Structure Learning of Probabilistic Graphical Models
Probabilistic Graphical models provide a strong and clear formalism for studying conditional independence relations, probabilistic reasoning, and decision making that arise in different research areas. Originally, graphs with a single type of edge were used i.e., undirected graphs and directed acyclic graphs (DAGs). However, in the case of undirected graphs only symmetric relations i.e., correlation between variables can be represented and in the case of DAGs only asymmetric relations i.e., cause and effect relation between variables can be represented. Chain graphs were introduced as a unification of directed and undirected graphs to model systems containing both symmetric and asymmetric relations.
Graphical models are the backbone of structural causal models and at the center of robust artificial intelligence. The construction of these models is a challenging task that would be greatly facilitated by automation. In this page, I present some of my recent work addressing challenges in learning the structure of probabilistic graphical models from data.
AMP Chain Graphs: Minimal Separators and Structure Learning Algorithms
August 09, 2020
Journal of Artificial Intelligence Research (JAIR 2020)
Learning LWF Chain Graphs: A Markov Blanket Discovery Approach
June 20, 2020
Uncertainty in Artificial Intelligence (UAI 2020)