Validation of voting committees
Neural Computation
Machine Learning
Linearly Combining Density Estimators via Stacking
Machine Learning
Stacked generalization: when does it work?
IJCAI'97 Proceedings of the Fifteenth international joint conference on Artifical intelligence - Volume 2
Issues in stacked generalization
Journal of Artificial Intelligence Research
Hi-index | 0.00 |
Estimation of confidence intervals for neural network outputs isimportant when the uncertainty of a neural network system must beaddressed for safety or reliability. This paper presents a newapproach for estimating confidence intervals, which can help usersvalidate neural network outputs. The estimation of confidenceintervals, called error estimation by series association, isperformed by a supplementary neural network trained to predict theerror of the main neural network using input features and theoutput of the main network. The accuracy of this approach is shownusing a simple nonlinear mapping and more complicated, realisticnuclear power plant fault diagnosis problems. The resultsdemonstrate that the approach performs confidence estimationsuccessfully.