The focus is on the statistical analysis of non-Euclidean data, where random objects denote random variables and data samples that take values in general metric spaces. Data of such general type are increasingly prevalent in statistics and machine learning. The workshop covers theory, methodology and applications, with emphasis on applications to brain connectomics. Methodology for metric statistics is aimed at samples of networks, distributions, compositional data, covariance matrices and surfaces, tree data and data on manifolds, as well as complex Euclidean data such as high-dimensional and functional data that can be viewed as elements of a metric space.
Topics of interest include quantification of means, barycenters and variance and conditional Fréchet means and their implementation by regression models such as Fréchet regression and its variants. The workshop will also cover extensions of metric statistics and the analysis of random objects towards metric versions of principal component analysis, uncertainty quantification, deep learning, causal inference and longitudinal and time series methodology.
Lightning Talks and Poster Session
This workshop will include lightning talks and a poster session for early career researchers (including graduate students). If accepted, you will be asked to do both. In order to propose a lightning session talk and a poster, you must first register for the workshop, and then submit a proposal using the form that will become available on this page after you register. The registration form should not be used to propose a lightning session talk or poster.
The deadline for proposing is September 30, 2026. If your proposal is accepted, you should plan to attend the event in-person.
In-Person Registration
Seats are limited at the venue, which means that in-person registration may be capped prior to the workshop start date. If capacity is reached, a waitlist will be imposed, which the registration form will reflect. Early registration is strongly encouraged.
All in-person registrants must wait to receive an invitation to attend in-person from IMSI before traveling, which generally begin to be sent out 4-6 weeks in advance.
All registrants (online and in-person) will receive zoom links and are welcome to attend online.
Registration Fee
A non-refundable registration fee will be payable by credit card or debit card for any participants invited to attend this workshop in-person. In-person participants agree to pay the non-refundable fee by the deadline given by IMSI. Failure to pay the fee by the deadline may mean that the invitation to attend in-person is revoked.
Nabarun Deb
University of Chicago Booth School of Business
D
K
Daisuke Kuriso
University of Tokyo, Graduate School of Economics
K
L
Kuang-Yao Lee
Temple University
H
L
Hongzhe Li
University of Pennsylvania
E
L
Eardi Lila
University of Washington
T
N
Tom Nye
University of Newcastle
B
P
Byeong Park
Seoul National University
A
P
Alex Petersen
Brigham Young University
X
S
Xiaofeng Shao
Washington University in St. Louis
B
S
Bharath Sriperumbudur
Pennsylvania State University
Z
Y
Zhigang Yao
National University of Singapore
H
Z
Hang Zhou
University of North Carolina
Y
Z
Yidong Zhou
University of Minnesota, School of Statistics
C
Z
Changbo Zhu
University of Notre Dame
H
Z
Hongtu Zhu
University of North Carolina
V
Z
Vadim Zipunnikov
Johns Hopkins University
Y
W
Yichao Wu
University of Illinois Chicago
R
O
Ryo Okano
University of Tokyo
Z
Z
Zhengwu Zhang
University of North Carolina
Schedule
Monday, November 30, 2026
8:30-8:55 CST
Sign-In & Light Refreshments
8:55-9:00 CST
Welcome & Housekeeping
9:00-9:45 CST
Lifted PCA in the Wasserstein Space
Speaker: Jérémie Bigot (University of Bordeaux)
Principal Component Analysis (PCA) is one of the fundamental tools for statistical dimension reduction. Nevertheless, no canonical extension of PCA to distribution-valued data has emerged within the framework of optimal transport. While Wasserstein barycenters and Wasserstein least-squares regression have recently been shown to arise as canonical liftings in the Wasserstein space of Euclidean averaging and standard linear regression, an analogous lifting principle for PCA has not been considered so far. In this paper, we introduce Lifted PCA, a canonical extension of Euclidean PCA to the Wasserstein space obtained through the framework of lifting functionals in the space of probability measures by convex extension. Starting from the classical variational formulations of PCA on the Stiefel and Grassmann manifolds, we derive two equivalent minimization problems over probability measures supported on these manifolds that correspond to Wasserstein barycenters with generalized costs. We prove that these formulations are linked by an exact analogue of the equivalence between the Stiefel and Grassmann formulations of Euclidean PCA, thereby preserving the original variational structure of PCA on these manifolds after lifting. Building upon the generalized Wasserstein barycenter problem, our analysis further shows that Lifted PCA admits dual formulations and optimality conditions for its characterization, as well as equivalent multimarginal optimal transport and matching-for-teams formulations. We also establish statistical consistency for empirical estimators and develop computational algorithms based on the Wasserstein barycenter formulation over the Stiefel manifold. Our results show that while Euclidean PCA summarizes a collection of vectors by projections onto an optimal low-dimensional linear subspace, Lifted PCA summarizes a collection of probability distributions using an optimal probability measure supported on multiple low-dimensional subspaces. Numerical experiments on distributional datasets illustrate the practical performance of the proposed methodology and its benefits for distributional data analysis.
