Rahul Ghosal and Enakshi Saha
Atmospheric Environment (2021)
Enakshi Saha, Veronika Rockova and Kenichiro McAllin
Journal of Applied Econometrics (in Revision), (2021+)
Veronika Rockova and Enakshi Saha
The 22nd International Conference on Artificial Intelligence and Statistics. PMLR (2019)
Enakshi Saha, Soham Sarkar and Anil K. Ghosh
Journal of Multivariate Analysis 161 (2017): 83-95.
A Theoretical Perspective on Generalized BART Models
Adaptibility of BART: on Theoretical Optimality for Classification, Causal Inference and Survival Analysis
Bayesian Latent-Variable Modeling of the COVID-19 Infection Rate
Enakshi Saha, Carlo Graziani and Marieme Ngom
Abstract: COVID-19 infection rates are severely underreported because of insufficient testing and huge proportion of asymptomatic infections. The goal of this project is to develop a software pipeline that will estimate true infection rates of COVID-19 from viral RNA testing data, by removing the underreporting bias using a Bayesian latent variable model. Our methodology performs epidemic surveillance tasks similar to those for which serological antibody tests are designed, using qualitatively different data, therefore providing a valuable cross-validation addressing the same key questions addressed by serology: (1) How close are we to herd immunity? and (2) What vaccination rate is sufficient to achieve herd immunity? Given the uncertainties and limitations attending serological testing, such a cross-validation may be essential. Additionally, our methodology can forecast - with quantified uncertainties - local epidemic states a few weeks into the future, providing an essential tool for risk management and scarce-resource deployment.
Dynamic Recommender Systems
Abstract: The goal of a recommender is to suggest its users items that they might like. Traditional recommender systems look at existing user ratings for a set of given items and try to predict how a user will rate a new item, if recommended. However not all users respond in similar way to recommendations. As Wang et al. (2020) have pointed out, some users might like an item so much that they will buy that regardless of whether it is recommended. On the flip side, the user might never buy an item even if it is recommended, for example an user who likes comedy movies might always skip movies by a particular director, even if it’s a comedy. So instead of looking at a single rating matrix, it might be useful to see how the rating matrix changes over several months, taking into account whether a particular user was recommended an item or not. In this paper we use this dynamic information on how different users react to recommendations, to build a hierarchical latent factor model that takes dynamic time varying rating matrices along with binary indicators for recommendation intervention as input data, and uses that to come up with recommendations.
with Elizabeth Huppert & Jean Decety, Department of Psychology, The University of Chicago
with Erin Boyle Anderson & Robert K. Ho, Department of Organismal Biology and Anatomy, The University of Chicago
with Elizabeth Huppert, Emma Levine & Jean Decety, Department of Psychology, The University of Chicago
Data Analysis Project Report (STAT 349), The University of Chicago
with Anil K. Ghosh, Indian Statistical Institute, Kolkata