Schedule for ICOSDA26

Abstract list (back to top)

• Barry Arnold (University of California, Riverside)
  Title: On bivariate power-function distributions and related models
  Abstract: We introduce a new bivariate distribution for modeling two dimensional data supported on the unit square. The one dimensional marginals of this bivariate distribution follow the univariate Power Function Distributions (PFDs), leading us to name this new distribution the Bivariate Power Function Distribution (BPFD). BPFD is constructed using a common-shock mechanism analogous to the Marshall–Olkin bivariate exponential model. Marshall–Olkin and related extensions are three parameter models in which the common shock affects both components symmetrically through the same parameter, which leads to a positive probability of ties. By contrast, BPFD is a four-parameter model that allows the shared shock to impact the two variables asymmetrically and, for almost all parameter choices, produces no ties. The only exception arises when a specific pair of parameters coincide, in which case ties occur. Some related bivariate models will also be discussed.
• Ruth Pfeiffer (National Cancer Institute)
  Title: Predicting Second Cancer Risk in Cancer Survivors: From Model Development to Clinical Validation
  Abstract: Predicting a patient’s future risk of developing cancer or other serious health outcomes can help guide surveillance, prevention, and treatment decisions. I will discuss a practical framework for developing and evaluating absolute risk prediction models for cancer survivors, an increasingly large and important patient population. Absolute risk models estimate an individual’s probability of developing a second cancer over a fixed time interval, e.g. the next 5 or 10 years, while accounting for competing mortality. I will focus on the development and validation of a breast cancer risk prediction model for female survivors of Hodgkin lymphoma that incorporates detailed information about prior cancer treatments. I will describe how such models can be tested in independent patient populations before they are considered for clinical use. I will discuss two methodological challenges that arose during model development and the solutions we developed to address them. These approaches allowed us to make better use of existing data and to incorporate cancer incidence rates from the general population obtained from registries into the model. Together, these methods improved the accuracy and applicability of risk predictions and may be useful for developing risk models in other settings.
• HK Tony Ng (Bentley University)
  Title: When Is a More Flexible Model Actually Better? Lessons on Model Complexity and Model Selection from Degradation Data Analysis
  Abstract: The development of increasingly flexible statistical distributions and stochastic models remains a major theme in modern statistics. While additional parameters often improve goodness-of-fit, they may also increase estimation uncertainty, reduce interpretability, and complicate scientific conclusions. An important question is whether greater flexibility necessarily leads to better models for real-world applications. Using degradation modeling as a case study, this talk examines how model complexity should be allocated and evaluated. Recent developments in Wiener-process-based degradation models show that flexibility can be introduced through nonlinear time transformations, random effects, and increasingly sophisticated degradation mechanisms. However, empirical evidence suggests that improved fit alone does not guarantee better reliability prediction or more useful scientific insight. In many situations, model components that are most important for the application, such as the operational-time structure governing degradation, contribute far more to predictive performance than additional layers of distributional complexity. Through examples from degradation data analysis, the presentation highlights the trade-offs among flexibility, complexity, interpretability, and practical utility, and discusses broader implications for the development of new statistical distributions and applied statistical models.
• Nilanjan Chatterjee (Johns Hopkins University)
  Title: Integrating Disparate Studies: Transfer and Federated Learning Across Heterogeneous Model Spaces
  Abstract: In this talk, I will present a body of work from our group on building complex regression models by integrating information across multiple studies with disparate designs and available information. A central feature of the framework is its ability to incorporate studies that provide only summary-level information, such as parameter estimates from a variety of reduced-order models. I will describe applications to generalized linear models, the Cox proportional hazards model, and their high-dimensional extensions, along with recent computational advances that make these methods scalable to large-scale genomic data. Finally, I will illustrate a recent application to genome-wide association studies in admixed individuals, where modeling that accounts for local ancestry can substantially enhance gene mapping but access to the required individual-level data is often limited. This setting highlights the practical value of a framework that can draw on summary-level information alone.