9:45-9:55 CST
Tech Break
9:55-10:40 CST
Additive Frechet Regression for Random Objects
Speaker: Byeong U. Park (Seoul National University)
Regression analysis for complex data taking values in a general metric space has gained increasing attention in recent years, particularly in the context of Frechet regression with Euclidean predictors $bX in R^{p}$. However, local Frechet regression, while more flexible than global Frechet regression, suffers from the curse of dimensionality for the case of multivariate predictors. To address this issue while maintaining model flexibility, we introduce a framework of additive structured nonparametric regression when responses are situated in a general metric space. Overcoming the challenge of the
lack of vector space structure in general metric spaces where the responses reside requires a novel approach that incorporates the additive structure via projection operators. This leads to a unified framework of additive regression for a wide range of response types, including one-dimensional distributions in Wasserstein space, network data represented by graph Laplacians and spherical data equipped with geodesic distances, among others. We establish pointwise and uniform consistency, including convergence rates, for the proposed additive Frechet regression estimators, using smooth backfitting. The practical utility of this additive model is demonstrated through applications to brain connectivity network analysis using resting-state fMRI data and distributional physical activity data.
10:40-11:15 CST
Coffee Break
11:15-12:00 CST
Which Point on My Brain Matches Which Point on Yours? Registering Cortical Surfaces by Structural Connectivity
Speaker: Zhengwu Zhang (University of North Carolina, Chapel Hill)
"Comparing brain Connectivity across people presupposes that cortical locations correspond, yet stand registration matches folding patterns, which are only loosely related to the brain wiring. This talk presents two methods that align cortical surfaces using structural connectivity itself. Representing connectivity as a continuous atlas-free density over pairs of cortical locations, encore uses a square-root transform and Fisher-Rao geometry to obtain an inverse-consistent penalty-free registration on the product of two spheres. ConSeal keeps that geometry but moves the observed streamline endpoints directly, making the alignment grid-robust and diffeomorphic by construction. On Human Connectome Project data, alignment by wiring improves prediction of cognition. I will end with open statistical problems on registration uncertainty and point processes on manifolds. "
12:00-13:00 CST
Lunch Break
13:00-13:45 CST
TBA
Speaker: Alex Petersen (Brigham Young University)
13:45-13:55 CST
Tech Break
13:55-14:40 CST
Deep additive Fréchet regression
Speaker: Changwon Choi (Seoul National University)
14:40-15:15 CST
Coffee Break
15:15-16:00 CST
Optimal Transport in Statistics: Barycenter Optimization and Regression Analysis
Speaker: Changbo Zhu (University of Notre Dame)
Optimal transport provides a geometric framework for analyzing distribution-valued data. This talk explores two fundamental tasks: averaging probability distributions and modeling their relationships with covariates. I first present scalable algorithms for computing unregularized Wasserstein barycenters of distributions represented on high-resolution grids. I then introduce Kantorovich regression, which uses transport potentials relative to Wasserstein barycenters to model distribution-valued responses with mixed distributional and Euclidean predictors. Numerical examples and applications illustrate the computational advantages and interpretability of these methods.
