Skip to content

Weighted Sums of Random Kitchen Sinks: Replacing Minimization with Randomization in Learning

Authors: Ali Rahimi, Benjamin Recht

Published: 2008 (Conference Paper)

Source: Advances in Neural Information Processing Systems

Algorithm: Random Kitchen Sinks

Summary

Shows that randomly-weighted sums of random kitchen-sink basis functions (random Fourier features) can uniformly approximate any target function in a reproducing kernel Hilbert space, replacing costly convex minimization with a randomized learning algorithm. Establishes approximation bounds and demonstrates competitive empirical performance. Companion to the authors' 2007 random features paper.

Abstract

Primary

Paper papers.nips.cc

Tags

  • Random features

  • Kernel methods

  • Randomization

  • Function approximation

  • Learning theory

  • Machine learning