• Thomas Mathew (University of Maryland)
  Title: A Joint Confidence Set for a Ranking Based on Ordered Means
  Abstract: This talk presents the development of a measure of uncertainty associated with an estimated ranking of several populations, following up on the novel ideas proposed in Klein, Wright, and Wieczorek (2020, JRSS-C), to be referred to as KWW, Wright (2024, Census Bureau Research Report) and in Wright (2025, Statistics and Probability Letters). In their work, KWW have developed a measure of uncertainty based on confidence intervals for the population means, and by considering the overlap/nonoverlap of the confidence intervals. Later, Wright (2024) derived a new joint confidence region for a ranking based on confidence intervals for the population mean differences, focusing on the overlap/non-overlap of the intervals with the number zero. The present work explores possible improvements as follows: (i) consider differences among the ordered population means, rather than the population means themselves, (ii) compute lower confidence limits for the ordered population mean differences (i.e., an orthant confidence set), and (iii) tighten the joint confidence set by distributing the confidence level unequally among the differences of the ordered population means, with a smaller difference receiving a smaller confidence level. Because the tightness of the confidence set is determined by the number of lower confidence limits that are equal to zero, the strategy proposed in (iii) results in a tighter confidence set because it produces a larger lower confidence limit for a smaller difference, thereby reducing the chance of a lower limit that is zero. The methodology will be applied for ranking the different states in the country in terms of the mean travel time to work, based on an American Community Survey dataset.
• Susmita Datta (University of Florida)
  Title: Deciphering Disease Mechanisms: Statistical Modeling of Cell–Cell Communication Using Spatial Transcriptomics
  Abstract: Spatially resolved transcriptomics enables the characterization of cellular microenvironments and intercellular communication in disease. However, many existing computational methods rely on ad hoc communication scores without rigorous statistical modeling. We developed Bayesian Tweedie Modeling of Communications (BATCOM), a generalized linear regression framework that infers cell–cell communication from spot-based spatial transcriptomics data while accounting for spatial distance. We also developed TWCOM, a scalable frequentist approach based on generalized additive models and implemented in R. Using single-cell and spatial RNA-sequencing data from cutaneous squamous cell carcinoma, we demonstrate the improved performance of these methods in identifying biologically meaningful cellular interactions. BATCOM and TWCOM provide accessible statistical tools for studying disease mechanisms and identifying potential therapeutic targets.
• Narayanaswamy Balakrishnan (McMaster University)
  Title: Discrete-time Signatures
  Abstract: Signatures, originally developed by Samaniego (1985), are well-known tools for the study of coherent systems as they provide convenient mixture representations (in terms of distributions of order statistics) for the distribution of system lifetime. They are, however, based on the assumption of iid continuous lifetimes. In this talk, I will introduce the notion of signatures for systems with discrete lifetimes using the theory of discrete order statistics. I will then discuss some interesting properties of them. Next, I will introduce the notions of discrete-time joint signatures and dynamic signatures and establish some of their properties and applications. I will present several numerical examples to illustrate all the theoretical results discussed. Finally, I will close the talk by making some concluding remarks.
• George Yanev (University of Texas Rio Grande Valley)
  Title: Empirical Bayes Estimation for Mixture of Bore-Tanner Distributions
  Abstract: This talk concerns empirical Bayes estimation of the reproduction parameter for mixtures of Borel--Tanner distributions. As particular examples, we consider the generalized Poisson distribution and compound Bartlett distributions. The main result establishes asymptotic optimality of the proposed estimators and derives their rates of convergence. Monotonized versions are constructed using the method of van Houwelingen and isotonic regression.
