Home
Publications
Software
Teaching

Boxiang Wang

alt text 

Welcome to my home page!

I have been Associate Professor and Director of Graduate Studies in the Department of Statistics and Actuarial Science at The University of Iowa since 2024. I am also a Center Faculty at Center for Advancing Multimorbidity Science.

I obtained my Ph.D. in the School of Statistics at the University of Minnesota under the supervision of Professor Hui Zou. I got my master's degree in the Department of Applied Statistics and Operation Research at the Bowling Green State University, and my bachelor's degree in the School of Mathematics at Nankai University.

My main research focus is at the crossroads of statistics, machine learning, and optimization. I specialize in supervised learning algorithms such as support vector machines, unsupervised learning algorithms including model-based clustering, model assessment, tensor data analysis, and optimal experimental design. I am also interested in actuarial science and insurance applications.

  • Address: 261 Schaeffer Hall, Iowa City, IA 52242

  • Phone: 319-335-2294

  • E-mail: boxiang-wang at uiowa.edu

Publications

Software

R package: hdqr 

The R package implements an efficient algorithm to obtain exact solutions for penalized quantile regression models based on a finite smoothing algorithm.

R package: hdsvm 

The R package implements an efficient algorithm for sparse penalized support vector machine models using the generalized coordinate descent algorithm. Designed to handle high-dimensional datasets effectively, with emphasis on precision and computational efficiency

R package: PIE 

The R package implements a novel predictive model, Partially Interpretable Estimators (PIE), which jointly trains an interpretable model and a black-box model to achieve high predictive performance as well as partial model.

R package: fastkqr 

The R package efficiently fits and tunes kernel quantile regression models based on the majorization-minimization method. It can also fit multiple quantile curves simultaneously with a novel non-crossing penalty.

R package: dcsvm 

The R package implements an efficient algorithm for solving sparse-penalized support vector machines with kernel density convolution. This package is designed for high-dimensional classification tasks, supporting lasso (L1) and elastic-net penalties for sparse feature selection and providing options for tuning kernel bandwidth and penalty weights. The package is applicable to fields such as bioinformatics, image analysis, and text classification, where high-dimensional data commonly arise.

R package: ARTtransfer 

The R package implements a flexible framework for transfer learning that integrates information from auxiliary data sources to improve model performance on primary tasks. It is designed to be robust against negative transfer by including the non-transfer model in the candidate pool, ensuring stable performance even when auxiliary datasets are less informative.

R package: ktweedie 

The R package fits kernel-based 'Tweedie' compound Poisson gamma models using high-dimensional predictors for the analyses of zero-inflated response variables. The package features built-in estimation, prediction and cross-validation tools and supports choice of different kernel functions.

R package: kerndwd 

The R package kerndwd uses the majorization-minimization principle to solve the linear DWD. It also formulates the kernel DWD in an reproducing kernel Hilbert space and develops the same algorithm for linear DWD. The package involves very fast tuning procedures and delivers prediction accuracy that is highly comparable as the kernel SVM.

R package: sdwd 

The R package sdwd uses coordinate descent to solve sparse distance weighted discrimination for high-dimensional classification. The package computes the entire solution path for lasso, elastic net, and adpative lasso/elastic net penalites. The implementation is efficient involving computational tricks such as strong rule, warm start, and active set. The R package sdwd is extremely fast, as compared with some sparse support vector machines.

R package: CUSUMdesign 

The R package CUSUMdesign employs the Markov chain algorithm to compute the average run length and the decision interval when the CUSUM charts are designed. The CUSUM chart is widely used for detecting small but persist shifts in statistical process control.

Honors & Awards