Tuesday, December 1, 2026
8:30-9:00 CST
Check-In & Light Refreshments
9:00-9:45 CST
Generation-Powered Inference for Distribution-Valued Outcomes
Speaker: Hongzhe Lee (University of Pennsylvania)
Modern generative models increasingly produce distribution-valued outputs, such as predicted cellular responses to genetic perturbations in single-cell genomics. While these models provide valuable auxiliary information, they are inherently imperfect, creating a need for statistical methods that leverage their predictions without relying on their correctness. We propose generation-powered inference (GPI), a general framework for improving inference on distribution-valued parameters using auxiliary generative models. Focusing on Wasserstein barycenters and related distributional functionals, we introduce a function-valued bridge representation that transforms inference in the nonlinear Wasserstein space into estimation of a mean function in a Hilbert space, enabling an augmented estimation framework analogous to prediction-powered inference. We develop a family of GPI estimators with optimal information borrowing, establish consistency, asymptotic normality, and simultaneous confidence bands, and derive valid inference for linear functionals and Wasserstein distances. Simulation studies demonstrate efficiency gains over labeled-data-only methods and robust performance under generative model misspecification. We illustrate the proposed framework using a Perturb-seq study of K562 cells, where synthetic perturbation responses generated by the State foundation model are used to improve inference for pathway-level consensus gene expression distributions associated with perturbations of the 40S ribosome module.
9:45-9:55 CST
Tech Break
9:55-10:40 CST
Fréchet Kriging for Metric Space-Valued Spatial Data
Speaker: Ryo Okano (Hitotsubashi University)
Spatial prediction of non-Euclidean data, such as probability distributions, symmetric positive definite matrices, and networks, arises in many modern applications. Such data can be naturally modeled as elements of general metric spaces, where classical kriging methods relying on vector space structures are not directly applicable. We propose Fréchet kriging, a general framework for spatial prediction of metric space-valued data. The proposed method replaces the weighted average in classical ordinary kriging with a weighted Fréchet mean, where the weights are determined from the spatial dependence structure characterized by a generalized variogram. This approach preserves the intrinsic geometry of the response space without requiring an embedding into a linear space. We establish theoretical properties of the proposed predictor and demonstrate its effectiveness through simulations and real data applications.
10:40-11:15 CST
Coffee Break
11:15-12:00 CST
Variable Selection for Partially Global Frechet Regression
Speaker: Yichao Wu (University of Illinois at Chicago)
Variable selection for Fréchet regression presents a unique challenge, as the model is defined via a weighted Fréchet mean without explicit regression coefficients. In this talk, I will review variable selection techniques for global Fréchet regression and Nadaraya-Watson Fréchet regression first. After that, I will introduce our new variable selection method for partially global Fréchet regression.
12:00-13:00 CST
Lunch Break
13:00-13:45 CST
From limits of the matrix exponential to a geodesic metric space of graphs
Speaker: Tom Nye (Newcastle University)
I will describe the construction and geometry of pipework space, a stratified geodesic metric space of edge-weighted
graphs. A key step is understanding limits of transition matrices of random walks on the vertices of graphs.
13:45-13:55 CST
Tech Break
13:55-14:40 CST
Geometry as Signal: Manifold Fitting and the Structure of Data
Speaker: Zhigang Yao (National University of Singapore)
14:40-15:15 CST
Coffee Break
15:15-16:00 CST
Geometry of covariance shrinkage
Speaker: Karthik Bharath (University Nottingham)
The manifold of symmetric positive-definite matrices, equipped with the affine-invariant metric, is a nonpositively curved symmetric space of noncompact-type. In this talk, its geometry is used to provide a geometric interpretation of the classical James–Stein covariance shrinkage estimator. In particular, the role of horospherical curvature and restricted roots in determining the shrinkage correction is explained. Busemann functions, indexed by directions and flags at infinity, are then used to construct new families of shrinkage estimators, with risk improvements established through an intrinsic Stein identity. An associated SDE formulation reveals how the optimal James–Stein correction cancels the geometric drift of Brownian motion in a Busemann coordinate, connecting covariance shrinkage to diffusions and potential theory on symmetric spaces.
