3 ALGORITHMS FOR ESTIMATING THE DOMAIN OF VALIDITY OF FEEDFORWARD NEURAL NETWORKS

被引:39
作者
COURRIEU, P
机构
[1] Centre National de la Recherche Scientifique, Université de Provence
关键词
INTERPOLATION DOMAIN; NEURAL NETWORKS; CONVEX POLYTOPE; CIRCUMSCRIBED SPHERE; GENERALIZATION RELIABILITY;
D O I
10.1016/0893-6080(94)90065-5
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
This article presents three simple algorithms for determining the distance between any point, and the domain of interpolation associated with a cluster of control points of a vectorial function. The first algorithm uses the convex hull polytope of the cluster in the support space to accurately estimate the domain. The second algorithm is a neuron-like good approximation of the first. When the number of vertices of the polytope is large, a more economical approach is to approximate the domain by its circumscribed sphere, which is what the third algorithm does. It is also shown that there is a significant relation between these three measures of the distance between any test point and a set of learning points, and the generalization errors made by an artificial neural network.
引用
收藏
页码:169 / 174
页数:6
相关论文
共 7 条