Jyotishka Datta
Virginia Polytechnic Institute and State University, VA·Mathematics & Statistics·Early-career
Many kinds of scientific data contain rare but very large values alongside many quantities near zero: star distances in astronomy, species counts in ecology, signals in finance and networks. Heavy-tailed probability distributions describe such data well, and statisticians use them to build models that tolerate extreme observations, capture non-linear patterns, and pick out the few important variables among many.
- Fields
Mathematics & Statistics
Computer Science
Astronomy & Astrophysics
- Topics
Bayesian inference
heavy-tailed priors
hierarchical models
- Grants
- 1 known grant, through Aug 2030
What they work on
- Develop Bayesian independent component analysis with heavy-tailed priors and novel latent space representations, unifying feature extraction methods and controlling higher-order moments when separating hidden sources from mixtures.
- Improve astronomical distance estimation from parallax measurements using hierarchical Bayesian models that incorporate measurement error and non-linear relationships.
- Construct heavy-tailed prior distributions for high-dimensional compositional (simplex-valued) data that adapt to arbitrary sparsity and dependence, applied to ecological models of species diversity and affinity.
- Validate the methods in astronomy, blind source separation, community detection and ecology, with theoretical guarantees and openly accessible R package implementations.
More about the research
This project asks how to build Bayesian methods around heavy-tailed priors, the starting assumptions a model makes before seeing data, that are fast, rigorously justified and usable by working scientists.
Why it matters
Astronomers, ecologists, machine learning practitioners and other scientists could gain reliable, openly available Bayesian tools for messy, high-dimensional data containing extreme values, including more accurate stellar distances from parallax and better models of species composition. Releasing the methods as R packages aims to narrow the gap between new statistical theory and the tools scientists actually use.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Lei Wu
Stevens Institute of Technology, NJ·Engineering
Design likelihood functions and prior distributions for Bayesian models that capture realistic measurement errors and physical characteristics of smart electricity distribution systems.
Alexander Aue
University of California-Davis, CA·Mathematics & Statistics
Derive new probability results describing the structure of nonlinear, possibly heavy-tailed functional time series, where each observation is an entire curve.
Vahid Tarokh
Duke University, NC·Engineering
Develop the theory of score-based quickest change detection for independent, identically distributed single-stream data under Bayesian, generalized Bayesian and minimax formulations, and establish that performance depends on the Fisher distance between distributions.
The school
Virginia Polytechnic Institute and State University: 35 labs on PI Pages, $657M a year on research, 496 research doctorates a year.
More about Virginia Polytechnic Institute and State University
Reach out
A start for your email. Fill in the brackets.