Spatial Variable Selection
Localized and multiresolution variable-selection methods for spatial regression, spatial point processes, and heterogeneous high-dimensional systems.
Spatial Statistics • Statistical Learning • Spatial Data Science
Debjoy Thakur, Ph.D.
Assistant Professor, Ahmedabad University
I am an Assistant Professor at Ahmedabad University, India. My research lies at the intersection of spatial statistics, high-dimensional statistical learning, spatial extremes, variational inference, and modern machine learning.
A central theme of my work is the development of localized, multiresolution, and scalable statistical learning methods for complex spatial and spatio-temporal data. I am particularly interested in methodology that combines interpretable statistical modeling, rigorous theoretical guarantees, and computational scalability.
Research
Research Themes
My research develops statistical methodology for heterogeneous, high-dimensional, dependent, and geometrically complex data.
Spatial Extremes
Statistical and machine-learning methodology for threshold exceedances, environmental extremes, nonstationary risk surfaces, and extreme spatial events.
Statistical Learning Theory
Theoretical foundations for localized learning, penalized estimation, spatial dependence, nonstationarity, and asymptotic guarantees.
Variational Inference
Scalable variational approximations for latent spatial models, structured random effects, and restricted maximum likelihood estimation.
Spatial Genomics & fMRI
Statistical learning and feature-selection methodology for high-dimensional spatial transcriptomics, biomedical, and neuroimaging data.
Network Data & Spatial AI
Graph-based, network-aware, and deep-learning methods for transportation systems, environmental networks, and complex spatial processes.
Current Work
Current Research Directions
Local & Multiresolution Variable Selection
Developing localized LASSO, SCAD, and multiresolution methods that allow predictor relevance to vary across geographic space.
Spatial Point Processes
High-dimensional intensity modeling and localized feature selection for spatial event data, including transportation safety and crime applications.
Spatial Statistics on Manifolds
Developing intrinsic intensity and pair-correlation methodology for point processes observed on Riemannian manifolds.
Spatial Deep Learning
Combining neural networks, spatial regularization, covariance modeling, and graph structure for nonstationary spatial and spatio-temporal prediction.
Selected Work
Recent Research
Variational Approximated Restricted Maximum Likelihood Estimation for Spatial Data
Debjoy Thakur. Accepted in Statistics & Probability Letters.
Multi-Resolution Analysis of Variable Selection in Spatial Intensity Function for Risky Road in St. Louis
Debjoy Thakur and Soumendra N. Lahiri.
Revised and resubmitted.
Local Variable and Neighborhood Selection in Spatial Firearm Fatality in Southeastern USA
Debjoy Thakur, Lyrian Zhao, and Soutir Bandyopadhyay.
Trustworthy Intrinsic Pair Correlation Function Estimation on Riemannian Manifold
Debjoy Thakur and Jorge Mateu.
Manuscript in preparation.
Teaching
Teaching Experience
I have taught undergraduate and graduate courses in Survival Analysis, Linear Models, Probability, and Stochastic Processes.
My teaching emphasizes the connection between mathematical foundations, statistical methodology, computation, and applications to real data.
Teaching Portfolio →Software
Research Software & Code
I develop research software and reproducible code accompanying methodological work in spatial statistics and statistical machine learning.
Current software projects include
localLASSO
and
gpdpenCNN,
along with reproducibility repositories for research papers.
Academic Background
Appointments & Education
Appointments
2026 – Present
Assistant Professor
Ahmedabad University, India
2023 – 2026
Postdoctoral Lecturer
Washington University in St. Louis, USA
2023
Visiting Scientist
Indian Statistical Institute, Kolkata, India
Education
2019 – 2023
Ph.D. in Statistics
Indian Institute of Technology Tirupati
Advancements of Space-Time Modeling in Environmental Statistics
2017 – 2019
M.Sc. in Statistics
Banaras Hindu University
2014 – 2017
B.Sc. in Statistics
University of Calcutta
Contact
Research & Collaboration
I welcome discussions and collaborations in spatial statistics, statistical learning, spatial extremes, spatial AI, high-dimensional modeling, and related interdisciplinary applications.