Quadrature-based features for kernel approximation
Machine Learning
2018-10-31 v4 Machine Learning
Abstract
We consider the problem of improving kernel approximation via randomized feature maps. These maps arise as Monte Carlo approximation to integral representations of kernel functions and scale up kernel methods for larger datasets. Based on an efficient numerical integration technique, we propose a unifying approach that reinterprets the previous random features methods and extends to better estimates of the kernel approximation. We derive the convergence behaviour and conduct an extensive empirical study that supports our hypothesis.
Cite
@article{arxiv.1802.03832,
title = {Quadrature-based features for kernel approximation},
author = {Marina Munkhoeva and Yermek Kapushev and Evgeny Burnaev and Ivan Oseledets},
journal= {arXiv preprint arXiv:1802.03832},
year = {2018}
}
Comments
Accepted to NIPS 2018; 9 pages, 3 figures, Appendix: 4 pages, 2 figures