Quadrature-based features for kernel approximation

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.

arXiv commentAccepted to NIPS 2018; 9 pages, 3 figures, Appendix: 4 pages, 2 figures

Cited by in corpus (10)

• Random Features for Kernel Approximation: A Survey on Algorithms, Theory, and Beyond
• Kaleidoscope: An Efficient, Learnable Representation For All Structured Linear Maps
• Pixelated Butterfly: Simple and Efficient Sparse training for Neural Network Models
• Fast Learning in Reproducing Kernel Krein Spaces via Signed Measures
• Gauss-Legendre Features for Gaussian Process Regression
• No-Regret Algorithms for Private Gaussian Process Bandit Optimization
• Towards a Unified Quadrature Framework for Large-Scale Kernel Machines
• Denoising Score Matching with Random Fourier Features
• Preconditioning Kaczmarz method by sketching
• Generalization Properties of hyper-RKHS and its Applications