• Fatemeh Salboukh (University of Massachusetts at Dartmouth)
  Title: A Transformer-Based Framework for System Resilience Prediction
  Abstract: Resilience engineering plays a critical role in ensuring the sustained performance of complex systems under disruptions, with predictive modeling serving as a cornerstone for proactive decision-making. Accurate long-term prediction of resilience metrics, such as recovery duration or performance degradation, remains a persistent challenge due to the dynamic behavior of systems and uncertainty in future influencing factors. Traditional predictive models often assume full knowledge of future covariates, which limits their reliability and practical applicability, especially in real-world forecasting scenarios. In this study, we address this gap by proposing a forecasting framework that improves prediction accuracy while accounting for the unknown nature of future covariates. Using a real-world dataset spanning ten years of crude oil prices and six historical covariates, we implement a disjoint sliding-window approach with TimeGPT, a state-of-the-art time series foundation model. Our framework exclusively utilizes historical covariates, ensuring that predictions are made under realistic constraints without access to future inputs. The results demonstrate stable and competitive predictive performance across various non-overlapping forecasting window configurations, with the best results achieved using the shortest window of 10 time points and a balanced 5-5 split between training and testing. In this setting, the model reached a mean absolute error of 1.2 and an R^2 value of up to 0.99, highlighting its robustness in long-term prediction under partial information.
• Kun Gou
  Title: Statistical Inference and Nonlinear Parameter Estimation for Quantifying Mechanical Damage and Compliance Changes in Human Umbilical Arteries
  Abstract: This presentation highlights the central role of statistical methods in characterizing the mechanical alterations of human umbilical arteries (HUAs) during vascular graft preparation. Using pressure-diameter inflation data from five paired samples across untreated HUA, collagenized HUA, and decellularized HUA configurations, we employ a hyperelastic damage model with two damage parameters (for isotropic matrix and collagen fiber components) plus a swelling parameter. Parameter estimation is performed via constrained nonlinear optimization based on pressure match and axial force match between lab and computational data. The parameter estimation produces high goodness-of-fit. Paired Student’s t-tests and repeated-measures ANOVA demonstrate that stored strain energy W increases significantly after abluminal removal but shows only a mild, non-significant decrease post-decellularization, while compliance C exhibits overall group differences by ANOVA without pairwise significance. Mean and standard deviation summaries, and stress-strain curves further demonstrate that W outperforms conventional compliance C as a sensitive statistical index of vessel adaptability. These statistical applications provide a robust framework for validating tissue integrity to inform graft-host mechanical matching and guide future larger-cohort studies in tissue-engineered vascular scaffolds.
• Tomoaki Imoto (University of Shizuoka, Japan)
  Title: New toroidal distribution and its application to the hidden Markov model
  Abstract: Toroidal data, which consist of two circular observations, often appear in various fields. For example, pass and shot angles in sports, directions of two animal movements in ecology, torsional angles of amino acid residues in protein structure in biology, and so on. Sometimes, toroidal data is analyzed as time series data, e.g., time series of wind directions at two points in meteorology. In this talk, a new toroidal distribution is proposed in which the marginal and conditional distributions belong to the same circular distribution. The probability density function is expressed in a tractable form, and maximum likelihood estimation can be easily carried out using the EM algorithm. Using the distribution, a hidden Markov model is constructed for analyzing toroidal time series data, and a fitting example for wind direction data is shown.
• Souparno Ghosh
  Title: On inferential procedures for machine learning models
  Abstract: Machine/Deep learning models have become indispensable tools across scientific, industrial, and policy domains. Yet, in most applications, these models are used to generate point estimates. This talk offers protocols for inferential procedures that bridge the gap between algorithmic prediction and classical statistical inference. We examine frameworks for quantifying uncertainty in model outputs via constructing prediction intervals. Through theoretical exposition, simulation studies, and real data applications, we demonstrate how these procedures enhance transparency of machine/deep learning models.
• Mozhdeh Forghaniarani
  Title: Multivariate Scale Mixtures of Normal Distributions for Modeling Stock Returns with Machine Learning Benchmarks
  Abstract: In this study, we utilize infinite multivariate scale mixtures of Normal distributions to model the density of the continuously compounded returns (CCR) of five major American stocks. The approach is based on a finite discretized version of the density, estimating the multivariate parameters from univariate components using UNMIX, a newly developed program for fitting univariate infinite scale mixtures. The performance of the fitted density is evaluated against both classical Bayesian estimation and modern machine learning approaches for density estimation, such as Normalizing Flows and Variational Autoencoders (VAEs). Machine learning approaches in error calculations will be explored as well. Using weekly closing stock prices for major equities (including Google, Amazon, ExxonMobil, Apple, and Fannie Mae), we analyze the models' abilities to capture heavy tails and asymmetric behavior. Additionally, we evaluate how incorporating machine learning benchmarks clarifies the distinct effects of correlation on multivariate fits.
