IJCNN 2015 Tutorial on Robust EM Methods (T05)
- Where: Killarney Convention Center in Killarney, Ireland
- When: 8:00am–9:45am, Sunday, July 12, 2015
Yixin Chen, Ph.D.
Department of Computer and Information Science,
University of Mississippi
Yixin Chen is an associate professor in the Department of Computer and Informatics Science at the
University of Mississippi. His research interests include machine learning, data mining, computer
vision, bioinformatics, and robotics and control. Dr. Chen serves as an associate editor of Pattern Recognition and is a member of the ACM, the IEEE, the IEEE Computer Society, the IEEE Neural
Networks Society,and the IEEE Robotics and Automation Society. Dr. Chen has published more than 100 publications including Journal of Machine
Learning Research, IEEE Transactions on Pattern Analysis and Machine Intelligence, IEEE
Transactions on Image Processing, IEEE Transactions on Automatic Control, IEEE Transactions
on Robotics and Automation, IEEE Transactions on Robotics, IEEE Transactions
on Fuzzy Systems, IEEE Transactions on Control Systems Technology, BMC Bioinformatics, NIPS, ACM SIGMM, CVPR, ICDM, ICIP, ICRA.
Xin Dang, Ph.D.
Department of Mathematics,
University of Mississippi
Xin Dang is an associate professor in the Department of Mathematics at the
University of Mississippi. Her research interests include robust and nonparametric statistics, statistical and numerical computing, bioinformatics
and multivariate data analysis. In particular, she has focused on statistical learning, data depth and application, and robust procedure computation. Dr. Dang has published more than 20 journal papers including IEEE Transactions on Pattern Analysis and Machine Intelligence, Pattern Recognition, IEEE Transactions on Knowledge and Data Engineering, BMC Bioinformatics, Knowledge Based Systems, Journal of Nonparametric Statistics, Journal of Multivariate Analysis, Statistics in Medicine, Journal of Statistical Planning and Inference.
Finite mixture models are powerful and flexible to represent arbitrarily complex probabilistic
distribution of data. Mixture model-based approaches have been increasingly popular.
Applications in a wide range of fields have emerged in the past decades.
They are used for
density estimation in unsupervised clustering, for estimating class-conditional densities
in supervised learning settings, and for outlier detection purposes. Usually parameters
of a mixture model are estimated by the maximum likelihood estimate (MLE) via the
expectation maximization (EM) algorithm. It is well known that the MLE and hence EM
can be very sensitive to outliers. To overcome this limitation, various robust alternatives
have been developed.The goal of this tutorial is
to make audiences understand the sensitivity problem of EM;
to review various robust EM methods;
to elaborate Spatial-EM and its algorithm;
to illustrate application of robust EM on supervised and unsupervised learning;
- to introduce concepts of robust statistics to machine learning community.
- Topics to be covered:
Review of the EM algorithm
Finite mixture elliptical models
EM algorithm based on Gaussian mixture model
Robust EM methods
Maximizing other types of likelihood
Replace location and scatter estimators with robust ones in M-step
Robustness measure: breakdown point
Algorithm, convergence and R package
Connection with K-median and K-medoid, connection with Kotz EM
Robust Model-based Novelty Detection
Outlyingness, two type errors and AUC criterion
Estimating the number of components: one standard error rule
New species discovery in taxonomic Research
Robust Model-based Clustering
Selecting the number of clusters: BIC, MML, NEC
Probabilistic (soft) clustering vs outright (hard) clustering
- Presentation slides: [PDF]
Disclaimer: The opinions expressed in this web page and
the presentation slides are that of the organizers, not of the IJCNN conference or IEEE, or any other entity.
Updated: Monday March 2 2:27:42 CDT 2015