Location: Public Affairs 2270

Jian Kang, Professor and Associate Chair for Research
School of Public Health, University of Michigan

Abstract:

Recent advances in neuroimaging have produced massive and heterogeneous datasets, ranging from fMRI with high spatial resolution to EEG with high temporal resolution, characterized by complex spatiotemporal correlations and substantial inter-subject variability. Traditional regression models and Gaussian process (GP) approaches with fixed parametric kernels often fail to model such complex data effectively while maintaining scalability and interpretability. This talk introduces a family of modern Bayesian GP frameworks that integrate deep kernel learning, neural network priors, and geometric modeling for large-scale neuroimaging analysis. An example is the Deep Kernel Learning Process (DKLP), which embeds deep neural networks within GP priors to learn data-adaptive covariance structures directly from imaging data. DKLP provides a unified modeling foundation for image-on-scalar, scalar-on-image, and image-on-image regression, supported by theoretical guarantees and efficient posterior computation. Applications to fMRI data from the Adolescent Brain Cognitive Development (ABCD) study reveal reproducible cortical activation patterns associated with cognitive ability, while analyses of EEG-based brain–computer interface data demonstrate robust neural decoding under high noise. I will also discuss scalable heat-kernel GPs on manifolds and thresholded GP–based spatially varying neural network priors, which together expand the scope of Bayesian inference for complex neuroimaging data.

Bio:

Jian Kang is Professor and Associate Chair for Research in the Department of Biostatistics at the University of Michigan, Ann Arbor. He received his B.S. in Statistics from Beijing Normal University in 2005, M.S. in Mathematics from Tsinghua University in 2007, and Ph.D. in Biostatistics from the University of Michigan in 2011. His research focuses on Bayesian modeling, statistical machine learning, and large-scale data integration, with applications in neuroimaging, metabolomics, and precision medicine. Dr. Kang is a Fellow of the American Statistical Association and has served as Associate Editor for the Journal of the American Statistical Association, Annals of Applied Statistics, Biometrics, and Statistics in Medicine. He received the 2025 University of Michigan School of Public Health Excellence in Research Award, the 2024 International Chinese Statistical Association President’s Citation Award, and the 2022 “Best Paper in Biometrics” Award from the International Biometric Society.

Location: Mathematical Sciences 8359

José A. Carrillo, Professor of Nonlinear Partial Differential Equations
Mathematical Institute, University of Oxford

Abstract:

The Stein Variational Gradient Descent method is a variational inference method in statistics that has recently received a lot of attention. The method provides a deterministic approximation of the target distribution, by introducing a nonlocal interaction with a kernel. Despite the significant interest, the exponential rate of convergence for the continuous method has remained an open problem, due to the difficulty of establishing the related so called Stein-log-Sobolev inequality. Here, we prove that the inequality is satisfied for each space dimension and every kernel whose Fourier transform has a quadratic decay at infinity and is locally bounded away from zero and infinity. Moreover, we construct weak solutions to the related PDE satisfying exponential rate of decay towards the equilibrium. The main novelty in our approach is to interpret the Stein-Fisher information, also called the squared Stein discrepancy, as a duality pairing between H⁻¹(ℝⁿ) and H¹(ℝⁿ), which allows us to employ the Fourier transform. We also provide several examples of kernels for which the Stein-log-Sobolev inequality fails, partially showing the necessity of our assumptions.

Bio:

José A. Carrillo is currently Professor of the Analysis of Nonlinear Partial Differential Equations at the Mathematical Institute and Tutorial Fellow in Applied Mathematics at The Queen’s College, University of Oxford. He mainly works on kinetic and nonlinear nonlocal diffusion equations. He has contributed to the theoretical and numerical analysis of PDEs, and their simulation in different applications such as granular media, semiconductors, collective behaviour, and lately in plasmas and tissue modelling. He is currently officer at large at the International Council for Industrial and Applied Mathematics 2024-2028 and head of the Division of the European Academy of Sciences, Section Mathematics where he was elected in 2018. He received the 2022 Echegaray Medal of the Royal Spanish Academy of Sciences and was a plenary speaker at the International Congress on Industrial and Applied Mathematics in Tokyo 2023.

