Injecting knowledge into the solution of the two-spiral problem

被引：28

作者：

Alvarez-Sánchez, JR ^{[1
]}

机构：

[1] Univ Nacl Educ Distancia, Fac Ciencias, Dept Inteligencia Artificial, E-28040 Madrid, Spain

来源：

NEURAL COMPUTING & APPLICATIONS | 1999年 / 8卷 / 03期

关键词：

backpropagation; benchmark; generalisation; knowledge based design; neural networks; two spirals of Wieland;

D O I：

10.1007/s005210050029

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

Wieland's two-spiral problem is often used as a test for comparing the quality of different supervised-learning algorithms and architectures. In this paper, we use this two-spiral problem to illustrate the advantages obtained from using all the additional knowledge about the problem domain in designing the neural net which solves a given problem. The characteristics of the knowledge-based net, with regard to complexity, number of elements, training speed and generalisation quality, make it appreciably better than alternative nets which make no use of this knowledge.

引用

页码：265 / 272

页数：8