PhD Workshop:

Model-Based Clustering via Finite Mixture Models: Foundations and Applications with R

Volodymyr Melnykov | Professor of Statistics, Rogers-Spivey Faculty Fellow
Department of Information Systems, Statistics, and Management Science
The University of Alabama

Finite mixture modeling is an increasingly popular technique that is primarily known for its flexibility in modeling heterogeneity in data.  One of its main applications, known as model-based clustering, is a method of unsupervised learning that provides an intuitive and  highly interpretable way to define clusters and partition the data. Under this formulation, a population is viewed as a collection of a finite number of subpopulations, each represented by a subsample commonly referred to as a data group or cluster. 
Applications of finite mixture modeling can be found in physics, sociology, linguistics, and many other disciplines.

In a series of lectures, we consider the foundations of finite mixture models, such as the expectation-maximization algorithm, 
elliptical and non-elliptical mixture models, mixture order estimation, assessment of clustering complexity and uncertainty, 
as well as the choice of an initialization strategy. Some recent developments in mixture modeling, such as matrix-variate mixtures, 
cluster-weighted models, semi-supervised cluster analysis, and clustering categorical sequences, will also be presented. 
These techniques will be thoroughly motivated and illustrated using modern real-life applications and relevant R packages.
 

VENUE: Department of Economics, Doctoral Common Room, second floor

Calendar: 17 Nov. 9:30-13:30; 18 Nov. 9:30-12:30; 25 Nov. 9:30-12:30;