An empirical study of the ability of back propagation to approximate posterior probabilities

Michael L. Kalish, Catherine L. Harris

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

1 Scopus citations

Abstract

Proofs which show backpropagation to produce outputs equal to the posteriors of the training data have left open the effect of reduced resources on the accuracy of estimation. The authors empirically explore the effects of reduced resources on the ability of networks to estimate the posterior likelihoods of data in two simple classification problems, one with independent and one with dependent cues. They contrast the effects of restricting hidden units and training cycles for classifying the different cues. Marginal probabilities tend to be incorrectly estimated, and dependencies among the cues affect both the course and outcome of training. For the dependent cue case it was found that even a slight difference between the posteriors for the odd and the even patterns can impair estimation of the posteriors.

Original languageEnglish (US)
Title of host publication91 IEEE Int Jt Conf Neural Networks IJCNN 91
PublisherIEEE Computer Society
Pages2247-2252
Number of pages6
ISBN (Print)0780302273
StatePublished - Dec 1 1991
Externally publishedYes
Event1991 IEEE International Joint Conference on Neural Networks - IJCNN '91 - Singapore, Singapore
Duration: Nov 18 1991Nov 21 1991

Publication series

Name91 IEEE Int Jt Conf Neural Networks IJCNN 91

Other

Other1991 IEEE International Joint Conference on Neural Networks - IJCNN '91
CitySingapore, Singapore
Period11/18/9111/21/91

ASJC Scopus subject areas

  • Engineering(all)

Fingerprint Dive into the research topics of 'An empirical study of the ability of back propagation to approximate posterior probabilities'. Together they form a unique fingerprint.

Cite this