Hello, I'm Shitao Fan
Ph.D. Candidate in Statistics · Bayesian Methods · Variational Inference · Generative AI
Download CV
Ph.D. Candidate in Statistics · Bayesian Methods · Variational Inference · Generative AI
Download CV
I am a Ph.D. candidate in Statistics at the University of Maryland, specializing in Variational Bayesian methods. Experienced with generative AI, deep learning, and nonparametric statistics; strong emphasis on reproducible pipelines and theory–practice integration.
An ensemble-style correction for mean-field variational inference that addresses covariance underestimation and lost dependencies, improving posterior uncertainty quantification and predictive performance under model misspecification.
Time-dependent structured variational families with learned inference maps that capture temporal and cross-component dependencies. Includes risk bounds for the amortized posterior and a decomposition of the amortization gap.
A geometric-median aggregation of split-data variational posteriors (VM-posterior) that yields strong robustness to outliers and contamination, with posterior contraction rates and Bernstein–von Mises guarantees in scalable settings.
Email: sfan211@umd.edu
GitHub: waterism211