UCLA Statistics & Data Science
Algorithms & Innovations in Data Science
Celebrating the Legacy of Jan de Leeuw
Celebrating the Legacy of Jan de Leeuw
Join us for a one-day workshop celebrating the remarkable contributions of Jan de Leeuw to statistics, data science, and computational methods. Leading researchers will present talks spanning dimension reduction, multilevel analysis, optimization algorithms, and modern machine learning. A poster session offers an opportunity for emerging scholars to share their work.
Space is limited. RSVP now to reserve your spot, or submit a poster proposal to present your work at the workshop.
Dr. Jan de Leeuw is a renowned Dutch statistician and psychometrician who is Distinguished Professor Emeritus and the founding Chair of the Department of Statistics at UCLA. After earning his Ph.D. cum laude from Leiden University in 1973, he began his career at Bell Labs and Leiden University, eventually joining UCLA in 1987 as a professor of psychology and mathematics. Dr. de Leeuw's methodological contributions span nonlinear multivariate analysis, multilevel modeling, and data theory. His extensive editorial leadership includes serving as the founding Editor-in-Chief of the Journal of Statistical Software, as well as Editor-in-Chief of the Journal of Multivariate Analysis and the Journal of Educational and Behavioral Statistics. Widely recognized for his impact on the field, he is a past President of the Psychometric Society and an elected Fellow of the Royal Statistical Society, the International Statistical Institute, the Institute of Mathematical Statistics, and the American Statistical Association, in addition to being a member of the Royal Netherlands Academy of Arts and Sciences.
Since the seminal paper of Box and Cox (1964, JRSS-B), the Box-Cox model has been a must-taught topic in any applied regression course. However, its broader practical impact has remained limited. Even within the academic circle some had raised serious concerns. We challenge those concerns by demonstrating the Box-Cox model's power in modern high-dimensional large-scale data analysis. We propose a nonparametric Box-Cox model and a novel composite likelihood approach for ultra-high dimensional inference. It is demonstrated that the nonparametric Box-Cox regression can significantly improve prediction accuracy over the standard high-dimensional regression, while maintaining its core advantages: computational efficiency and model transparency. We therefore suggest that the nonparametric Box-Cox model should be a routine choice for high-dimensional regression analysis.
Dr. Hui Zou is Dr. Lynn Lin Distinguished Professor at University of Minnesota. His research interests include high-dimensional statistics, machine learning and statistical optimization. Many of his papers are highly cited. In 2025 Hui received the International Chinese Statistics Association Pao-Lu Hsu award and the International Statistical Institute Founders of Statistics prize. Hui is an elected Fellow of IMS, ASA and AAAS.
As a tribute to the seminal contributions of Jan de Leeuw, this talk will survey applications of the MM principle, a framework for constructing monotone optimization algorithms in high-dimensional models. The MM principle transfers optimization from the objective function to a surrogate function and simplifies matters by: (a) separating the variables of a problem, (b) avoiding large matrix inversions, (c) linearizing a problem, (d) restoring symmetry, (e) dealing with equality and inequality constraints gracefully, and (f) turning a nondifferentiable problem into a smooth problem. The art in devising an MM algorithm lies in choosing a tractable surrogate function g(x | x_n) that hugs the objective function f(x) as tightly possible. The EM principle from statistics is a special case of the MM principle. Modern mathematical themes such as sparsity and parallelization mesh well with the MM principle.
Dr. Kenneth Lange is the Rosenfeld Professor of Computational Genetics in the Departments of Computational Medicine, Human Genetics, and Statistics at UCLA. He was elected to the National Academy of Sciences in 2021. Lange served as chair of the Department of Computational Medicine for 9 years and as chair of the Department of Human Genetics for 12 years. During his academic career, he has mentored 24 doctoral students and 12 postdoctoral fellows and authored seven advanced textbooks on applied mathematics and statistics. Lange won the Snedecor award from the Joint Statistical Societies in 1993 and the Education Award from the American Society of Human Genetics in 2020. He was an invited speaker at the International Congress of Mathematicians in 2014. His de Leeuw number is 1.
Among other things, Jan is known as the principal architect of the Gifi system and for his contributions to multidimensional scaling (MDS). Gifi is a dimension reduction framework in which optimal scaling is the key ingredient, allowing many classical multivariate techniques, such as principal component analysis (PCA), to be applied to mixed input data. MDS is another dimension reduction technique that represents proximities among objects as distances between points in a low-dimensional space. In this talk, I will discuss connections between Gifi and MDS, review Jan's contributions to both frameworks, examine the relevance of these techniques in the modern data analysis era, and present some recent developments we have been working on.
Dr. Patrick Mair earned his PhD in Statistics from the University of Vienna in 2005 and completed his Habilitation (Venia Legendi) in Statistics in 2010. From 2005–2011 he worked as an Assistant Professor at the Department of Statistics and Mathematics, WU Vienna University of Economics and Business. From 2007–2008 he was a Research Fellow at the Department of Statistics/Department of Psychology at UCLA. Since 2013 he has been working as Senior Lecturer in Statistics at the Department of Psychology at Harvard University. His research focuses on computational and applied statistics, with a particular emphasis on psychometric methods.
Many types of data are nested. A typical example is a study of educational test outcomes, with students nested within schools. Multilevel analysis studies such data, and especially heterogeneity among the higher-level units (schools). For example, to what extent does time spent on homework affect test outcomes differently across schools, and can those differences be explained by characteristics of the schools? Jan de Leeuw is one of the founding fathers of this field. This presentation introduces the field and highlights De Leeuw's contributions to the field.
