2022 UPDATE: My personal website is now hosted at https://reetamm.com. This site is currently here for archival purposes only.
I’m a PhD candidate in Statistics at the University of Maryland, Baltimore County. My dissertation involves extensions to hidden Markov models, including variational Bayes, copula specifications for multivariate emissions distributions without a natural joint specification, and dealing with dimensionality using a multivariate Markov chain state process. Most of the problems are in the context of spatiotemporal precipitation models, but are applicable to HMMs with high dimensional emission distributions.
I am also a graduate research assistant with UMBC’s High Peformance Computing Facility. My RA work is split between research facilitation at the HPCF and a JCET project to model forest cover change over Panama.
My research interests include the spatiotemporal analysis of geostatistical data, Bayesian inference, stochastic simulations, Machine Learning, Deep Learning and High Performance Computing. In addition, I am also interested in state space models and in modeling extreme weather events. I am actively looking to be involved in the Earth Sciences domain.
My CV, updated Nov 2021. You can find me over at LinkedIn, or reach me via email.
Research Projects
- Variational Bayes Estimation of a Hidden Markov Model for Daily Precipitation at a Single Location (VB-HMM)
Stochastic precipication generators which use HMMs for single site and multi-site rainfall are fairly common; however, they all use the EM-like Baum-Welch algorithm which is a maximum likelihood method and doesn’t deal with model complexity well. I have been looking at adapting variational optimization to parameter estimation in HMMs for precipitation. Precipitation in this model is specified as a semi-continuous distribution with a point mass at zero and a mixture of two exponential distributions for postitive precipitation. Instead of the Baum-Welch algorithm, we use structured variational inference for parameter estimated. This is a first pass for just univariate emissions. Currently extending it to multi-site precipitation processes. -
Hidden Markov models with Gaussian copulas (HMM-GC)
This is an extension of the HMM for precipitation proposed by Hughes and Guttorp (1994) and Robertson et al (2006). It is based around the MVNHMM application designed by Sergey Kirshner which can run the Baum-Welch and Viterbi algorithm for HMMs, as well as simulate precipitation data. I’ve extended this with Gaussian copulas to model the emission distribution correlations, since it is not possible to specify a joint distribution for multiple locations, and the spatial correlation induced by the state process is insufficient. Code is written in R and this version is for the GPM-IMERG dataset over the Chesapeake Bay watershed. -
Fast Kalman filtering and immplementation using R
Final project for my Time Series Analysis class. Mostly the review of a paper which discusses FKF and smoothing via a low-rank perturbative approach. Also implementation of the FKF algorithm using R with a simulation study. Code needs a bit of cleanup. - Variance component estimation for the one-way random effects model
Final project for Mathematical Statistics II course, done jointly with my classmates Yewon Kim and Mark Ramos. The project deals with various ways to estimate the variance component of a one-way model, which is generally a tricky process and does not have a nice, clean solution. I worked on the bootstrap algorithms, as well as encoding the methods discussed in Hartung and Knapp.
Publications
- Reetam Majumder, Nagaraj K. Neerchal, and Amita Mehta. Variational Bayes Estimation of Hidden Markov Models for Daily Precipitation with Semi-Continuous Emissions. (In preparation).
- Reetam Majumder, Erika Podest, and Amita Mehta. Monitoring forest cover in Panama based on Landsat-8 and Sentinel-2 imagery. (In preparation).
- (Peer Reviewed Journal) Jeffrey X. Xie, Xiaoming Fan, Christopher A. Drummond, Reetam Majumder, Yanmei Xie, Tian Chen, Lijun Liu, Steven T. Haller, Pamela S. Brewster, Lance D. Dworkin, Christopher J. Cooper, and Jiang Tian. MicroRNA profiling in kidney disease: Plasma versus plasma-derived exosomes. Gene, 627:1-8, 2017. [www]
- (Refereed Proceedings) Reetam Majumder, Amita Mehta, and Nagaraj K. Neerchal. Copula-based correlation structure for multivariate emission distributions in hidden Markov models. In JSM Proceedings, Section for Statistics and the Environment. Alexandria, VA: American Statistical Association, 2020.
- (Refereed Proceedings) Gerson Kroiz, Reetam Majumder, Nagaraj K. Neerchal, Matthias K. Gobbert, Amita Mehta, and Kel Markert. A hidden Markov model with correlated emissions for daily precipitation generation for the Potomac river basin. Proceedings in Applied Mathematics and Mechanics (PAMM), in press (2020).
- (Technical Report) Gerson Kroiz, Jonathan Basalyga, Uchendu Uchendu, Reetam Majumder, Matthias K. Gobbert, Nagaraj K. Neerchal, Amita Mehta, and Kel Markert. Stochastic precipitation generation for the Potomac river basin using hidden Markov models. Technical Report HPCF-2020-11, UMBC High Performance Computing Facility, University of Maryland, Baltimore County, 2020. [pdf]
- (Technical Report) Reetam Majumder, Redwan Walid, Jianyu Zheng, Carlos Barajas, Pei Guo, Chamara Rajapakshe, Aryya Gangopadhyay, Matthias K. Gobbert, JianwuWang, Zhibo Zhang, Kel Markert, Amita Mehta, and Nagaraj K. Neerchal. Assessing water budget sensitivity to precipitation forcing errors in Potomac river basin using the VIC hydrologic model. Technical Report HPCF-2019-11, UMBC High Performance Computing Facility, University of Maryland, Baltimore County, 2019. [pdf]