BAYESIAN LEARNING OF LOGICAL HIDDEN MARKOV MODELS

T. Raiko1,2 - K. Kersting1 - J. Karhunen2 - L. De Raedt1
minipage[b]7cm center 1Institute for Computer Science
Machine Learning Lab
Albert-Ludwigs University of Freiburg
Georges-Koehler-Allee, Building 079
79112 Freiburg, Germany minipage[b]7cm center 2Helsinki University of Technology
Laboratory of Computer and
Information Science,
P.O. Box 5400,
02015 HUT, Finland

Abstract:

Logical hidden Markov models (LOHMMs) are a generalisation of hidden Markov models to analyze sequences of logical atoms. Transitions are factorized into two steps, selecting an atom and instantiating the variables. Unification is used to share information among states, and between states and observations. In this paper, we show how LOHMMs can be learned using Bayesian methods. Some estimators are compared and parameter estimation is tested with synthetic data.