Ray Bai

#IPAM (the institute for pure and applied mathematics) is facing a critical shortfall for operating expenses due to an unexpected suspension of NSF funding www.ipam.ucla.edu/news/nsf-fun... . Donations for emergency continuity of operations funding can be made at

giving.ucla.edu/Campaign/Donat

I have some related posts on this at mathstodon.xyz/@tao/1149568... and mathstodon.xyz/deck/@tao/11...

New preprint "Parameter expanded variational Bayes for well-calibrated high-dimensional linear regression with spike-and-slab priors"!

Read the full paper here: www.researchgate.net/publication/...

Posing with the Cocky the Gamecock statue

Standing in front of the LeConte College sign (where the USC Statistics Department is located)

Today is my last "official" day at University of South Carolina. Next month, I will be moving to the D.C. area and continuing my academic career at George Mason University. However, I really cherish my time as a faculty member at USC. #ForeverToThee

Read my full post here: tinyurl.com/4fpzjeru

Thank you for all of the support the last few years, @uofscstatistics.bsky.social. I will greatly cherish my time as a faculty member at USC where I got to collaborate with great colleagues, teach and mentor excellent students, and grow as a researcher and educator!

A Unified Three-State Model Framework for Analysis of Treatment Crossover in Survival Trials We present a unified three-state model (TSM) framework for evaluating treatment effects in clinical trials in the presence of treatment crossover. Researchers have proposed diverse methodologies to...

"A Unified Three-State Model Framework for Analysis of Treatment Crossover in Survival Trials" (with my former PhD student Zile Zhao and our co-authors) was published in the most recent issue of Statistics in Biopharmaceutical Research! Check it out here: doi.org/10.1080/1946...

SHORT ABSTRACT: In this work, we propose a novel Bayesian approach for biclustering binary datasets called Binary Spike-and-Slab Lasso Biclustering (BiSSLB). Our method is based on a logistic matrix factorization model with spike-and-slab lasso priors on the latent spaces. We further incorporate an Indian Buffet Process (IBP) prior to automatically determine the number of biclusters. To avoid the high computational cost of EM algorithms, we propose a novel coordinate ascent algorithm with proximal steps which allows for scalable computation. The performance of our proposed approach is assessed via simulations and two real applications on leukemia gene expression data and protein-protein interaction (PPI) data, where BiSSLB is shown to outperform other state-of-the-art binary biclustering methods.

Hard at work this summer before the Fall semester begins. Another new preprint "BiSSLB: Binary spike-and-slab biclustering for binary datasets" with my PhD student Sijian Fan should be available in the next few weeks!

ABSTRACT: Chronic obstructive pulmonary disease (COPD) is one of the leading causes of hospitalization and death in the United States. Although progress has been made in understanding both patient-level and community-level risk factors associated with COPD, there is little work on measuring the causal effects of community exposures on COPD risk. In particular, it remains unclear whether neighborhood income or neighborhood exposure to vape shops causes the number of emergency department (ED) visits for COPD to increase. In this study, we introduce a Bayesian spatial causal inference model to determine the average causal effect of median income and vape shop density on the number of COPD ED visits. Using county-level data from the state of North Carolina in 2023, we found that an increase in a county's median income caused a significant decrease in that county's COPD ED visits. On the other hand, greater exposure to vape shops did not cause a significant change in the number of COPD ED visits. Our findings enhance understanding of how community exposures impact COPD hospitalizations and may aid in place-based interventions to reduce the number of ED visits for COPD.

Another new preprint coming soon, hopefully in the next few weeks! This is work from the summer REU that I supervised this past summer. The undergrad students did a great job on this!

Title: Parameter-Expanded Variational Bayes for Well-Calibrated High-Dimensional Linear Regression with Spike-and-Slab Priors. Abstract: As scientific problems grow in complexity, there is a pressing need for robust and scalable computational methods for fitting high-dimensional statistical models. Variational Bayes (VB) provides an approximate alternative to traditional sampling-based Bayesian inference, often reducing computation time from days to hours or minutes. VB typically minimizes the Kullback-Leibler divergence via coordinate ascent under a mean-field assumption. Its performance can be highly sensitive to prior specifications, particularly in sparse high-dimensional regression with spike-and-slab priors. A significant limitation of standard VB is its tendency to produce poorly calibrated predictions; that is, the predicted values often exhibit a systematic bias relative to the observed outcomes, failing to accurately reflect the true conditional expectation. Motivated by this, we apply parameter expansion to VB and propose a sparse parameter-expanded VB (spexvb) algorithm that improves robustness to prior settings and enhances predictive calibration. Compared to standard VB, spexvb delivers more stable and accurate posterior estimates. We evaluate its performance through extensive simulations and a real-world application, demonstrating the practical advantages of parameter expansion in variational inference for high-dimensional regression.

