Title :
Neural network design based on decomposition of decision space
Author_Institution :
Dept. of Mech. Eng., Brunel Univ., Uxbridge, UK
Abstract :
The networks considered are multi-layer perceptrons. The method developed by the author describes how points in a classification volume could be separated by using Voronoi polygons and then removing the separation boundaries between polygons of the same type. Such an operation leaves behind a boundary that is necessary to separate the different types of regions. Unfortunately, the separation surface (decision surface) itself provides sufficient information to associate the surface with the network
Keywords :
computational geometry; decision theory; multilayer perceptrons; network synthesis; neural net architecture; pattern classification; separation; Voronoi polygons; classification volume; decision space decomposition; decision surface; multilayer perceptrons; neural network design; region types; separation boundaries; separation surface; Geometry; Mechanical engineering; Neural networks; Solid modeling; Testing; Transfer functions;
Conference_Titel :
Neural Information Processing, 1999. Proceedings. ICONIP '99. 6th International Conference on
Conference_Location :
Perth, WA
Print_ISBN :
0-7803-5871-6
DOI :
10.1109/ICONIP.1999.844015
