##Data Science Seminar##\\ Hosted by Department of Mathematical Sciences
* Date: Tuesday, March 26, 2020
* Time: 12:00pm -- 1:00pm
* Room: WH-100E
* Speaker: Wangshu Tu (Binghamton University)
* Title: A family of mixture models for biclustering
**//Abstract//**
Biclustering allows for simultaneous clustering of the observations and
variables. Martella et. al (2008) introduced biclustering in a model-based
clustering framework by utilizing a structure similar to a mixture of
factor analyzer structures such that observed variables are modelled using
a latent variable that is assumed to be from a MVN(0, I). In Martella et.
al (2008), clustering of variables was introduced by imposing constraints
on the entries of the factor loading matrix to be 0 and 1. However, this
approach restricts the non-zero off-diagonal entries of the covariance
matrix to be 1, which is very restrictive. Here, we assume the latent
variable to be from a MVN(0,T) where T is a diagonal matrix and hence, the
non-zero off-diagonal entires of the covariance matrix are not restricted
to be equal to 1. A family of models are developed by imposing constraints
on the components of the covariance matrix. An alternating expectation
conditional maximization(AECM) algorithm is used for parameter estimation.
Proposed method will be illustrated using simulated and real datasets. The
presentation will conclude with some on-going work and future research
directions.