New preprint coming soon! Variational Bayes (VB) is a scalable approach for conducting Bayesian variable selection when p>>n, but the standard mean-field CAVI algorithm results in poorly calibrated predictions. We introduce a novel parameter-expanded sheme to correct this issue.

The Summer REU I'm supervising wraps up next week! My students did an excellent job studying spatial models for disease mapping and causal inference, and implementing these models on real public health data from North Carolina! Looking forward to their final report and presentation!

Celebrating 8 years of #love and #pride! Happy Pride Month! 🧑‍🤝‍🧑🏳️‍🌈

I had a truly lovely time at BNP 14: the 14th International Conference on Bayesian Nonparametrics at UCLA this past week! I gave a talk, listened to many interesting talks, & chatted with a number of folks. This was my first time attending a BNP conference -- I'll definitely attend again in future!

I'm headed to Los Angeles, CA today to attend the BNP 14 Conference on Bayesian nonparametrics at UCLA! I will be in L.A. this whole upcoming week. Please feel free to reach out if you would like to meet up!

(2/2) We will now be extending our work to a Bayesian spatial causal inference model. We will try to publish a paper from this -- excited to see how this turns out, and grateful for the opportunity to engage with bright undergrad students on challenging statistical problems.

(1/2) My summer REU students did such a great job! They used county-level data and a Bayesian conditional autoregressive (CAR) model to estimate the North Carolina counties' relative risks for COPD ED visits. Not an easy project for undergrads but they did it!

(cont...)

Excited to give a talk for Merck Oncology this Friday! I'll be talking about my work with my former PhD student Zile Zhao on treatment switching in survival trials. I hope that the research team at Merck finds our statistical framework useful for their clinical trials!

Session 04.M2.172: Recent advances in high-dimensional modeling at the IISA 2025 Conference at the University of Nebraska-Lincoln is from 10:45 am-12:15 pm this Sunday, June 15.

Abstract of talk "Misspecified Yet Credible: A Generalized Bayes Framework for Uncertainty Quantification in High-Dimensional Bayesian Vector Autoregressive Models": Vector autoregressive (VAR) models are widely used to capture linear dependencies in multivariate time series, yet Bayesian inferential theory for high-dimensional VAR remains largely undeveloped. We propose a generalized Bayes framework that automatically adapts to sparsity and is robust to misspecification of both the error distribution and covariance structure. Under mild regularity conditions, we show that this approach yields reliable uncertainty quantification for the VAR transition matrices in very high dimensions. As a corollary, the same strategy also delivers valid inference for sparse high-dimensional stochastic regressions with serially corelated errors.

If anyone is attending the IISA 2025 Conference at U. Nebraska-Lincoln this week, check out Session 04.M2.I72: Recent advances in high-dimensional modeling on Sun., Jun 15, 10:45 am-12:15 pm! My recent work on Bayesian VAR models (w/ Partha Sarkar of FSU) will be presented then!

Sparse high-dimensional linear mixed modeling with a partitioned empirical Bayes ECM algorithm - Statistics and Computing High-dimensional longitudinal data is increasingly used in a wide range of scientific studies. To properly account for dependence between longitudinal observations, statistical methods for high-dimens...

Our paper "Sparse high-dimensional linear mixed modeling with a partitioned empirical Bayes ECM algorithm" in Statistics and Computing is now available online! Read at the link below. 👇

Link: doi.org/10.1007/s112...

Excited to work with summer REU students for the next 6 weeks! We will use statistical methods to create beautiful maps of both estimated disease prevalences and counterfactual prevalences under a spatial causal model.

Although I'm moving to GMU at end of July, I'm still so excited to be leading a 6-week summer REU at UofSC starting this Monday! I'll be working on spatial causal inference models for disease mapping with a small team of undergrad researchers. Excited to see what we can accomplish. :)

SHORT ABSTRACT: High-dimensional longitudinal data is increasingly used in a wide range of scientific studies. To properly account for dependence between longitudinal observations, statistical methods for high-dimensional linear mixed models (LMMs) have been developed. However, few packages implementing these high-dimensional LMMs are available in the statistical software R. Additionally, some packages suffer from scalability issues. This work presents an efficient and accurate Bayesian framework for high dimensional LMMs. We use empirical Bayes estimators of hyperparameters for increased flexibility and an Expectation-ConditionalMinimization (ECM) algorithm for computationally efficient maximum a posteriori (MAP) estimation of parameters. The novelty of the approach lies in its partitioning and parameter expansion as well as its fast and scalable computation. A real-world example is provided using data from a study of lupus in children with 15,424 genetic and clinical predictors.

