Title: Structure learning of continuous time Bayesian networks via penalized likelihood methods
Authors: Maryia Shpak - Uniwersytet Marii Curie-Sklodowskiej w Lublinie (Poland) [presenting]
Blazej Miasojedow - University of Warsaw (Poland)
Abstract: A Continuous Time Bayesian Network (CTBN) is a time homogeneous Markov process, which is decomposed into processes whose transition intensities depend on the other processes in the network. The dependence structure between the intensities is encoded by a directed graph. CTBNs are widely used for modeling different phenomena from biology, chemistry, social sciences among many others. The problem of learning the structure of the CTBN is a challenging task. We present the solution to this problem based on the penalized maximum likelihood method. Our method can be applied to both fully and partially observed data. In the case of partially observed data we propose the efficient MCMC algorithm to solve the underlying optimization problem. Both theoretical and numerical results will be presented.