Wednesday, December 2, 2026
8:30-9:00 CST
Check-In & Light Refreshments
9:00-9:45 CST
TBA
Speaker: Hongtu Zhu (University of North Carolina, Chapel Hill)
9:45-9:55 CST
Tech Break
9:55-10:40 CST
Functional linear and single-index regression via RKHS spectral regularization
Speaker: Bharath Sriperumbudur (Pennsylvania State University)
Functional linear and single-index models are core regression methods in functional data analysis and are widely used for performing regression in a wide range of applications when the covariates are random functions coupled with scalar responses. In this work, we consider the estimation of the index vector of a functional single-index model that relates the functional covariate, which is a sample path of a zero-mean Gaussian process, to a scalar-valued response. Based on the infinite-dimensional extension of Gaussian Stein's identity, which allows for estimating the index while being oblivious to the link function, we propose an estimator that involves solving a spectral regularized regression problem in a reproducing kernel Hilbert space (RKHS). We establish convergence rates for this estimator in the regimes of commutativity and non-commutativity of the kernel integral operator and the covariance operator of the Gaussian process. Moreover, when the model is linear, we show these rates to be minimax optimal for both the prediction and estimation errors.
10:40-11:15 CST
Coffee Break
11:15-12:00 CST
Geodesic Synthetic Control Methods
Speaker: Daisuke Kurisu (The University of Tokyo)
We introduce a geodesic synthetic control method for causal inference with panel outcomes that are random objects in a geodesic metric space. The proposed method replaces Euclidean weighted averages by weighted Fréchet means and defines treatment effects as geodesics connecting untreated and treated potential outcomes. We develop a causal model with stochastic perturbations for object-valued untreated outcomes, establish consistency of the estimated weights for the perturbed population target, derive a bound on post-treatment prediction error, and give sufficient conditions for consistent recovery of the untreated counterfactual. For uncertainty quantification, we propose an intrinsic metric conformal calibration procedure that yields prediction regions for untreated counterfactual objects, confidence sets for geodesic treatment effects, confidence intervals for their magnitudes, and global no-effect tests. The method is illustrated through simulation studies for networks and symmetric positive-definite matrices, and through applications to employment composition changes following the 2011 Great East Japan Earthquake and the impact of abortion liberalization on fertility patterns in East Germany.
12:00-13:00 CST
Lunch Break
13:00-13:50 CST
Special Discussion: Artificial Intelligence and the Future of Non-Euclidean Data Analysis
13:50-13:55 CST
Tech Break
13:55-14:55 CST
Lightning Talks
15:00-16:00 CST
Poster Session/Social Hour at IMSI
Thursday, December 3, 2026
8:30-9:00 CST
Check-In & Light Refreshments
9:00-9:45 CST
Online Monitoring for Object-valued Time Series
Speaker: Xiaofeng Shao (Washington University in St. Louis)
9:45-9:55 CST
Tech Break
9:55-10:40 CST
Distributional Data Analysis for Digital Health Technologies Data
Wearables, heart rate sensors, continuous glucose monitors, and smartphones produce dense data streams that are usually reduced to a few summary statistics, losing much of their information. The same data can be viewed in three ways at once: as a time series, as a function of time of day, and as a distribution of values. Each view captures different structure. This talk reviews these approaches for digital health. It covers how to choose among representations of a distribution (density, quantile, hazard, and others) and the geometry each one implies. It surveys regression methods and discusses how to combine the distributional view with the time-series and functional views. Applications include physical activity, heart rate, glucose monitoring, sleep, and mental health. I will close with open problems and connections to metric statistics for random objects.
10:40-11:15 CST
Coffee Break
11:15-12:00 CST
Statistical Inference for Random Objects
Speaker: Hang Zhou (University of North Carolina, Chapel Hill)
Random objects are complex random variables taking values in general metric spaces. Although such data are increasingly common in scientific research, current statistical methodology and theory remain limited. The primary challenge in analyzing such data lies in the absence of vector space operations, such as addition, subtraction, scalar multiplication, and inner products, which are fundamental tools in conventional statistical methodologies. This talk explores object data with distance profiles and their application to conformal prediction and independence testing.
