Mixture models for data exploration and prediction

Padhraic Smyth (UC Irvine) (S)

Finite mixture models often provide a relatively simple yet effective way to model large complex data sets. We begin with a brief tutorial overview on the representational capabilities of mixtures and the use of the Expectation-Maximization (EM) procedure for fitting such models. We will then proceed to discuss a broad variety of mixture model applications involving large data sets, including the generalization of the "classic" multivariate mixture model to temporal and spatial data settings. Applications from atmospheric science, computational biology, epidemiology, computer vision, and astronomy will be discussed, time permitting.

Back to Mathematical Challenges in Scientific Data Mining