Wednesday, August 08, 20123:00 PM - 4:00 PMCNLS Conference Room (TA-3, Bldg 1690)|
Maxentropic Markov Chains
Andrew M. FraserLos Alamos National Laboratory
I describe an algorithm that calculates, for any strongly connected directed graph, the set of branching probabilities for the edges that yields a Markov process with the maximum possible entropy rate. I developed the algorithm as a tool for quantifying uncertainty about an equation of state. I will introduce the entropy maximization problem in terms of Information Theory as I first encountered it in the homework problems for chapter 4 of Cover and Thomas’ text (problems 4.7 and 4.16 in the second edition). Although I will present the set of polynomial equations from Graph Theory that the solution must satisfy, the operationally useful algorithm relies on the power method for calculating eigenvalues.
Host: Garrett Kenyon, email@example.com, 7-1900, IS & T