Location: Public Affairs 2270

Helen Hao Zhang, Professor and Chair
Statistics and Data Science GIDP, University of Arizona

Abstract:

Transfer learning has become a cornerstone of modern machine learning, enabling the transfer of knowledge from data-rich source domains to data-scarce target domains. However, determining whether transfer learning will be beneficial prior to implementation remains a critical challenge. This work proposes a novel performance-based framework to measure similarity between source and target datasets and quantify transferability. Theoretical justifications are provided by connecting the new measure to the cosine similarity between decision boundaries in supervised learning. Key advantages of the proposed framework include its nonparametric and flexible nature, easy implementation, and computational scalability without requiring estimation of the underlying data distribution. We further suggest practical guidance that categorizes source datasets into positive, ambiguous, or negative zones based on their transferability. Finally, we extend this approach to encoder-head architectures in deep learning. Numerical results and real-world applications are presented to demonstrate the empirical effectiveness of the framework.

Bio:

Dr. Hao Helen Zhang is Professor of Mathematics and Chair of the Statistics and Data Science Interdisciplinary Program at the University of Arizona. Her research focuses on nonparametric models, statistical machine learning, and high-dimensional data, with applications in biomedical and engineering fields. She has published over 100 research articles and received numerous funding support from NSF, NIH, and NSA, including the NSF CAREER and NSF TRIPODS awards. Dr. Zhang has served as Editor-in-chief of STAT, and Associate Editor of JASA and JRSSB. She is Fellow of IMS, Fellow of ASA, and an IMS Medallion Lecturer.

Location: Public Affairs 2270

Ji Zhu, Susan A. Murphy Collegiate Professor
Department of Statistics, University of Michigan

Abstract:

Most statistical models for networks focus on pairwise interactions between nodes. However, many real-world networks involve higher-order interactions among multiple nodes, such as co-authors collaborating on a paper. Hypergraphs provide a natural representation for these networks, with each hyperedge representing a set of nodes. The majority of existing hypergraph models assume uniform hyperedges (i.e., edges of the same size) or rely on diversity among nodes. In this work, we propose a new hypergraph model based on non-symmetric determinantal point processes. The proposed model naturally accommodates non-uniform hyperedges, has tractable probability mass functions, and accounts for both node similarity and diversity in hyperedges. For model estimation, we maximize the likelihood function under constraints using a computationally efficient projected adaptive gradient descent algorithm. We establish the consistency and asymptotic normality of the estimator. Simulation studies confirm the efficacy of the proposed model, and its utility is further demonstrated through edge predictions on several real-world datasets.

Bio:

Ji Zhu is Susan A. Murphy Collegiate Professor of Statistics at the University of Michigan, Ann Arbor. He received his B.Sc. in Physics from Peking University, China in 1996 and M.Sc. and Ph.D. in Statistics from Stanford University in 2000 and 2003, respectively. His primary research interests include statistical machine learning, statistical network analysis, and their applications to health and natural sciences. He received an NSF CAREER Award in 2008 and was elected as a Fellow of the American Statistical Association in 2013 and a Fellow of the Institute of Mathematical Statistics in 2015. From 2014 to 2020, he was recognized as an ISI Highly Cited Researcher by Web of Science, which annually lists leading researchers in the sciences and social sciences worldwide. In 2022, he received the International Chinese Statistical Association Pao-Lu Hsu Award. He served as the Editor-in-Chief of the Annals of Applied Statistics from 2022 to 2024.

Location: Public Affairs 2270

Lingzhou Xue, Professor
Department of Statistics, The Pennsylvania State University

Abstract:

Random objects are complex non-Euclidean data taking values in general metric spaces, possibly devoid of any underlying vector space structure. Such data are becoming increasingly abundant with the rapid advancement in technology. Examples include probability distributions, positive semidefinite matrices, and data on Riemannian manifolds. However, except for regression for object-valued responses with Euclidean predictors and distribution-on-distribution regression, there has been limited development of a general framework for object-valued responses with object-valued predictors in the literature. To fill this gap, we introduce the notion of a weak conditional Fréchet mean based on Carleman operators and then propose a global nonlinear Fréchet regression model through the reproducing kernel Hilbert space (RKHS) embedding. Furthermore, we establish the relationships between the conditional Fréchet mean and the weak conditional Fréchet mean for both Euclidean and object-valued data. We also show that the state-of-the-art global Fréchet regression developed by Petersen and Müller (Ann. Statist. 47 (2019) 691–719) emerges as a special case of our method by choosing a linear kernel. We require that the metric space for the predictor admit a reproducing kernel. In contrast, the intrinsic geometry of the metric space for the response is utilized to study the asymptotic properties of the proposed estimates. Numerical studies, including extensive simulations and a real application, are conducted to investigate the finite-sample performance.

