CONVERGENCE PROPERTIES OF A LEARNING ALGORITHM

In the learning process described by the algorithm, observations are made on individuals one at a time and the current estimate of the required partitioning may be adjusted after each observation, on the basis of knowledge of the category to which the individual observed belongs. At any given time,...

Full description

Saved in:
Bibliographic Details
Main Authors: Breiman,Leo, Wurtele,Zivia S
Format: Report
Language:English
Subjects:
ABSTRACT ALGEBRA
ALGEBRA
BOOLEAN ALGEBRA
INFORMATION THEORY
LEARNING
PERCEPTRONS
STATISTICAL SAMPLES
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In the learning process described by the algorithm, observations are made on individuals one at a time and the current estimate of the required partitioning may be adjusted after each observation, on the basis of knowledge of the category to which the individual observed belongs. At any given time, the current estimate of the partitioning is all that is held in memory; past history is lost except insofar as it has been incorporated into the present estimate. The learning process of perceptrons, as well as that of other artificial intelligence, is of this general form. It is assumed that each individual is a member of one and only one of two categories. The results obtained are applicable to the more general case, however, for they may be applied to appropriate partitions of a set of three or more categories into two subsets. Each individual in the population is characterized by an attribute vector X in m-dimensional Euclidean space; S1, S2 are the sets of vectors attributed to members of the first and second categories, respectively. (Author)