| |
Abstract:
This paper introduces the Mixture of Trees, a probability
model that can account for sparse, but dynamically changing
dependence relationships between the variables of the domain under
study. We present a family of efficient algorithms that use EM and
the Maximum Spanning Tree algorithm to find the Maximum Likelihood
and the MAP Mixture of Trees for a variety of priors, including the
Dirichlet and the MDL priors.
|
