Cumulative Distribution Networks: Inference, Sampling and Learning
Stefan Webb
ARTIFICIAL INTELLIGENCE SEMINARDATE: 2013-05-17
TIME: 14:00:00 - 15:00:00
LOCATION: RSISE Seminar Room, ground floor, building 115, cnr. North and Daley Roads, ANU
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
The cumulative distribution network (CDN) (Huang & Frey, 2011) is a recently developed class of probabilistic graphical models (PGMs) permitting a copula factorization, in which the CDF, rather than the density, is factored. Despite there being much recent interest within the machine learning community about copula representations, there has been scarce research into the CDN, its amalgamation with copula theory, and no evaluation of its performance. We explain how inference and learning are performed, present novel sampling and learning algorithms, and evaluate the model on a real-world data set.
Graphical models (Koller & Friedman, 2009) are used as a general framework for compactly representing high-dimensional probability distributions, for which there exist efficient algorithms for learning and inference. They have found diverse applications in, to name a few, information extraction, medical diagnosis, speech recognition, and computational biology, and have demonstrated superior performance over earlier techniques.
One advantage of the CDN is that it allows the factors to be parameterized as copulae (Silva, Blundell, & Teh, 2011), combining the benefits of graphical models with those of copula theory. In brief, the use of a copula parameterization enables greater modelling flexibility by separating representation of the marginals from the dependence structure, permitting more efficient and robust learning. For example, the marginals can be learnt nonparametrically by kernel density estimation, while the dependence structure learnt as a parametric copula.
BIO:
Stefan is a student from statistics who has been doing an honours project with Steve Gould.
