Bayesian Statistics Seminar
North Carolina State University
presents
David Dunson
Duke University
"Nonparametric Bayesian Learning from Big Data"
ABSTRACT
Our ongoing interest is on developing methods for learning low-dimensional structure in high-dimensional data using flexible Bayesian probability models, which accommodate uncertainty in learning of the subspace or lower-dimensional representation. Motivated by applications in genetic epidemiology and machine learning-aided restorations of famous paintings, we focus is this talk on flexible methods for classification based on conditional tensor factorizations and multiscale representations of massive scale data. Basic details and some theoretical justification is provided on the probability modeling frameworks, and we apply the methods to several interesting real data applications, while highlighting ongoing research on scaling up computation to huge data settings.
Thursday, October, 25, 2012
4:00 - 5:00 pm
1216 SAS Hall