistic loss. In this paper, we extend these analyses to general convex Lipschitz losses, which may lack smoothness, such as the hinge loss used in support vector machines. Our main results show the existence of various scenarios where computational gains can be achieved without sacrificing learning performance. When specialized to smooth loss functions, our analysis recovers most previous results. Moreover, it allows to consider classification problems and translate the surrogate risk bounds into classification error bounds. Indeed, this gives the opportunity to compare the effect of Nystr\"om approximations when combined with different loss functions such as the hinge or the square loss.