Some other good news! My paper "Sparse high-dimensional linear mixed modeling with a partitioned empirical Bayes ECM algorithm" (with Anja Zgodic, Jiajia Zhang, Peter Olejua, and
Alexander McLain) has been accepted by Statistics and Computing!

Link to paper: tinyurl.com/yu2f8rrv

✨Some news!✨ I have decided to accept a new faculty position at George Mason University, starting this August! I'm looking forward to moving to the Washington D.C. area and continuing my research at GMU.

I greatly valued my time as faculty in @uofscstatistics.bsky.social! Forever to thee!

SHORT ABSTRACT: Periodontal pocket depth is a widely used biomarker for diagnosing risk of periodontal disease. However, pocket depth typically exhibits skewness and heavy-tailedness, and its relationship with clinical risk factors is often nonlinear. Motivated by periodontal studies, this paper develops a robust single-index modal regression framework for analyzing skewed and heavy-tailed data. Our method has the following novel features: (1) a flexible two-piece scale Student-t error distribution that generalizes both normal and two-piece scale normal distributions; (2) a deep neural network with guaranteed monotonicity constraints to estimate the unknown single-index function; and (3) theoretical guarantees, including model identifiability and a universal approximation theorem.

New paper (with Qingyang Liu, Shijie Wang, and Dipankar Bandyopadhyay)! Motivated by periodontal studies, we introduce a new neural network-based robust single-index model. Our method is suitable for skewed and heavy-tailed data.

Read the preprint here: arxiv.org/abs/2505.02153

Congrats to my student Sijian for being one of the winners for Best Student Talk at the SC-ASA Palmetto Symposium today! I really enjoyed all of the student talks. Keep up the great work!

Short abstract for "A Montonic Single-Index Model for Skewed and Heavy-Tailed Data: A Deep Neural Network Approach Applied to Periodontal Studies": Periodontal pocket depth is a widely used biomarker for diagnosing risk of periodontal disease. However, pocket depth typically exhibits skewness and heavy-tailedness, and its relationship with clinical risk factors is often nonlinear. Motivated by periodontal studies, this paper develops a robust single-index modal regression framework for analyzing skewed and heavy-tailed data. Our method has the following novel features: (1) a deep neural network (DNN) with guaranteed monotonicity constraints to estimate the unknown single-index function; (2) a flexible two-piece scale Student-t error distribution that generalizes both normal and two piece scale normal distributions; and (3) theoretical guarantees, including model identifiability and a universal approximation theorem.

New preprint coming out later this month! I'm excited for this work, as I have been continuing to explore deep learning for challenging statistical problems - both from methodological/theoretical and applied perspectives.

Prof. David Hitchcock at the College of Arts and Sciences Tenure & Promotion reception being presented with a certificate by Dean Joel Samuels. Dr. Hitchcock was recently promoted from Associate to Full Professor of Statistics.

Prof. David Hitchcock was recently promoted to Full Professor of Statistics! 🎉🥳Here he is at the @uofsccas.bsky.social Tenure & Promotion reception. Dr. Hitchcock is currently Director of the Graduate program, has co-authored over 50 articles, and has supervised over 30 PhD+MS theses!

A mixed-effects Bayesian regression model for multivariate group testing data ABSTRACT. Laboratories use group (pooled) testing with multiplex assays to reduce the time and cost associated with screening large populations for infecti

New paper by Dr. Christopher McMahan (PhD '12), Prof. Joshua Tebbs, and their co-authors, "A mixed-effects Bayesian regression model for multivariate group testing data," has been published in the most recent issue of Biometrics! Read the paper below.

academic.oup.com/biometrics/a...

Zoom screenshot for the virtual panel "Navigating the Academic Job Market for (Bio)Statistics Faculty & Postdoc Positions"

I had a great time participating in this virtual panel yesterday! We had over 60 people in attendance from the U.S., Canada, and U.K.! Always happy to help out (aspiring) PhD students and junior faculty in any way that I can.

Virtual panel discussion "Academic Job Search for (Bio)statistics Faculty and Postdoc Positions" happening THIS Tuesday, April 1, at 3:30 PM EST! Below are the details for joining the event.

Zoom Meeting: lnkd.in/evxiJbPq
ID: 943 5676 5580
Passcode: 362974

Really enjoying supervising a student on her Senior Honors Thesis. We came up with a novel approach for estimating maternal mortality rates in SC counties with suppressed data. She learned all about Bayesian hierarhical models, CAR models, and model-fitting in Stan! We will try to publish this work.