I am a PhD student in Statistics at the University of Connecticut, advised by HaiYing Wang and Jun Yan. My work centers on massive-data subsampling, model averaging, and survival analysis, with a steady interest in building practical statistical software.
Efficient statistical modeling for large-scale data, software development, and real-world application.
Maintainer of the R package subsampling on CRAN.
I study principled subsampling strategies that preserve statistical efficiency while reducing computational cost. This includes work on generalized linear models, quantile regression, rare events, and rare binary features.
My background includes accelerated failure time models, additive hazards models, and frequentist model averaging methods for censored outcomes.
I am especially interested in work where theoretical development, computation, and application remain tightly connected, including clinical trial design and other collaborative data problems.
I am developing a first-authored methodology paper for regression settings with rare binary covariates, including theory clarifying why estimation becomes unstable and a balanced subsampling framework that improves rare-feature representation without pilot sampling.
I implemented scalable subsampling methods in R for generalized linear models, rare-feature settings, softmax regression, rare event logistic regression, and quantile regression, with documentation and reproducible examples.
I coauthored work on predictive-modeling-assisted interim analysis for censored time-to-event trials, including covariate-informed prediction for censored participants and evaluation metrics for conditional power accuracy and futility decisions.
I helped frame and prepare a study of how social determinants of health relate to frailty and accelerated aging among breast cancer survivors. My work focused on retrieving, cleaning, and integrating All of Us EHR, survey, and physical-measurement data for subsequent statistical analysis.
I analyzed fluorescence response data from porphyrinoid sensors using clustering and statistical summaries to support compound differentiation and sensor selection.
I am contributing to a survey project on time series explainability with an emphasis on LLM-enabled semantic explanations, including benchmark curation and related research synthesis; the manuscript is currently under review.
Selected papers and links to PDFs.
Formal training and academic background.
Advisors: HaiYing Wang and Jun Yan.
Advisor: Hui Zhao.
Thesis: Variable Selection and Model Averaging Methods of Accelerated Failure Time Models.
Thesis: Automatic Marking System based on Convolutional Neural Networks.