A STOCHASTIC A POSTERIORI UPDATING ALGORITHM FOR PATTERN RECOGNITION.
Abstract
The pattern recognition problem is viewed as a sequential two-class classification problem. An algorithm is given which updates the a posteriori distribution of membership in one class at stage n+1 based on the previous n observations and their respective classifications. This updated estimate of the a posteriori distribution is then used to classify the n+1st observation. It is shown that under very general assumptions on the distributions involved the expected squared-error of the estimated a posteriori distribution and true a posteriori distribution conditional on the past observations and classifications approaches zero in probability as the number of past observations increases. From this result it is shown that the probability of misclassification using the estimated rule conditional on the past observations approaches in probability the minimal probability of misclassification using the optimal Bayes rule. Comparisons with other algorithms in the literature are also discussed. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1966
- Accession Number
- AD0642612
Entities
People
- J. Van Ryzin
Organizations
- Stanford University