A Comparison of Some Error Estimates for Neural Network Models
- 1 January 1996
- journal article
- Published by MIT Press in Neural Computation
- Vol. 8 (1) , 152-163
- https://doi.org/10.1162/neco.1996.8.1.152
Abstract
We discuss a number of methods for estimating the standard error of predicted values from a multilayer perceptron. These methods include the delta method based on the Hessian, bootstrap estimators, and the “sandwich” 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.Keywords
This publication has 3 references indexed in Scilit:
- A neural network architecture that computes its own reliabilityComputers & Chemical Engineering, 1992
- A Practical Bayesian Framework for Backpropagation NetworksNeural Computation, 1992
- Connectionist learning proceduresArtificial Intelligence, 1989