Optimizing multilayer perceptrons by discriminatory component analysis

Yue Wang, Zuyi Wang, Jianhua Xuan, Junying Zhang, Eric P. Hoffman, Robert Clarke, Javed Khan

Research output: Chapter in Book/Report/Conference proceedingConference contribution

1 Scopus citations

Abstract

Multilayer perceptions offer an integrated procedure for feature extraction and Bayes classification by learning the decision boundary. Its feed-forward autoassociative architecture can also be used to construct subspaces in a supervised or unsupervised model. On the other hand, multiclass linear discriminant analysis provides a multivariate prediction by estimating the density function. Its linear subspaces obtained by the weighted Fisher criteria under a standard finite normal mixture model retain most closely the intrinsic Bayes separability. Here we show a twofold connection between multilayer perceptrons and linear discriminant analysis using discriminatory component analysis. This theoretical observation immediately suggests a possible clustering-model supported optimization mechanism for multilayer perceptrons: the weights between the input and hidden layers are related to eigenvectors of the weighted Fisher scatter matrix, the number of the hidden layer neurons is justified by the corresponding significant eigenvalues, and the weights connected to the output neurons are obtained from the centers of the classes in the extracted feature subspaces.

Original languageEnglish (US)
Title of host publicationMachine Learning for Signal Processing XIV - Proceedings of 2004 IEEE Signal Processing Society Workshop
EditorsA. Barros, J. Principe, J. Larsen, T. Adali, S. Douglas
Pages273-282
Number of pages10
StatePublished - 2004
Externally publishedYes
EventMachine Learning for Signal Processing XIV - Proceedings of the 2004 IEEE Signal Processing Society Workshop - Sao Luis, Brazil
Duration: Sep 29 2004Oct 1 2004

Publication series

NameMachine Learning for Signal Processing XIV - Proceedings of the 2004 IEEE Signal Processing Society Workshop

Conference

ConferenceMachine Learning for Signal Processing XIV - Proceedings of the 2004 IEEE Signal Processing Society Workshop
CountryBrazil
CitySao Luis
Period9/29/0410/1/04

Fingerprint Dive into the research topics of 'Optimizing multilayer perceptrons by discriminatory component analysis'. Together they form a unique fingerprint.

Cite this