Brief Bio (Murali Haran)

Brief Bio (Murali Haran) Graduate students, academic background etc.

Graduate Students

Current

Jaewoo Park (expected 2019) Computational Methods for Intractable Likelihoods, with Applications to Infectious Disease Models
Ben Seiyon Lee (pre-comp) Sequential Monte Carlo and Climate Science
Xiaoxiao Li (pre-comp) Disease Modeling and Point Processes: Neonatal Sepsis and Hydrocephalus in Africa
Claire Kelling (pre-comp) Point Processes and Social Science
John Mattiace (pre-comp) Vaccine Hesitancy

Ph.D. Completed

Virginia Recta (2009; joint with J.L. Rosenberger): A model-based analysis of semi-continuous spatial data. Position after graduation: Mathematical Statistician, Food and Drug Administration, VA.
K. Sham Bhat (2010): Inference for complex computer models and large multivariate spatial data with applications to climate science. Position after graduation: Mathematical Scientist, Los Alamos National Laboratories, Los Alamos, NM.
Matthew Tibbits (2011; joint with J.C. Liechty): Parallel Markov chain Monte Carlo. Position after graduation: Mathematical Statistician, Washington DC
John Hughes (2011; joint with J. Fricks): Motor Proteins and Non-Gaussian Areal Data: Advances in Stochastic Modeling and Computation. Position after graduation: Assistant Professor, Department of Biostatistics, University of Minnesota.
Roman Jandarov (2012): Inference with Implicit Likelihoods for Infectious Disease Models. Position after graduation: Postdoctoral Fellow, Department of Biostatistics, University of Washington--Seattle. Now: Assistant Professor, Department of Environmental Health, University of Cincinnati
Won Chang (2014; joint with K. Keller, Department of Geosciences, Penn State): Climate Model Calibration using High-dimensional and Non-Gaussian Spatial Data. Position after graduation: Postdoctoral Fellow, Department of Statistics, University of Chicago. Now: Assistant Professor, Department of Mathematical Sciences, University of Cincinnati
Josh Goldstein (2015): Compartmental, Spatial and Point Process Models for Infectious Diseases. Position after graduation: Social and Decision Analytics Laboratory, Virginia Tech.
James Russell (2016; joint with Ephraim Hanks): Space-time Models for Animal Movement Data Position after graduation: Assistant Professor, Dept of Mathematics and Computer Science, Muhlenberg College, Allentown, PA.
Yawen Guan (2017): Reduced-dimensional Non-Gaussian Spatial Models and Statistical Methods for Studying the West Antarctic Ice Sheet Position after graduation: Postdoctoral fellow, SAMSI (Statistics and Applied Mathematical Sciences Institute)/North Carolina State University.

M.S.

Muhammad Atiyat (2008). Ph.D., Department of Statistics, Penn State University. Now: United Nations.
Chris Groendyke (2008). Ph.D., Department of Statistics, Penn State University. Now: Robert Morris College.
Matthew Tibbits (2009). Ph.D., Department of Statistics, Penn State University. Now: Mathematical Statistician, Washington DC

Background

B.S. Computer Science (with minors in statistics, mathematics and film), Carnegie Mellon University, 1997.
M.S., Ph.D. Statistics, University of Minnesota, 2003.
Postdoctoral fellow, NISS (courtesy appointment at Duke University) (2003--2004).
New Research Fellow, SAMSI (2009--2010).
Visiting Associate Professor (sabbatical, 2011--2012) Department of Statistics, University of Washington.

I was jointly advised by Luke Tierney* and Brad Carlin** on my dissertation "Efficient Perfect and MCMC Sampling Methods for Bayesian Spatial and Components of Variance Models" (see my math genealogy ). I worked on computation for Bayesian spatial models and some hierarchical models. In particular, I investigated sampling techniques for hierarchical models that use Gaussian Markov Random Fields (GMRFs). A major portion of my thesis contains work on perfect sampling approaches for Bayesian disease mapping models. For more on perfect sampling, see D.B.Wilson's website . While in graduate school, I also studied hierarchical Bayesian modeling techniques used for disease mapping and particulate matter exposure studies. As a postdoctoral fellow I studied statistical approaches for software engineering problems.

Honors :

2016 Fellow of the American Statistical Association
2015 Abdel El-Shaarawi Young Researcher (AEYR) Award given by The International Environmetrics Society to ``recognize and honor outstanding contributions to the field of environmetrics''
2014 Young Investigator Award given by the American Statistical Association (ASA) Section on Statistics and the Environment (ENVR) for contributions to statistical methods for environmental science.
New Researcher Fellow, Statistical and Applied Mathematical Sciences Institute (SAMSI), Fall 2009.
Outstanding Poster Award, ``Case Studies in Bayesian Statistics'' conference. Carnegie Mellon University, Pittsburgh. 2007.
Student Service Award, Statistics, University of Minnesota. 2003.
School of Computer Science Honors, Carnegie Mellon University. May 1997.
Inducted into Lambda Sigma National Honor Society. 1994.

