Hi, I'm Miheer Dewaskar
I am a postdoc in the Department of Statistical Science at Duke University, working with David Dunson on developing methods for robust inference in parametric models and non-parametric Bayesian clustering. I obtained my PhD in Statistics and Operations Research from UNC Chapel Hill in 2021, under the direction of Shankar Bhamidi, Amarjit Budhiraja, and Andrew Nobel in which we developed a method for groupwise correlation mining and applied it to the data-integration problem of eQTL analysis in genomics, and asymptotically studied the power-of-choice load-balancing scheme.
Research in Data and Systems science
My current research develops general principled frameworks for robust statistical inference and exploratory data analysis that have applications in a variety of fields like genomics, neuroscience, and social science. I am developing theoretically motivated Bayesian and frequentist frameworks to robustly fit likelihood-based models that may be misspecified. This methodology is relevant in the current age where we tend to have big data that is potentially biased or corrupted (e.g. arising in surveys, health records, etc.). I have also developed new exploratory data analysis methods, e.g. those based on novel Bayesian density-based clustering and iterative-testing to find combinatorial structures (e.g. networks) for integration of multi-view data. Given my strong background in stochastic processes and large deviation theory, in the future I aim to develop methods for analysis of time series data and for data assimilation problems that arise in climate modeling.Publications
* indicates joint first authors.= indicates primary contribution and alphabetical author order.
Peer-reviewed
- Finding Groups of Cross-Correlated Features in Multi-View data. Dewaskar M , Palowitch J, He M, Love M.I., Nobel A.B. Accepted into the Journal of Machine Learning Research, VOL. 24.
- Near Equilibrium Fluctuations for Supermarket Models with Growing Choices (2022). Bhamidi S, Budhiraja A, and Dewaskar M=. Annals of Applied Probability, VOL. 32 (NO. 3), 2083-2138.
- NExG: Provable and Guided State Space Exploration of Neural Network Control Systems using Sensitivity Approximation (2022) Goyal M, Dewaskar M, Duggirala PS. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2022, doi: 10.1109/TCAD.2022.3197524.
- Controlling a population (2019). Bertrand N, Dewaskar M=, Genest B=, Gimbert H, and Godbole A. Logical Methods in Computer Science, Vol. 15, Issue 3.
- Controlling a population (2017). Bertrand N, Dewaskar M=, Genest B=, Gimbert H. 28th International Conference on Concurrency Theory (CONCUR 2017).
- Robustifying likelihoods by optimistically re-weighting data. Dewaskar M *, Tosh C*, Knoblauch J, Dunson D.B. Under revision at the Journal of American Statistical Association.
- Bayesian Level-set Clustering. Buch D*, Dewaskar M*, Dunson D.B. Submitted to the Journal of American Statistical Association.
Courses taught:
- STA 211: Mathematics of Regression (Fall 2023, Spring 2024) at Duke University.
- Introduction to Statistical Modeling, (Fall 2019) at UNC Chapel Hill.
Software Packages:
- Correlation Bi-community Extraction (CBCE): R package for multi-view data analysis.
- Optimistically Weighted Likelihood (OWL): Python package to robustly fit likelihood-based models.
Awards and Honors:
- Cambanis-Hoeffding-Nicholson award (2017) by Department of Statistics and Operations Research, UNC Chapel Hill.
- Medal of Excellence (2016) by Chennai Mathematical Institute.
- Charpak Scholarship (2015) by French Embassy in India.
- INSPIRE Scholarship (2011-14) by DST, India.