SUPERIMPOSED RANDOM CODING OF STIMULUSRESPONSE CONNECTIONS.
Abstract
The problem of economically linking a large number of stimuli with a large number of potential responses is considered to resemble a problem of efficient retrieval of documents (the responses) on the basis of their characterization by descriptors (the stimuli to which the responses are appropriate). In this retrieval problem, a method whereby the codes for descriptors are random positions in a coding field, and whereby codes for all applicable descriptors are superimposed in the same field, seems to be the simplest way of avoiding serious difficulties of retrieval. After a review of this method, the possibility is considered that very simple neural mechanisms could embody the essential features of the method. The aim of the discussion is to learn whether very simple structures and patterns of reinforcement would be adequate to carry out useful information processing in the brain, and to show some conceivable functions of simple neural networks that the experimenter might keep in mind. The discussion also shows how the structure of a simple 'perceptron-like' network is suggested by the requirements of a retrieval task. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 10, 1965
- Accession Number
- AD0625759
Entities
People
- Peter H. Greene
Organizations
- System Development Corporation