##Statistics Seminar##\\ Department of Mathematical Sciences
^ **DATE:**|Thursday, Month 31, 2017 |
^ **TIME:**|1:15pm -- 2:15pm |
^ **LOCATION:**|WH 100E |
^ **SPEAKER:**|Yuan Fang, Binghamton University |
^ **TITLE:**|Bayesian Approach to Parameter Estimation |
\\
**Abstract**
Bayesian Approach to Parameter Estimation and Clustering for
the Mixtures of Multivariate Normal Inverse Gaussian Distributions
Abstract: Increasingly, non-Gaussian mixture models are gaining
attention for mixture model-based clustering particularly when dealing
with data that exhibit features such as skewness and heavy tails. One
such mixture distribution is the mixtures of multivariate normal
inverse Gaussian (MNIG) distribution. MNIG arises from a mean-variance
mixture of a multivariate Gaussian distribution with the inverse
Gaussian distribution. A mixture of MNIG distributions has the
flexibility to represent both skewed and symmetric clusters as well as
their mixture, which makes them suitable for a wide range of datasets.
In this talk, I will focus on an approach for parameter estimation of
mixtures of MNIG distributions in a Bayesian framework via a Gibbs
scheme. Novel approaches to simulate univariate generalized inverse
Gaussian (GIG) random variables and matrix generalized inverse
Gaussian (MGIG) random matrices will be provided. The proposed
algorithm will be applied to both simulated and real data. Some future
work on extending finite mixture of MNIG distributions to an infinite
mixture model framework will also be discussed.