Probabilistic graphical models are graphical representations of joint probability distributions of the variables in the context. They represent graphically, conditional (in)dependencies inherited in the joint probability distribution. Therefore, graph theoretic properties can be exploited for doing efficient probabilistic inferences in the context. These elegant models have many scientific and engineering applications such as probabilistic expert systems, models of uncertainty reasoning, enhanced classifiers, in self-adaptive and self-organizing systems, etc.
This tutorial covers theory, methods and algorithms associated with probabilistic graphical models. Mainly directed acyclic graphical models (Bayesian networks) and undirected graphical models (Markov networks) are discussed; their representations (conditional independence relationships), inference methods with related algorithms such graph moralization, junction tree algorithm, etc. It also includes model structure learning methods and parameter estimation methods. Firstly, some network structure learning algorithms are discussed. Then different parameter estimation methods are discussed. Finally, some applications of graphical models relevant to SASO and ICAC communities are presented. Mainly, free software such as R is used for modelling and inference tasks.
Priyantha Wijayatunga (Umeå University, Sweden)
Since probabilistic inferences are fundamentally important for SASO community this tutorial is suitable for them, especially for PhD students, researchers and practitioners. Also, it may relate to sections of ICAC community. If someone is interested in modelling probabilistic (in)dependence structure of problem domain and communicate it effectively with applied user then this tutorial discusses ways to do it. Contents are introductory and intermediate levels.
Lecture slides and example sheets will be made available as pdfs (in soft-form).
Basic knowledge in probability theory and statistics. Basic level knowledge in using statistical software R.