IKA: Independent Kernel Approximator

TL;DR

This paper introduces IKA, a novel low-rank kernel approximation method that constructs the approximation as a linear combination of arbitrary functions, outperforming Nyström in experiments on STL-10.

Contribution

The paper presents IKA, a new kernel approximation technique that offers more flexible function bases and demonstrates superior performance over Nyström in empirical tests.

Findings

IKA outperforms Nyström on STL-10 dataset

The method produces more flexible kernel approximations

Numerical results are reproducible with available source code

Abstract

This paper describes a new method for low rank kernel approximation called IKA. The main advantage of IKA is that it produces a function $\psi(x)$ defined as a linear combination of arbitrarily chosen functions. In contrast the approximation produced by Nystr\"om method is a linear combination of kernel evaluations. The proposed method consistently outperformed Nystr\"om method in a comparison on the STL-10 dataset. Numerical results are reproducible using the source code available at https://gitlab.com/matteo-ronchetti/IKA