Two other papers pertaining to my doctoral research are now published

I am delighted to announce that two other papers of mine have been recently published in the Annals of Applied Statistics and the American Journal of Preventive Medicine. While both papers are related to my doctoral research, they address completely different research questions with one proposing more theoretical developments, and the other is more focused on application. Hereby, I would like to highly thank my advisers at the University of Michigan, Professors Michael, R. Elliott, Carol A. C. Flannagan, Brady T. West, and Neil K. Mehta, for all their support and encouragement, scientifically and financially.

The one entitled “Robust Bayesian Inference for Big Data: Combining Sensor-based Records with Traditional Survey Data” proposes an alternative doubly robust approach to Chen et al (2019) for inference based on non-probability samples. As the main advantage, this method allows for more flexible algorithmic predictive tools as well as Bayesian non-parametric models to be used for predicting both outcomes as well as propensity scores, which offers additional shields against model misspecification. In this paper, I particularly employ Bayesian Additive Regression Trees (BART) as the underlying modeling approach. BART not only automatically accounts for non-linear as well as high-order interaction effects, but also permits direct quantification of the uncertainty by simulating the posterior predictive distribution.

The second paper is entitled “Incidence in US Children and Young Adults: A Pooled Analysis“. In this research, we analized pooled data from five 5 high-quality nationally representative panel surveys—National Longitudinal Survey of Youth 1979 and 1997, National Longitudinal Study of Adolescent Health, and Early Childhood Longitudinal Study-Kindergarten cohorts of 1998 and 2011. We employed discrete-time survival analysis to model the dynamics of the obesity risk. Our model utilized a penalized spline function to smooth the estimate of incidence over time. Check out the following link if you find it interesting, and want to know further about our research findings.

arafei

Seasoned statistician and data scientist with 10+ years of combined industry and academic experience in coding, deploying ML models and algorithms with more focus on data fusion, causal inference, anomaly detection, and Bayesian computations using petascale organic data; expert in the design and analysis of complex samples, NN & DL, feature selection, model evaluation, NLP, Monte Carlo simulations, network analysis, A/B testing, uncertainty quantification, time-series analysis, revenue optimization, and responsible AI; author of multiple open-source projects (GitHub) and 40+ research articles published in top tier journals with 1,250+ citations received (GScholar); Strongly passionate about developing scalable methods that establish a unified framework for mitigating fairness and propagating uncertainty when training deep generative models using Bayesian approaches.