Sparsity-accuracy trade-off in MKL

TL;DR

This paper empirically explores the balance between sparsity and accuracy in multiple kernel learning, revealing how the optimal trade-off depends on kernel spectrum sparsity, dependence, and sample size.

Contribution

It provides new insights into how the trade-off parameter in MKL should be chosen based on data properties, using elastic-net regularization.

Findings

Optimal trade-off depends on kernel spectrum sparsity

Dependence among kernels affects the trade-off

Sample size influences the sparsity-accuracy balance

Abstract

We empirically investigate the best trade-off between sparse and uniformly-weighted multiple kernel learning (MKL) using the elastic-net regularization on real and simulated datasets. We find that the best trade-off parameter depends not only on the sparsity of the true kernel-weight spectrum but also on the linear dependence among kernels and the number of samples.