On Networks, Optimised Feature Extraction and the Bayes Decision
Abstract
In this paper we address the problem of multi-class pattern classification using adaptive layered networks. We view such networks as performing generalized linear discriminant analysis in which a particular parametric form is assumed for the nonlinear functions. Training the network consists of a least-square approach which combines a generalized inverse computation to solve for the final layer weights, together with a nonlinear optimization scheme to solve for parameters of the nonlinearities. Such an approach performs feature extraction and classification simultaneously, in which the feature extraction is (optimally) matched to the classification scheme. We derive a general analytic form for the feature extraction criterion and interpret it for specific forms of target coding and error weighting. A particular aspect of the approach is to exhibit how a priori information regarding nonuniform class membership, uneven distribution between train and test sets and misclassification costs may be exploited in a regularized manner in the training phase of networks. Keywords: Great Britain.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 05, 1989
- Accession Number
- ADA219796
Entities
People
- Andrew R. Webb
- David Lowe
Organizations
- Royal Signals and Radar Establishment