A Leave-One-out Cross Validation Bound for Kernel Methods with Applications in Learning

Authors:
Tong Zhang
Affiliations:
-
Venue:
COLT '01/EuroCOLT '01 Proceedings of the 14th Annual Conference on Computational Learning Theory and and 5th European Conference on Computational Learning Theory
Year:
2001

Citing 8
Cited 1

The nature of statistical learning theory

The nature of statistical learning theory
Error estimates for interpolation by compactly supported radial basis functions of minimal degree

Journal of Approximation Theory
Algorithmic stability and sanity-check bounds for leave-one-out cross-validation

Neural Computation
An introduction to support Vector Machines: and other kernel-based learning methods

An introduction to support Vector Machines: and other kernel-based learning methods
Relative Loss Bounds for Temporal-Difference Learning

ICML '00 Proceedings of the Seventeenth International Conference on Machine Learning
Relative Expected Instantaneous Loss Bounds

COLT '00 Proceedings of the Thirteenth Annual Conference on Computational Learning Theory
A Sequential Approximation Bound for Some Sample-Dependent Convex Optimization Problems with Applications in Learning

COLT '01/EuroCOLT '01 Proceedings of the 14th Annual Conference on Computational Learning Theory and and 5th European Conference on Computational Learning Theory
Stability and generalization

The Journal of Machine Learning Research

A Sequential Approximation Bound for Some Sample-Dependent Convex Optimization Problems with Applications in Learning

COLT '01/EuroCOLT '01 Proceedings of the 14th Annual Conference on Computational Learning Theory and and 5th European Conference on Computational Learning Theory

Quantified Score

Hi-index	0.01

Visualization

Abstract

In this paper, we prove a general leave-one-out style cross-validation bound for Kernel methods. We apply this bound to some classification and regression problems, and compare the results with previously known bounds. One aspect of our analysis is that the derived expected generalization bounds reflect both approximation (bias) and learning (variance) properties of the underlying kernel methods. We are thus able to demonstrate the universality of certain learning formulations.