Probabilistic Graphical Models

Catalog description

Probabilistic graphical models provide a flexible framework for modeling large, complex, heterogeneous collections of random variables. After a brief introduction to their representational power, we provide a comprehensive survey of state-of-the-art methods for statistical learning and inference in graphical models. We discuss a range of efficient algorithms for approximate inference, including optimization-based variational methods, and simulation-based Monte Carlo methods. Several approaches to learning from data are explored, including conditional models for discriminative learning, and Bayesian methods for controlling model complexity. Programming experience required for homeworks and projects, which integrate mathematical derivations with algorithm implementations. PREREQUISITES: CSCI1420 or APMA1690.