The parallel degeneration of a system of classification hyperplanes

Author

Junying, Zhang ; Zheng, Bao

Author_Institution

Inst. of Electron. Eng. Res., Xidian Univ., Xi´´an, China

Volume

2

fYear

1998

fDate

1998

Firstpage

1265

Abstract

It is known that a neuron with hard limitation activation function functions as a classification hyperplane (CH), and that binary training samples can be used to train the network such that the classification hyperplanes of hidden neurons are in a parallel pattern. This paper deals with a basic property of such a system of classification hyperplanes, with its parallel degeneration defined, and its parallel degeneration condition revealed, being that if and only if there exists a subsystem of four adjacent classification hyperplanes not capable of parallel degeneration, the system of classification hyperplanes is not capable of parallel degeneration

Keywords

feedforward neural nets; learning (artificial intelligence); pattern classification; binary feedforward neural networks; binary training samples; classification hyperplanes; hard limitation activation function; hidden neurons; neuron; parallel degeneration; parallel pattern; pattern classification; Algorithm design and analysis; Feedforward neural networks; Neural networks; Neurons; Pattern classification; Sufficient conditions; Upper bound;

fLanguage

English

Publisher

ieee

Conference_Titel

Signal Processing Proceedings, 1998. ICSP '98. 1998 Fourth International Conference on

Conference_Location

Beijing

Print_ISBN

0-7803-4325-5

Type

conf

DOI

10.1109/ICOSP.1998.770849

Filename

770849