• Zheng Wei (Texas A&M University - Corpus Christi)
  Title: Shape-aware deep learning for models of production with panel data
  Abstract: Stochastic Frontier Modeling was first introduced by Aigner, Lovell, and Schmidt (1977) and Meeusen and van den Brocket (1977) as a parametric model to estimate production efficiency while accounting for statistical noise. Further developments in the field have resulted in a multitude of unbiased estimators which have applications when certain conditions are met. However, classical SFM requires the specification of a production function based on knowledge about the relationship between input and output variables. Furthermore, classical SFM estimators are also limited in that they require the data to be cross-sectional in nature. However, Battese-Coelli (1992) presented an estimator which is appropriate for panel data and showed its effectiveness through an empirical study using rice production data from paddy farmers in India. In this talk, we will review panel data SFMs in literature and their estimation algorithms for efficiency analysis with panel data. Further, we will discuss and propose a new Neural Network based SFM for efficiency analysis with panel data, and demonstrate its capability of efficiency analysis for both the Cobb-Douglas and Translog production functions through a simulation study, followed by an empirical analysis using rice production panel data from the Philippines. Finally, in order to further display the neural network based model's flexibility, the Ind-31 dataset utilized by Malikov (2024) will be examined.
• Jongwook Kim (Ball State University)
  Title: Statistical Modeling using Intrinsic Random Functions for Non-Homogeneous Spatio-Temporal Random Processes on the Sphere
  Abstract: Analyzing global spatio-temporal data requires statistical models that accommodate non-Euclidean geometry and non-stationary behaviors. Classical approaches rely on spatial homogeneity and temporal stationarity, assumptions frequently violated in practice. Furthermore, Euclidean techniques fail to account for the unique geometric properties of a spherical domain.
• Jeonghwa Lee (University of North Carolina Wilmington)
  Title: Fractional binomial regression model
  Abstract: Recently, the fractional binomial distribution (FB) was developed by summing a sequence of dependent Bernoulli variables, called a generalized Bernoulli process (GBP). The fractional binomial distribution was further extended as a generalized linear model, called the fractional binomial regression model. In this talk, we first introduce the GBP and FB distribution, and examine their properties, theoretically and empirically, using simulations. We then introduce the FB regression model, investigate the performance of the estimators, and provide applications to count data with excess zeros.
• Yifan Hsu (Texas A&M University-Kingsville)
  Room: Maryland Room
  Title: A Beta Distribution Framework for Detecting Latent Bias in AI Models
  Abstract: Pre-trained models such as CLIP achieve strong performance in representation learning, but their latent embeddings may contain hidden biases that reduce downstream accuracy. In this talk, I present a statistical framework using propensity scores to detect and mitigate latent bias in CLIP representations. We show that samples with low propensity scores often lie near decision boundaries and are more likely to be misclassified. Removing these samples improves classification performance and highlights the importance of data quality. I further investigate the distributional behavior of propensity scores across classes, where many scores concentrate near one with varying skewness. To better characterize these patterns, I discuss modeling propensity scores using Beta distributions and related simulations to study tail behavior, outlier detection, and misclassification risk. These simulations provide insight into how latent bias propagates through representation spaces. Finally, I discuss applications of this framework to modern AI systems, including deep learning and large language models.
