Pattern Recognition with Continous Parameter, Observable Markov Chains.
Abstract
The paper develops Bayesian learning and decision-making algorithms for the following pattern recognition problem. Each of M pattern classes is described by a continuous-parameter, discrete-state Markov chain having a finite number of states. All states and times of transition between states can be observed perfectly. The transition rate matrices, which establish the properties of the chains, are not known a priori. A Bayesian learning algorithm using a fixed amount of memory digests the training patterns which consist of a member function from each chain. This leads to an iterative, computationally simple, decision-making algorithm for classifying any portion of a member function. The Bhattacharyya bound and the probability of error are derived for the 2-state, 2-chain problem when the transition rate matrices are known. The last section reports on a computer simulation of a 3-state, 2-chain problem with varying amounts of training data. An appendix summerizes the pertinent facts about Markov chains. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 25, 1970
- Accession Number
- AD0720837
Entities
People
- Erdal Panayirci
- Richard C. Dubes
Organizations
- Michigan State University