Bootstrap for model selection: linear approximation of the optimism

Authors:
G. Simon;A. Lendasse;M. Verleysen
Affiliations:
Université Catholique de Louvain, DICE, Louvain-la-Neuve, Belgium;Université Catholique de Louvain, CESAME, Louvain-la-Neuve, Belgium;Université Catholique de Louvain, DICE, Louvain-la-Neuve, Belgium
Venue:
IWANN'03 Proceedings of the Artificial and natural neural networks 7th international conference on Computational methods in neural modeling - Volume 1
Year:
2003

Citing 3
Cited 1

Introduction to statistical pattern recognition (2nd ed.)

Introduction to statistical pattern recognition (2nd ed.)
Pattern Recognition and Neural Networks

Pattern Recognition and Neural Networks
Natural Gradient and Multiclass NLDA Networks

ICANN '02 Proceedings of the International Conference on Artificial Neural Networks

Fast bootstrap methodology for regression model selection

Neurocomputing

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this work we shall discuss how to apply classical input relevance results for linear Fisher discriminants to measure the relevance of the linear last hidden layer of a Non Linear Discriminant Analysis (NLDA) network. We shall quickly review first possible ways to extend classical and non linear Fisher analysis to multiclass problems and introduce a criterion function very well suited computationally to NLDA networks. After defining a relevance statistic for linear NLDA units, we shall numerically illustrate the resulting procedures on a synthetic 3 class classification problem.