A COMPARISON OF SOME ERROR-ESTIMATES FOR NEURAL-NETWORK MODELS

Authors
TIBSHIRANI R
Citation
R. Tibshirani, A COMPARISON OF SOME ERROR-ESTIMATES FOR NEURAL-NETWORK MODELS, Neural computation, 8(1), 1996, pp. 152-163
Citations number
13
Categorie Soggetti
Computer Sciences","Computer Science Artificial Intelligence",Neurosciences
Journal title
Neural computation → ACNP
ISSN journal
08997667
Volume
8
Issue
1
Year of publication
1996
Pages
152 - 163
Database
ISI
SICI code
0899-7667(1996)8:1<152:ACOSEF>2.0.ZU;2-J
Abstract
We discuss a number of methods for estimating the standard error of pr edicted values from a multilayer perceptron. These methods include the delta method based on the Hessian, bootstrap estimators, and the ''sa ndwich'' estimator. The methods are described and compared in a number of examples. We find that the bootstrap methods perform best, partly because they capture variability due to the choice of starting weights .