AN ANALOG LINEAR CLASSIFICATION NETWORK.
Abstract
A linear network was built to classify analog signals consisting of a large number of parallel inputs. These inputs were derived by a feature abstracting system using synthetic nerve networks. In this classification network the signals pass simultaneously through a maximum amplitude filter and then are classified by a resistive memory matrix. The maximum amplitude filter attenuates smaller inputs much more than larger ones and serves to 'pick out' predominant features. This is a way of utilizing the pandemonium concept of Selfridge. Classification by linear hyperplanes is discussed briefly and then the operation and design of the maximum amplitude filter is covered. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 27, 1966
- Accession Number
- AD0641178
Entities
People
- Robert J. Biegalski
Organizations
- Naval Ordnance Laboratory