PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Bagged Structure Learning of Bayesian Networks
Gal Elidan
In: AIStats 2011, 11-13 Apr 2011, Fort Lauderdale, Florida.

Abstract

We present a novel approach for density esti- mation using Bayesian networks when faced with scarce and partially observed data. Our approach relies on Efron's bootstrap frame- work, and replaces the standard model selec- tion score by a bootstrap aggregation objec- tive aimed at sifting out bad decisions dur- ing the learning procedure. Unlike previ- ous bootstrap or MCMC based approaches that are only aimed at recovering specic structural features, we learn a concrete den- sity model that can be used for probabilis- tic generalization. To make use of our ob- jective when some of the data is missing, we propose a bagged structural EM proce- dure that does not incur the heavy com- putational cost typically associated with a bootstrap-based approach. We compare our bagged objective to the Bayesian score and the Bayesian information criterion (BIC), as well as other bootstrap-based model selection objectives, and demonstrate its eectiveness in improving generalization performance for varied real-life datasets.

PDF - Requires Adobe Acrobat Reader or other PDF viewer.
EPrint Type:Conference or Workshop Item (Paper)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Learning/Statistics & Optimisation
Theory & Algorithms
ID Code:9210
Deposited By:Gal Elidan
Deposited On:21 February 2012