• Ayaka Yagi (Hosei University, Japan)
  Title: T-squared-type Test Statistic for Testing the Adequacy in Growth Curve Model with Two-step Monotone Missing Data
  Abstract: The test for adequacy in growth curve model when the data set has a two-step monotone pattern of missing observations is considered. For the complete data case, there exist studies that use a Hotelling-type test statistic. In addition, T-squared-type tests for mean vectors with monotone missing data have been developed and investigated. We extend these results to the problem of testing the adequacy in growth curve model for data with missing values. Specifically, we propose a T-squared-type test statistic for the growth curve model with two-step monotone missing data. Since it is difficult to derive the exact null distribution of the proposed test statistic, we provide an approximate upper percentile for its null distribution. Finally, we numerically evaluate the accuracy of this approximation using Monte Carlo simulation. This is joint work with Takashi Seo (Tokyo University of Science).
• Hideki Nagatsuka (Chuo University, Japan)
  Title: Statistical Inference for Levy Processes Characterized by Non-Convolution-Closed Infinitely Divisible Distributions
  Abstract: Levy processes characterized by non-convolution-closed infinitely divisible distributions have received relatively little attention, mainly because they are mathematically more difficult to handle than classical convolution-closed Levy processes. In this talk, we present a unified framework for statistical inference for such Levy processes. The proposed methodology is applicable to a broad class of infinitely divisible distributions, including the generalized hyperbolic and generalized inverse Gaussian distributions. Numerical studies demonstrate the practical applicability of the proposed methodology. Potential applications to degradation modeling and reliability analysis will also be discussed.
• Hamdan Hasan (James Madison University )
  Title: Using Machine Learning as a Mapping Tool of Debris Flow Tracks on High Resolution Digital Elevation Models
  Abstract: The objective of this research is to test an automated mapping technique to map debris flow tracks using machine learning models. The models used five measures of surface roughness as independent variables calculated with ArcGIS on high resolution DEMs from Virginia’s Blue Ridge mountains. The models were trained on two first order stream groups, one consisting of known debris flow tracks and another group unaffected by debris flows. The logistic regression results look promising, different other techniques are used and compared. Future work will be focused on extending the results on identifying sink holes and wider range of landslides.
• Olusegun Michael Otunuga (Augusta University)
  Title: From Outbreak to Endemicity or Control: Tracking First Passage Time in Infectious Diseases
  Abstract: This study models how epidemics shift from outbreak to endemic control using the stochastic effective reproduction number R(t). It derives transition and stationary probability density functions for R(t)in a stochastic epidemic framework. These results are used to characterize first passage times (FPTs) for R(t)crossing a threshold Z. The FPT distribution quantifies when infection is likely to dominate or be controlled under random fluctuations. For a fixed Z, the crossing time indicates the first signal of sustained endemic behavior. The framework is extended to a moving threshold Z(t), capturing changing policy or epidemiological conditions. By combining epidemic dynamics with stochastic first-passage theory, the results offer new probabilistic tools for anticipating epidemic resurgence, assessing intervention durability, and designing adaptive public health strategies. Applied to U.S. rhinovirus data, lowering transmission reduces fluctuations and speeds entry to a non-epidemic regime, with modal FPT preceding the positivity peak and mean FPT aligning with sustained decline/plateau.
• Amos Natido (University of California at Santa Barbara)
  Title: The multiple-scaled generalized asymmetric Laplace distribution: Properties and parameter estimation
  Abstract: The Generalized Asymmetric Laplace (GAL) distribution has emerged as a useful model for multivariate data exhibiting skewness and heavy tails. However, the classical formulation relies on a single mixing variable, which may limit its flexibility in capturing heterogeneous tail behavior and asymmetry across different directions of variation. In this talk, we introduce the Multiple-Scaled Generalized Asymmetric Laplace (MSGAL) distribution, a flexible extension obtained by allowing distinct latent mixing variables to act along the principal directions of the covariance structure. We discuss key properties of the proposed model and develop an EM-based framework for parameter estimation. Potential applications to modeling complex multivariate data are also considered. This is joint work with T.J. Kozubowski.
