I present a class of graphical models for directly representing the joint cumulative distribution function (CDF) of many random variables, called "cumulative distribution networks" (CDNs). I will present properties of such graphical models, such as efficient marginalization, marginal and conditional independence and connections to extreme value theory. In order to perform efficient inference and estimation in such graphical models, I will describe the "derivative-sum-product" (DSP), "gradient-derivative-product" (GDP) and "junction tree differentiation" (JDiff) message-passing algorithms for computing conditional cumulative distributions where messages correspond to derivatives of the joint CDF. Time permitting, I will present results on various application settings in which CDNs have been used.
Jim C. Huang received his bachelor's degree in Electrical Engineering from McGill University in 2004 and his doctorate in Electrical and Computer Engineering from the University of Toronto in 2009. During 2006-2011, Jim has worked at Microsoft Research in the Cambridge, UK and Redmond, USA Machine Learning groups, collaborating on a variety of projects involving statistical machine learning and applications to the natural and applied sciences. He has received post-graduate fellowships from the Natural Sciences and Engineering Research Council of Canada and an outstanding conference paper award at UAI 2008.