When: Thursday, August 27, 2026 — 2:50 p.m. to 3:50 p.m.
Where: LeConte 224

Speaker: Dr. Nikos Ignatiadis, Department of Statistics, University of Chicago

Abstract: A common task in high-throughput biology is to screen for associations across thousands of units of interest, e.g., genes or proteins. Often, the data for each unit are modeled as Gaussian measurements with unknown mean and variance and are summarized as per-unit sample averages and sample variances. The downstream goal is multiple testing for the means. In this domain, it is routine to "moderate" (that is, to shrink) the sample variances through parametric empirical Bayes methods before computing p-values for the means. Such an approach is asymmetric in that a prior is posited and estimated for the nuisance parameters (variances) but not the primary parameters (means). Our work initiates the formal study of this paradigm, which we term "empirical partially Bayes multiple testing." In this framework, if the prior for the variances were known, one could proceed by computing p-values conditional on the sample variances---a strategy called partially Bayes inference by Sir David Cox. We show that these conditional p-values satisfy an Eddington/Tweedie-type formula and are approximated at nearly-parametric rates when the prior is estimated by nonparametric maximum likelihood. The estimated p-values can be used with the Benjamini-Hochberg procedure to guarantee asymptotic control of the false discovery rate. Even in the compound setting, wherein the variances are fixed, the approach retains asymptotic type-I error guarantees.

When: Thursday, September 3, 2026 — 2:50 p.m. to 3:50 p.m.
Where: LeConte 224

Speaker: Dr. Rebecca Killick, Department of Mathematics and Statistical Sciences, Clemson University

Abstract: Multiple changepoint analyses have become an important tool in modern statistics. Classical approaches to the problem include dynamic programming, binary segmentation procedures and their variants, and windowed approaches. Fast dynamic programming procedures that optimize penalized likelihoods only apply when all model parameters change at each and every changepoint time, which is often physically unrealistic. Similarly, windowed approaches either assume that all dynamics change at each changepoint time, or that aspects that do not change at the changepoint time vary across windows. Penalized likelihood methods for the general case, where only a subset of parameters are allowed to change at the changepoint times, require extensive computational searches, classically done via genetic algorithms, to locate the optimal changepoint configuration. This talk discusses a new method that rapidly estimate optimal penalized likelihood changepoint configurations in the general case, bypassing the slow computational (and stochastic) drawbacks of genetic algorithms. Consistency of the changepoint configuration and model parameters under infill asymptotics are proven; the procedure is shown to work well in finite samples via simulation. Applications to environmental and business problems are detailed.

When: Thursday, September 17, 2026 — 2:50 p.m. to 3:50 p.m.
Where: LeConte 224

Speaker: Dr. Lulu Kang, Department of Mathematics and Statistics, University of Massachusetts Amherst

Abstract: Gaussian process (GP) regression is a popular surrogate modeling tool for computer simulations in engineering and scientific domains. However, it often struggles with high computational costs and low prediction accuracy when the simulation involves too many input variables. In this talk, I will present two approaches for building Gaussian process surrogate models for experiments with high-dimensional inputs. First, I introduce an optimal kernel-learning approach to identify active variables, thereby overcoming GP model limitations and enhancing system understanding. This method approximates the original GP model's covariance function through a convex combination of kernel functions, each using low-dimensional subsets of input variables. Second, I introduce a Bayesian bridge GP regression approach, in which we impose a shrinkage penalty on the linear regression coefficients in the mean function and the correlation coefficients in the covariance function. Under a Bayesian framework, this is equivalent to using certain proper informative priors on these parameters. Using Spherical Hamiltonian Monte Carlo, we can directly sample from the constrained posterior distribution without the prior-distribution restrictions imposed in Bayesian bridge regression.

When: Thursday, September 24, 2026 — 2:50 p.m. to 3:50 p.m.

Where: LeConte 224

Speaker: Dr. Minjeong Jeon, Department of Education, University of California, Los Angeles

Abstract: In this talk, I will introduce a novel approach for longitudinal assessments that involve item responses from individuals at two or more time points. A key limitation of existing longitudinal models is their inability to capture item-by-person interactions that can change over time. To address this, I propose an interaction map approach that can capture and visualize time-varying person-by-item interactions, offering valuable insights into individuals’ progress over time. Furthermore, I will present a more structured version of the interaction map approach, which focuses on tracking individuals’ progress toward a measurement target directly within the map. Real-world examples will be shared to illustrate the practical applications of the proposed methodologies.

When: Thursday, October 8, 2026 — 2:50 p.m. to 3:50 p.m.
Where: LeConte 224

Speaker: Dr. Andrew Chen, Department of Public Health Sciences, Medical University of South Carolina

Abstract: National and international imaging consortia have formed with the goal of precisely characterizing the human brain across the lifespan. These consortia have succeeded in collecting large samples of brain magnetic resonance imaging (MRI) scans to estimate sex-specific trajectories of brain phenotypes across age, often called brain charts. The promise of brain charts is that future researchers and clinicians will be able to assess a new scan for deviations from this healthy trajectory. However, the implementation of brain charts in practice is severely limited by differences in image acquisition, preprocessing, and study populations. Here, we first discuss several projects in harmonization of MRI data tailored to this setting. Then, we present a new method to calibrate brain charts, leveraging conformal prediction. We compare these approaches in consortium data based on error rate in heathy individuals and sensitivity in Alzheimer's disease patients. We conclude by providing methodological recommendations for applying brain charts to new samples.

