Postdoctoral Research in Statistical Modeling, Machine Learning, and Privacy-Aware Data Analysis
I am a Postdoctoral Research Fellow at George Mason University. My recent work spans healthcare analytics, robust learning under distribution shift, large language models, transportation safety, and privacy-preserving statistical inference.
Professional Experience
A focused snapshot of my current postdoctoral work.
George Mason University. Working on robust machine learning, uncertainty quantification, and privacy-aware modeling for healthcare and transportation datasets.
Publications
When a public version is available, the title links directly to the paper.
Current manuscript on robust LLM training under epistemic uncertainty and dependent conversational logs.
Collaborative work on privacy risks and data quality issues in prior authorization records.
Privacy-aware estimation with robust inferential guarantees and new Hellinger-based privacy machinery.
Sharp asymptotics, Value at Risk, Expected Shortfall, and Gibbs conditioning for rare portfolio-loss events.
Current manuscript on local limits, oscillation behavior, and large deviations across growth regimes.
Robust and privacy-aware inference for Hawkes process data with a link to integer-valued autoregression.
Research
A mix of current postdoctoral work and recent statistical research. Each project page gives a short overview, key methods, and currently available materials.
Clinical classification with controlled false positive rates, stable biomarker selection, and repeated sample-splitting for heart transplant rejection outcomes.
Read projectRobust supervised fine-tuning under distribution shift using uncertainty-aware reweighting, paraphrase stress tests, and KL-constrained training.
Read projectAn LLM-powered crash analysis pipeline that enriches structured records with external context and produces analyst-friendly case reports.
Read projectA privacy framework that unifies and extends differential privacy notions while keeping estimation robust and statistically efficient.
Read projectThreshold-model analysis of portfolio losses with sharp extreme-risk estimates and a Gibbs conditioning view of obligor defaults.
Read projectRobust inference for event data with links between Hawkes process models and integer-valued autoregression.
Read projectSkills
Methods and tools most relevant to applied quantitative, machine learning, and research roles.
Background
Additional academic background and professional activity.