THE STATISTICAL AND COMPUTATIONAL ANALYSIS OF A SIMPLE LEARNING PROGRAM.
Abstract
A scheme for making an adaptive predictor was based upon a variable linear combination of a large number of non-linear functions of the available predicting variables generated by the use of a pseudo-random numbers. This allowed a regeneration of the predictors computationally each time they were needed rather than relying on massive storage. Thus a small fast computer could be used in lieu of one with a large memory. The elementary analysis of the statistical properties of the 'perceptron-like' program (suggested by the random nature of the predictor generator) was undertaken and tested numerically with results which partially confirmed the analysis. Further statistical investigations of the process were then undertaken. The results of these indicate the efficiency of the predictor is quite good in some cases and relatively poor in others. This lead the way to later analysis to explore schemes which improve upon 'perceptron-like' predictors in those cases where the predictor efficiency is poor. These results will be reported on elsewhere. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1967
- Accession Number
- AD0666932
Entities
People
- Max A. Woodbury
Organizations
- Duke University