DECISION MAKING NETWORKS IN PATTERN RECOGNITION,
Abstract
A feed forward switching network is described and then used as a basis for a pattern classifier. Rules associated with the network direct an arbitrary signal from a starting node to one of a group of terminal nodes, each of which is identified with a pattern class. The empirical distribution functions of a training set of pattern samples determine the manner in which the rules are constructed. The rules are functions of the input pattern binary variables. It is shown that, for a class of networks having the ordering property and statistically ordered rules at each node, increasing the size of the network decreases the probability of error. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1969
- Accession Number
- AD0695413
Entities
People
- Leonard J. Grantner
Organizations
- Columbia University