Editorial Work

Co-Editor

Bayesian Analysis (2016--)

Associate Editor

Biometrics (2009-2011)
Bayesian Analysis (2010--2015)
Journal of Agricultural, Biological and Environmental Statistics (2011--)
The American Statistician (2014--)

Guest Editor

Journal of Agricultural, Biological and Environmental Statistics Special Issue (jointly edited with Brian Reich) on Computer Models and Spatial Statistics for Environmental Sciences (2011)
Statistical Methodology Special Issue (jointly edited with Peter Craigmile) on Modern Statistical Methods in Ecology (2014)

Research Interests

Markov chain Monte Carlo (MCMC) algorithms
MCMC algorithms are very widely used for both Bayesian and frequentist statistical inference. I am interested in designing algorithms that have good properties so they produce accurate estimates quickly. I am particularly interested in algorithms that are automated, that is, where the user needs to provide minimal input.
- Exact/perfect sampling: This is a sophisticated class of algorithms that use Markov chains to produce i.i.d. draws from the target distribution. It has generally been difficult to get this to work for non-toy Bayesian models. I have worked on implementing these for a class of Gaussian Markov random field models.
- Stopping rules and standard errors for MCMC: My collaborators and I have shown that using a consistent batch means estimate of standard errors and then using the standard errors as a criteria for stopping an MCMC run works well in theory and practice.
- Fast mixing algorithms: I have worked on designing MCMC algorithms that explore the state space quickly, demonstrating their effectiveness via empirical studies and theoretical results wherever possible.
- Parallel computing: With Matt Tibbits (Ph.D. student) and John Liechty, I have recently been working on constructing MCMC algorithms that exploit the latest in parallel computing resources.
Spatial models
- Spatial models for non-Gaussian data: I am working on modeling and computation for `automodels' and spatial generalized linear mixed models for spatially dependent data that are non-Gaussian, for e.g. spatial binary (0-1) data, count data, and zero-inflated data.
Complex computer models: emulation, calibration, model averaging
- Computer experiments: complex computer models increasingly have output that is in the form of spatial fields. There are computational and modeling challenges involved in parameter inference/calibration based on such computer model output and physical observations of the process being modeled by the computer model. I have been working on flexible and computationally efficient approaches for these problems.
- Often multiple computer models are available for the same phenomenon. Bayesian model averaging (BMA) is an approach that allows for information from multiple models to be combined in a probabilistic fashion. I have worked on BMA approaches for computer models with spatial output.
Interdisciplinary work in environmental and ecological statistics
My methodological work in this area is motivated by the challenge of carrying out statistical inference when the scientific models, questions and data sets are complex. Statistical computing and spatial modeling often play crucial roles in these projects.
- Geosciences (and meteorology) on using large spatiotemporal data sets (e.g. from large oceanic observation systems) and complex climate computer models to learn about indicators of climate change. This work involves an interdisciplinary team including Prof.Klaus Keller in geosciences and myself, and graduate students and postdocs in statistics and geosciences. The statistical aspects of this work fall under the fields of space-time modeling and computer model calibration. I am also Co-Director of a Penn State Center led by Prof.Keller, Climate Risk Management Initiative (CLIMA) . This work has been supported by the National Science Foundation (NSF) and the US Geological Survey (USGS).
- Geography on studying how ecological factors can impact coastal aquatic ecosystems with Prof.Denice Wardrop (Assistant Director of Penn State Institute for Energy and Environment (PSIEE)). This work has been supported by the National Oceanic and Atmospheric Administration (NOAA) and USGS.
- Center for Infectious Disease Dynamics: I am working with a group of researchers from biology, physics, geography and mathematics on studying the prevalence, incidence and geographic spread patterns of past epidemics in developing countries, and will help prevent future infectious disease epidemics by optimizing vaccine strategies for particular diseases and regions ). I am currently working with a Ph.D. student and Prof. Ottar Bjornstad on using gravity models to infer the spatiotemporal dynamics of infectious disease. This work has been supported by a Gates Foundation grant.
- Social Science Research Institute on space-time models for studying the impact of socio-economic factors on public health.
- Plant pathology on estimating risks of crop epidemics.
- Modeling pollutants (PM2.5) with researchers at the School of Public Health, University of Minnesota.
Statistics in Software Engineering
I have worked with computer scientists on statistical issues in software engineering. In particular, we have explored the use of a classification technique called random forests to identify failing executions in large computer programs.

* A member of the R core group and author of Xlisp-stat .
** Author of the books Bayes and Empirical Bayes Methods for Data Analysis and Hierarchical Modeling and Analysis for Spatial Data .

Return to home page