Dr. Erik Meijer is a senior economist at the Center for Economic and Social Research at the University of Southern California. He received his PhD in Social Sciences from Leiden University. His methodological work includes measurement error and multilevel analysis, among others. His substantive work has spanned transportation, economics, health, and other topics. Much of his current research is in the study of aging: health, cognition, dementia, Social Security, retirement, and economic well-being. With Jan de Leeuw, he co-edited the Handbook of Multilevel Analysis (Springer, 2008).
We present a new coordinate descent algorithm for learning Bayesian networks from continuous observational data, generated according to a linear structural equation model. The proposed algorithm optimizes the L0-penalized maximum likelihood function, which is known to have favorable statistical properties but is computationally challenging to solve. The proposed algorithm approximates this estimator and achieves several remarkable properties: The algorithm converges to a coordinate-wise minimum, and despite the non-convexity of the loss function, as the sample size tends to infinity, the objective value of the coordinate descent solution converges to the optimal objective value of the L0-penalized maximum likelihood estimator, providing the first asymptotic optimality guarantees for coordinate descent for the non-convex Bayesian network estimation problem.
Dr. Ali Shojaie is the Norman Breslow Endowed Faculty Professor and Interim Chair of the Department of Biostatistics, as well as a Professor of Statistics, at the University of Washington (UW). In addition to his departmental leadership, he serves as the founding director of the UW Summer Institute for Statistics in Big Data (SISBID) and the Lead of the Data Management and Statistics (DMS) Core for the UW Alzheimer's Disease Research Center (ADRC). His research operates at the intersection of statistical machine learning and statistical network analysis, with key applications spanning the biological and social sciences. Recognized for his significant contributions to the field, Dr. Shojaie is an elected Fellow of both the American Statistical Association (ASA) and the Institute of Mathematical Statistics (IMS), and he is a recipient of the prestigious 2022 Leo Breiman Award from the ASA Section on Statistical Learning and Data Science (SLDS).
Tensor decomposition is a fundamental tool for analyzing multi-dimensional data and is widely used in fields such as biomedical imaging and signal processing. In multi-source studies, tensor data often exhibit substantial heterogeneity, yet most existing methods focus on estimating common patterns shared across datasets rather than dataset-specific structures. We propose an individualized Tucker decomposition framework that explicitly targets accurate recovery of dataset-specific tensor components. Each observed tensor is decomposed into an individualized component and a shared component, with the shared component serving as an auxiliary structure to enhance individual data prediction and classification through removing population-level effects. An efficient two-step estimation procedure is developed with data-driven Tucker rank selection and accurate estimation of the individualized components. We establish theoretical guarantees, including consistency of rank selection and non-asymptotic error bounds for both the shared and individualized components. Extensive simulations and a retinal imaging application demonstrate that our method effectively captures dataset-specific heterogeneity and consistently outperforms existing Tucker decomposition approaches in recovering individualized tensor structures.
Dr. Annie Qu is a Professor in the Department of Statistics and Applied Probability at UC Santa Barbara, having previously served as a Chancellor's Professor at UC Irvine and the Data Science Founder Professor at the University of Illinois Urbana-Champaign (UIUC). After earning her Ph.D. in Statistics from Pennsylvania State University in 1998, she built a distinguished career recognized by numerous prestigious honors, including the IMS Carver Medal (2025), the IMS Medallion Award and Lecture (2024), and an NSF CAREER Award. Dr. Qu is an Elected Fellow of the American Association for the Advancement of Science (AAAS), the Institute of Mathematical Statistics (IMS), and the American Statistical Association (ASA). Beyond her research and teaching excellence, she is a highly active leader within the statistical community; she currently serves as Co-Editor of the Journal of the American Statistical Association (JASA, Theory and Methods), IMS Program Secretary, and Chair of the ASA Council of Sections Governing Board.
Time series prediction inference is an important yet challenging task in economics and business, where existing approaches often rely on low-frequency, survey-based data. With the recent advances of large language models (LLMs), there is growing potential to leverage high-frequency online text data for improved time series prediction, an area still largely unexplored. This paper proposes LLM-TS, an LLM-based approach for time series prediction inference incorporating online text data. The LLM-TS is based on a joint time series framework that combines survey-based low-frequency data with LLM-generated high-frequency surrogates. The framework relies only on an error correlation assumption, combining a text-embedding-augmented ARX model for the observed gold-standard measurements with a VARX model for the LLM-generated surrogates. LLM-TS employs LLMs such as ChatGPT and the trained BERT models to construct LLM surrogates. Online text embeddings are extracted via LDA and BERT. We establish the asymptotic properties of the method and provide two forms of constructed prediction intervals.
Dr. Yingying Fan is currently Associate Dean for the PhD Program, Centennial Chair in Business Administration, and a Professor in both the Marshall School of Business and the Department of Economics at the University of Southern California. Since earning her Ph.D. from Princeton University in 2007, she has built a highly interdisciplinary research portfolio focusing on statistics, data science, artificial intelligence, large language models, and their applications in economics and business. Her numerous honors include the 10th International Congress of Chinese Mathematicians (ICCM) Best Paper Gold Award (2025), the Institute of Mathematical Statistics (IMS) Medallion Lecture (2023), the Royal Statistical Society Guy Medal in Bronze (2017), and the NSF Faculty Early Career Development (CAREER) Award (2012). Dr. Fan is a Fellow of the IMS and the ASA.