Bio:

Lingzhou Xue is a Professor of Statistics at Penn State. He received his Ph.D. in Statistics from the University of Minnesota in 2012, and he was a postdoctoral research associate at Princeton University from 2012 to 2013. His research interests include high-dimensional statistics, nonparametric statistics, machine learning, optimization, and statistical modeling in biomedical, environmental, and social sciences. He is a dedicated mentor to Ph.D. students and postdoctoral researchers, and five of his former advisees have become tenure-track faculty members in statistics. He became an Elected Fellow of the Institute of Mathematical Statistics (IMS) in 2024, an Elected Fellow of the American Statistical Association (ASA) in 2023, and an Elected Member of the International Statistical Institute (ISI) in 2016. He received the inaugural Committee of Presidents of Statistical Societies (COPSS) Emerging Leader Award in 2021, the inaugural Bernoulli Society New Researcher Award in 2019, and the International Consortium of Chinese Mathematicians Best Paper Award in 2019.

Location: Mathematical Sciences Building 8359

Houman Owhadi, Professor of Applied and Computational Mathematics and Control and Dynamical Systems
Computing and Mathematical Sciences Department, California Institute of Technology

Abstract:

We introduce a novel kernel-based framework for learning differential equations and their solution maps, which is efficient in terms of data requirements (both the number of solution examples and the amount of measurements from each example), as well as computational cost and training procedures. Our approach is mathematically interpretable and supported by rigorous theoretical guarantees in the form of quantitative worst-case error bounds for the learned equations and solution operators. Numerical benchmarks demonstrate significant improvements in computational complexity and robustness, achieving one to two orders of magnitude improvement in accuracy compared to state-of-the-art algorithms. This presentation is based on joint work with Yasamin Jalalian, Juan Felipe Osorio Ramirez, Alexander Hsu, and Bamdad Hosseini. A preprint is available at: https://arxiv.org/abs/2503.01036

Bio:

Houman Owhadi is an IBM professor of applied and computational mathematics and control and dynamical systems at the California Institute of Technology. His expertise includes uncertainty quantification, numerical approximation, statistical inference/learning, data assimilation, stochastic and multiscale analysis, and scientific machine learning. He was a plenary speaker at SIAM CSE 2015, SIAM UQ 2024 and EMI 2025, and a tutorial speaker at SIAM UQ 2016. He received the 2019 Germund Dahlquist SIAM Prize. He is a SIAM Fellow (class of 2022) and a Vannevar Busch Fellow (class of 2024).

Location: Public Affairs 2270

Heping Zhang, Susan Dwight Bliss Professor of Biostatistics
Yale University

Abstract:

Regression, clustering, and sequential analysis are fundamental techniques in statistics. Today, these same concepts are often relabeled as supervised learning, unsupervised learning, deep learning, reinforcement learning, or, more broadly, artificial intelligence. In this talk, I will present several of our statistical methods, developed in response to real-world applications, including the analysis of high-dimensional data for building-related occupant syndromes, inference of risk factors with uncertain frequencies from haplotype data, and residual diagnostics for generalized linear models. By revisiting these examples, I will highlight the essential ideas and techniques that our approaches share with modern AI methods. My goal is to reflect on why our statistical methods appear so “unadorned,” and to ask whether—and how—we might close the gap in how statistics and AI are recognized and valued.

Bio:

Heping Zhang, Ph.D., is the Susan Dwight Bliss Professor of Biostatistics at the Yale University School of Public Health. He also holds secondary appointments as Professor in the Child Study Center and the Department of Obstetrics, Gynecology, and Reproductive Sciences at the Yale School of Medicine, and in the Department of Statistics and Data Science at Yale University. He is the founding director of the Collaborative Center for Statistics in Science at Yale. Dr. Zhang is a Fellow of both the American Statistical Association and the Institute of Mathematical Statistics. He was the founding Editor-in-Chief of Statistics and Its Interface and previously served as an editor of the Journal of the American Statistical Association – Applications and Case Studies. His honors include the 2008 Myrto Lefkopoulou Distinguished Lecture at the Harvard School of Public Health, the 2011 IMS Medallion Lecture and Award, the 2022 Neyman Lecture and Award, the 2023 Distinguished Achievement Award from the International Chinese Statistical Association, and recognition as a 2023 Highly Cited Researcher by the Web of Science.

