THE LEARNING BEHAVIOR AND ERGODIC PROPERTY OF FINITE-STATE MARKOV CHAINS.
Abstract
The dynamic behavior of ergodic finite-state homogeneous Markov chains is studied. An ergodicity test of general natural is formulated. From this general test procedure, various test criteria are derived. One criterion is shown to be both necessary and sufficient. The ergodicity is found to be a topological property of the state flow graph of a finite-state homogeneous Markov chain or an equivalent stochastic automation. The dynamic behavior of an ergodic chain can be explained as a contraction mapping in the probability vector space. The norm of the induced transition probability matrix serves as pessimistic estimation of the learning rate.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1967
- Accession Number
- AD0653116
Entities
People
- H. H. Yeh
- J. T. Tou
Organizations
- Ohio State University