12:00-13:00 CST
Lunch Break
13:00-13:45 CST
Change-Point Detection for Distribution-Valued Sequences
Speaker: Hao Chen (University of California, Davis (UC Davis))
13:45-13:55 CST
Tech Break
13:55-14:40 CST
TBA
Speaker: Eardi Lila (University of Washington)
14:40-15:15 CST
Coffee Break
15:15-16:00 CST
Hypothesis Testing for Metric Point Processes
Speaker: Kuang-Yao Lee (Temple University)
Point process data arise widely in scientific applications, however, existing methods have been developed primarily for Euclidean domains, and there is few solution for replicated point processes observed on general metric spaces. We propose a novel, unified and fully nonparametric framework for hypothesis testing on metric point processes. We represent each process as a random finite counting measure and develop a two-layer kernel embedding architecture: the first layer embeds individual counting measures into a reproducing kernel Hilbert space to test equality of metric intensity measures, while the second layer embeds their probability laws into a nested Hilbert space to test equality of the full distributions. We establish integrally strictly positive definite kernels on broad classes of Polish spaces, and formulate the resulting procedures as maximum mean discrepancy tests. We derive asymptotic null distributions for both independent and paired two-sample designs. Extensive simulations across several metric spaces demonstrate accurate size control and strong power against a range of alternatives. We further illustrate the methodology with a neuroimaging application, where the proposed approach identifies structural differences that are not captured by conventional count-based summaries. (This is joint work with Zhengwu Zhang (University of North Carolina at Chapel Hill) and Lexin Li (University of California at Berkeley)).
Friday, December 4, 2026
8:30-9:00 CST
Check-In & Light Refreshments
9:00-9:45 CST
Measuring Association on Topological Spaces Using Kernels and Geometric Graphs
Speaker: Nabarun Deb (University of Chicago Booth School of Business)
In this work we propose and study a class of simple, nonparametric, yet interpretable measures of association between two random variables X and Y taking values in general topological spaces. These nonparametric measures — defined using the theory of reproducing kernel Hilbert spaces — capture the strength of dependence between X and Y and have the property that they are 0 if and only if the variables are independent and 1 if and only if one variable is a measurable function of the other. Further, these population measures can be consistently estimated using the general framework of geometric graphs which include k-nearest neighbor graphs and minimum spanning trees. Moreover, a sub-class of these estimators are also shown to adapt to the intrinsic dimensionality of the underlying distribution. Some of these empirical measures can also be computed in near linear time. Under the hypothesis of independence between X and Y, these empirical measures (properly normalized) have a standard normal limiting distribution. Thus, these measures can also be readily used to test the hypothesis of mutual independence between X and Y . In fact, as far as we are aware, these are the only procedures that possess all the above desirable properties. This is joint work with Promit Ghosal and Bodhisattva Sen.
9:45-10:15 CST
Coffee Break
10:15-11:00 CST
FréchetFlow: Flow Matching for Non-Euclidean Data
Speaker: Yidong Zhou (University of Minnesota)
Scientific observations such as probability distributions, trees and covariance matrices call for generative models that respect their non-Euclidean geometry. While flow matching has been extended to manifolds by utilizing geodesics and velocities, such manifold structures are not available in a general metric space. We introduce FréchetFlow, a generative modeling framework for metric space-valued data that requires neither linear operations nor a smooth manifold structure. A key feature of the proposed approach is that we utilize well-established Euclidean flows to learn the joint distribution of distances to a set of landmarks, whereupon new objects are reconstructed from generated landmark representations. The reconstruction step utilizes weighted Fréchet means of reference objects, drawing on Fréchet regression. We establish Wasserstein error bounds that connect accuracy of the generated object distributions to the placement of landmarks, flow approximation and reconstruction error and prove consistency under finite second moments. This leads to convergence rates that depend on the metric complexity of the target objects, tail moments and fitting accuracy. Experiments show competitive accuracy with established flow matching methods in Euclidean simulations and lower Wasserstein errors than specifically tailored methods for several scientific datasets. Across simulation settings, we find that generation error decreases as sample size increases.
IMSI is committed to making all of our programs and events inclusive and accessible.
Contact [email protected] to request
disability-related accommodations.
In order to register for this workshop, you must have an IMSI account and be logged in.