• Aye Aye Maung (University of Louisville)
  Title: Node Splitting SVMs for Survival Trees Based on an L2-Regularized Dipole Splitting Criteria
  Abstract: This work introduces a novel, node-splitting support vector machine (SVM) for creating survival trees, capable of nonlinearly partitioning survival data with continuous, right-censored outcomes. Our method improves on an existing non-parametric approach, which uses oblique splits to induce survival regression trees. In this prior work, oblique splits were created by minimizing a piecewise linear objective function, called a dipole splitting criterion, constructed from pairs of covariates and their associated survival information. We extend this method by enabling splits from a general class of nonlinear surfaces. We achieve this by ridge regularizing the dipole-splitting criterion to enable application of kernel methods analogous to classical SVMs. Using various kernels, we induce both linear and nonlinear survival trees on real and simulated data sets. Trees created by nonlinear splits, using polynomial and Gaussian kernels, show similar predictive power while often being smaller than trees created by univariate and oblique splits.
• Gozde Sert (Texas A&M University)
  Title: Bayesian semiparametric causal inference: Targeted doubly robust estimation of treatment effects
  Abstract: We propose a semiparametric Bayesian methodology for estimating the average treatment effect (ATE) within the potential outcomes framework using observational data with high-dimensional nuisance parameters. Our method introduces a Bayesian debiasing procedure that corrects for bias arising from nuisance estimation and employs a targeted modeling strategy based on summary statistics rather than the full data. These summary statistics are identified in a debiased manner, enabling the estimation of nuisance bias via weighted observables and facilitating hierarchical learning of the ATE. By combining debiasing with sample splitting, our approach separates nuisance estimation from inference on the target parameter, reducing sensitivity to nuisance model specification. We establish that, under mild conditions, the marginal posterior for the ATE satisfies a Bernstein-von Mises theorem when both nuisance models are correctly specified and remains consistent and robust when only one is correct, achieving Bayesian double robustness. This ensures asymptotic efficiency and frequentist validity. Extensive simulations confirm the theoretical results, demonstrating accurate point estimation and credible intervals with nominal coverage, even in high-dimensional settings. The proposed framework can also be extended to other causal estimands, and its key principles offer a general foundation for advancing Bayesian semiparametric inference more broadly.
• Hapuhinna Nelum Kasturiratna Dhanuja (Northern Kentucky University)
  Title: EDF goodness-of-fit tests for Multivariate Normal Distribution
  Abstract: A novel transformation that converts a random sample of size n drawn from a k-variate normal distribution with unknown mean vector and covariance matrix into a sample of size nk − k(k + 3)/2 from a uniform distribution on the interval (0, 1) will be presented. This transformation is then employed to construct an EDF (empirical distribution function) goodness-of-fit test for multivariate normality. Finally, the Monte Carlo method is used to evaluate the power of the proposed EDF goodness-of-fit test under various alternative distributions, demonstrating its effectiveness compared to existing methods.
• John Nolan (American University)
  Title: Generalized logistic extreme value distributions
  Abstract: Multivariate Fréchet laws are a class of extreme value distributions that exhibit heavy tails and directional dependence controlled by an angular measure. Multivariate generalized logistic laws are a recently described sub-class that are dense in a certain sense. It is shown that these laws are related to positive multivariate sum stable laws, which gives a way to simulate from these laws. The corresponding angular measure density is described, and expressions for the density of the distribution are given.
• Munni Begum (Ball State University)
  Title: Random Survival Forest for Predicting Survival Outcomes of Breast Cancer Patients in the Presence of Large Genomic Information
  Abstract: Cancer remains a leading cause of mortality worldwide, motivating research into prognostic modeling. We evaluate survival prediction for breast cancer patients using high-dimensional genomic data, comparing a machine learning approach, Random Survival Forests (RSF), with the traditional Cox Proportional Hazards (Cox PH) model, emphasizing dimensionality reduction. Non-Negative Matrix Factorization (NMF) is applied as a preprocessing step to reduce genomic dimensionality while preserving biologically meaningful structure. Analyses were conducted using two datasets: a full dataset of 177 significantly altered genes and a corresponding NMF-reduced dataset. Model performance was assessed using time-dependent Area Under the Curve (AUC). Both RSF and Cox PH achieve strong, stable predictive accuracy across datasets. RSF consistently shows slightly higher AUC values at all time points, though differences are modest. These findings demonstrate that NMF effectively reduces dimensionality without compromising predictive performance, suggesting that both machine learning and classical statistical approaches remain competitive with appropriate feature reduction techniques.