Location: Public Affairs 2270

Qing Nie, Distinguished Professor of Mathematics and Developmental & Cell Biology
University of California, Irvine

Abstract:

Cells make fate decisions in response to dynamic environments, and multicellular structures emerge from multiscale interplays among cells and genes in space and time. While single-cell omics data provides an unprecedented opportunity to profile cellular heterogeneity, the technology requires fixing the cells, often leading to a loss of spatiotemporal and cell interaction information. How to reconstruct temporal dynamics from single or multiple snapshots of single-cell omics data? How to recover interactions among cells, for example, cell-cell communication from single-cell gene expression data? I will present a suite of our recently developed computational methods that learn the single-cell omics data as a spatiotemporal and interactive system. Those methods are built on a strong interplay among systems biology modeling, dynamical systems approaches, machine-learning methods, and optimal transport techniques. The tools are applied to various complex biological systems in development, regeneration, and diseases to show their discovery power. Finally, I will discuss the methodology challenges in systems learning of single-cell data. Dr. Qing Nie is a University of California Presidential Chair and UCI Excellence in Teaching Chair, and a Distinguished Professor of Mathematics and Developmental & Cell Biology at University of California, Irvine. In research, Dr. Nie uses systems biology and data-driven methods to study complex biological systems with focuses on single-cell analysis, multiscale modeling, cellular plasticity, stem cells, embryonic development, and their applications to diseases. In training, Dr. Nie has supervised more than 60 postdoctoral fellows and PhD students, with many of them working in academic institutions now.

Bio:

Dr. Nie is a fellow of the American Association for the Advancement of Science (AAAS), a fellow of American Physical Society (APS), a fellow of Society for Industrial and Applied Mathematics (SIAM), and a fellow of American Mathematical Society (AMS). In 2025, Dr. Nie was ranked #1 based on the data analytics of publications and citations by ScholarGPS in the Highly Ranked Scholar list for two areas: Single-cell Transcriptomics and Transcriptomics Technologies, for the prior five years.

Location: Public Affairs 2270

Yingying Fan, Centennial Chair in Business Administration & Professor of Data Sciences and Operations
USC Marshall School of Business

Abstract:

We propose a unified theoretical framework for studying the robustness of the model-X knockoffs framework by investigating the asymptotic false discovery rate (FDR) control of the practically implemented approximate knockoffs procedure. This procedure deviates from the model-X knockoffs framework by substituting the true covariate distribution with a user-specified distribution that can be learned using in-sample observations. By replacing the distributional exchangeability condition of the model-X knockoff variables with three conditions on the approximate knockoff statistics, we establish that the approximate knockoffs procedure achieves the asymptotic FDR control. Using our unified framework, we further prove that an arguably most popularly used knockoff variable generation method–the Gaussian knockoffs generator based on the first two moments matching–achieves the asymptotic FDR control when the two-moment-based knockoff statistics are employed in the knockoffs inference procedure. For the first time in the literature, our theoretical results justify formally the effectiveness and robustness of the Gaussian knockoffs generator. Simulation and real data examples are conducted to validate the theoretical findings.

Bio:

Yingying Fan is Associate Dean for the PhD Program, Centennial Chair in Business Administration, Professor in Data Sciences and Operations Department of the Marshall School of Business at the University of Southern California. Her research interests include statistics, data science, machine learning, economics, and big data and business applications. Her latest works have focused on knockoff inference, causal inference, and LLM model applications. She is the recipient of the Institute of Mathematical Statistics Medallion Lecture, Fellow of Institute of Mathematical Statistics, Fellow of American Statistical Association, the Royal Statistical Society Guy Medal in Bronze, the American Statistical Association Noether Young Scholar Award, and the NSF Faculty Early Career Development (CAREER) Award. She is a Co-Editor of the Journal of Business and Economic Statistics and Statistics Surveys.