IASC-ARS Interim Conference 2024

Xuming He joined Washington University in July 2023 as the inaugural chair of the Department of Statistics and Data Science. Previously, he served as the H.C. Carver Collegiate Professor of Statistics at the University of Michigan. He is a renowned leader in the fields of robust statistics, quantile regression, Bayesian inference, and post-selection inference; he is also a proponent of interdisciplinary research in data science. Before joining the University of Michigan in 2011, He held positions at the National University of Singapore and the University of Illinois at Urbana-Champaign and served as program director of statistics at the National Science Foundation. He is a fellow of the American Association for the Advancement of Science and the American Statistical Association. Currently, He serves as President (2023-2025) of the International Statistical Institute. He received his bachelor's of science from Fudan University and his master's (mathematics) and PhD (statistics) from the University of Illinois at Urbana-Champaign.

Some Recent Developments in Expected Shortfall Regression

Expected shortfall, measuring the average outcome (e.g., portfolio loss) above a given quantile of its probability distribution, is a common financial risk measure. The same measure can be used to characterize treatment effects in the tail of an outcome distribution, with applications ranging from policy evaluation in economics and public health to biomedical investigations. Expected shortfall regression is a natural approach of modeling covariate-adjusted expected shortfalls. Because the expected shortfall cannot be written as a solution of an expected loss function at the population level, computational as well as statistical challenges around expected shortfall regression have led to stimulating research. We discuss some recent developments in this area, with a focus on a new optimization-based semiparametric approach to estimation of conditional expected shortfall that adapts well to data heterogeneity with minimal model assumptions. The talk is based on joint work with Yuanzhi Li and Shushu Zhang.

Patrick J.F. Groenen is a professor of statistics at the Erasmus School of Economics (ESE). He currently is also dean of that school. Professor Groenen's work focuses on data science techniques and their numerical algorithms. He is the co-author of several textbooks on multidimensional scaling published by Springer and has published articles in the top peer-reviewed journals including, among others, the Journal of Machine Learning Research, the Journal of Marketing Research, Psychological Methods, Psychometrika, the Journal of Classification, Computational Statistics and Data Analysis, the British Journal of Mathematical and Statistical Psychology, and the Journal of Empirical Finance.

Effective MM-Algorithms for Statistics and Machine Learning

As an optimization method, majorization and minorization (MM) algorithms have been applied with success in a variety of models arising in the area of statistics and data science. A key property of majorization algorithms is guaranteed descent, that is, the function value decreases in each step. In practical cases, the function is decreased until it has converged to a local minimum. If the function is convex and coercive, a global minimum is guaranteed. The auxiliary function, the so-called majorizing function, is often quadratic so that an update can be obtained in one step. Here, we present a selection of useful applications of MM algorithms. We discuss its use multidimensional scaling, and in binary and multiclass classification such as logistic regression, multinomial regression, and support vector machines. In the case of regularized generalized canonical correlation analysis, its MM algorithm coincides with several partial least squares (PLS) algorithms thereby providing a previously unknown goal function for the PLS algorithms. We show how MM can also be effective in large scale optimization problems, such as the SoftImpute approach for dealing with missings in principal components analysis, and the MM algorithm for convex clustering.

Cathy W.S. Chen is a distinguished professor in the Department of Statistics at Feng Chia University, Taiwan. She has significantly contributed to the fields of Bayesian inference, diagnostics, model comparison, and forecasting. In addition, she has advanced Bayesian methodology in econometrics and applications of statistics in epidemiology research. She became an elected member of the International Statistical Institute (ISI) in 2008, and in 2010, received the Chartered Statistician (CStat) title from the Royal Statistical Society, a formal recognition of her statistical qualifications, professional training, and experience. She is also a fellow of the American Statistical Association, the Royal Statistical Society, and the International Society for Bayesian Analysis (ISBA). Professor Chen has made considerable contributions through her editorial work. Since 2021, she has served as co-editor for Computational Statistics and as an associate editor for several prestigious journals.

Advances in Spatial Integer-Valued Time Series Modeling for Dengue Fever

Female Aedes aegypti mosquitoes, the primary vectors of dengue, typically remain within 400 meters of their emergence site, while human movement facilitates the spread of the dengue virus to new areas, highlighting the importance of analyzing spatial and temporal patterns in dengue cases. This research integrates spatiotemporal dynamics into multivariate integer-valued GARCH models with generalized Poisson or zero-inflated generalized Poisson (ZIGP) distributions. By employing a flexible and continuous conceptualization of distance, the model represents spatial components without relying on a predefined spatial weight matrix. This approach effectively captures the non-separability of space and time, offering a more nuanced analysis of spatiotemporal relationships. In the second part, a spatial hurdle-type model is introduced, alongside the ZIGP model, incorporating two parameters to adjust the spatial weight decay rate and the shape of the decay curve between locations. These models are applied to time-series counts of dengue hemorrhagic fever within a Bayesian framework using Markov chain Monte Carlo (MCMC) algorithms. Performance evaluations through simulations and multivariate weekly dengue case data demonstrate their effectiveness in capturing spatial dependency, over-dispersion, and the high prevalence of zeros, offering a comprehensive framework for modeling the complex characteristics observed in dengue data.