When: Thursday, October 22, 2026 — 2:50 p.m. to 3:50 p.m.
Where: LeConte 224

Speaker: Dr. Jungeum Kim, Department of Statistics, North Carolina State University

Abstract: Probing the hypothesis space after seeing the data remains valid  under possibilistic inferential models (IMs), provided the significance level stays fixed. The price is computation, as each plausibility is a supremum of the possibility contour over the hypothesis, and the contour itself is approximated at each queried parameter value. We propose a possibilistic radial transport, which hides the contour value of a parameter in the radius of its source point. When a transport that maximizes within-shell entropy is picked, sampling parameters covering a confidence cut becomes a matter of truncating the radius. We provide a deep learning algorithm that enforces the contour depth condition while maximizing the entropy within each shell. Our amortization makes coverage and power assessments 
of the learned approximation practical as well as predictive check of new datasets. We also use the sampler to construct a Bel-Pl spectrum for comparing and selecting interpretable hypotheses that satisfy a prescribed Bel--Pl decision criterion. In simulations the learned contours match or improve on ellipsoidal approximations to the cuts, while the coverage and power track the exact reference. Finally, we probe hypotheses about ovarian aging using synthetic AMH records, asking for each woman how many more years her median AMH level will remain above a specified reference value.

When: Thursday, October 29, 2026 — 2:50 p.m. to 3:50 p.m.
Where: LeConte 224

Speaker: Dr. Ansu Chatterjee, Department of Mathematics and Statistics,  University of Maryland, Baltimore County

Abstract: Understanding the relationship between structural and functional brain connectivity remains a fundamental challenge in network neuroscience. While structural connectivity constrains neural interactions, functional connectivity emerges from complex, nonlinear dynamics that are not fully explained by anatomical wiring alone. Existing graph-based and deep learning approaches often focus on pairwise relationships, thereby overlooking higher-order topological organization in brain networks. We propose TopoGCN, a topology-informed graph convolutional network that integrates topological data analysis (TDA) into structure-function prediction. Persistent homology is used to extract multiscale topological features from structural connectomes, which are incorporated as node-level inputs. Our findings show that incorporating persistent-homology information into graph learning improves the modeling of structure-function relationships in the human brain. By preserving multiscale topological features beyond pairwise connections, TopoGCN provides an improved framework for topology-aware connectome prediction and offers new opportunities for studying brain organization and individual variability.

When: Thursday, November 5, 2026 — 2:50 p.m. to 3:50 p.m.
Where: LeConte 224

Speaker: Dr. Gonzalo Mena, Department of Statistics and Data Science, Carnegie Mellon University

Abstract: The problem of finding a transformation mapping one distribution into another is a relevant mathematical problem with several applications in physics, genomics, etc. When this transformation is assumed to be monotonic, the above problem corresponds to finding the so-called optimal transport map, for which a rich mathematical regularity theory is available and for which a recent non-parametric estimation theory has been established. These statistical results indicate that plug-in estimators of such maps converge faster than expected for Kernel density estimators, a consequence of the extra degree of smoothness of the optimal map compared to the original densities. Moreover, a central limit theorem has been established for such estimators under suitable bandwidth selection, enabling uncertainty quantification. The main drawback is that their computation is typically intractable as it relies on solving an optimal transport problem in the continuum, for which we can only obtain approximated solutions.

To deal with these issues, we propose rectified transport as an alternative to optimal transport. The rectified map (Liu et al., 2022) is a relaxation of optimal transport that recovers optimal transport if sufficient constraints are added. Unlike optimal transport, the rectified map is computed pointwise by solving an ordinary differential equation with a velocity field given by a conditional expectation. Moreover, the computation of a plug-in estimator for the rectified flow amounts to solving a sequence of non-parametric regression problems. Rectified maps are typically used in diffusion models, although little is still known about their regularity and sample properties.

We establish an elementary regularity theory, showing that the population rectified map also has more regularity than the underlying densities. Based on this theory, we derive a bias and variance analysis for this estimator, as well as a central limit theorem. Our results indicate that this estimator benefits from the enhanced regularity of the transport map. However, the benefits are more modest compared to optimal transport, presumably because the rectified map is typically less structured than the optimal transport.

When: Thursday, November 12, 2026 — 2:50 p.m. to 3:50 p.m.
Where: LeConte 224

Speaker: Dr. Matthias Katzfuss, Department of Statistics, University of Wisconsin-Madison

Abstract: We present a framework for the scalable modeling of spatial fields, tracing a progression from Gaussian process approximations using parametric Vecchia models to nonparametric modeling of non-Gaussian fields via autoregressive GPs. This evolution culminates in highly flexible, deep generative models for spatial fields based on geometry-aware autoregressive transformers. All of these approaches leverage the screening effect inherent in geospatial fields to achieve scalability by reducing the size of each conditioning set. In the Gaussian Vecchia case, this corresponds to the assumption of a sparse inverse Cholesky factor of the covariance matrix. For non-Gaussian models, it involves a sparse triangular transport map that facilitates transitions between the distribution of interest and a reference distribution. We demonstrate the efficacy and efficiency of these models through a series of numerical examples.