• Shimeng Huang (Purdue University)
  Title: A copula model for marked point process with a terminal event: An application in dynamic prediction of insurance claims
  Abstract: Accurate prediction of an insurer’s outstanding liabilities is crucial for maintaining the financial health of the insurance sector. We aim to develop a statistical model for insurers to dynamically forecast unpaid losses by leveraging the granular transaction data on individual claims. The liability cash flow from a single insurance claim is determined by an event process that describes the recurrences of payments, a payment process that generates a sequence of payment amounts, and a settlement process that terminates both the event and payment processes. More importantly, the three components are dependent on one another, which enables the dynamic prediction of an insurer’s outstanding liability. We introduce a copula-based point process framework to model the recurrent events of payment transactions from an insurance claim, where the longitudinal payment amounts and the time-to-settlement outcome are formulated as the marks and the terminal event of the counting process, respectively. The dependencies among the three components are characterized using the method of pair copula constructions. We further develop a stagewise strategy for parameter estimation and illustrate its desirable properties with numerical experiments. In the application we consider a portfolio of property insurance claims for building and contents coverage obtained from a commercial property insurance provider, where we find intriguing dependence patterns among the three components. The superior dynamic prediction performance of the proposed joint model enhances the insurer’s decision-making in claims reserving and risk financing operations.
• Nadeesha Jayaweera (University of Akron)
  Title: Adaptive Bayesian Spatio-Temporal Disease Prediction Using Likelihood-Based Weighted Smoothing
  Abstract: Accurate prediction of infectious disease spread is challenging due to evolving transmission patterns and regional variability in population characteristics. This study proposes an adaptive Bayesian spatio-temporal prediction framework that integrates likelihood-based weighted smoothing within the integrated nested Laplace approximation (INLA) methodology. The proposed approach improves short-term disease prediction by assigning greater importance to recent observations through calibrated likelihood weighting, allowing the model to adapt to non-stationary outbreak dynamics while incorporating population density information. Daily COVID-19 case count data from Massachusetts counties are used to evaluate predictive performance. Results demonstrate that the weighted Bayesian framework achieves improved forecasting accuracy compared with traditional smoothing and prediction methods. The proposed methodology provides a computationally efficient and flexible framework for real-time disease prediction, offering valuable support for public health surveillance, intervention strategies, and healthcare resource planning.
• Anna Panorska (University of Nevada, Reno)
  Title: From atmospheric rivers to flood risk: A multivariate model for extreme precipitation
  Abstract: Water resources and flood risk in the western United States are commonly driven by large scale precipitation events lasting several consecutive days. Some of the largest precipitation events are driven by the Atmospheric Rivers, the “rivers in the sky”. We propose a multivariate model for the extreme precipitation events describing their duration, intensity, and maximum. We also link it to the atmospheric flow patterns, like those producing the Atmospheric Rivers. As the most important questions related to the impact of extreme precipitation are about the risk or probability of these extreme events, our ultimate goal is to employ the statistical model along with the meteorological and statistical methods leading to estimation of these probabilities. This is joint work with Alexander Weyant.
• Qi Zheng (University of Louisville)
  Title: Censored Quantile Regression for Streaming Data
  Abstract: Censored quantile regression (CQR) provides a robust alternative to proportional hazards models in survival analysis, yet encounters practical challenges in streaming data scenarios due to the need for frequent estimation updates and constraints on storage capacity. In this paper, we develop a martingale-based streaming CQR framework that enables online quantile inference for survival data. By integrating kernel smoothing with a one-step linearization strategy, we derive a closed-form recursive estimator that resolves optimization challenges induced by non-smooth CQR objectives. Our online incremental update algorithm achieves both computational efficiency and low storage requirements through sufficient statistics, and the framework naturally extends to distributed computing environments. Theoretically, we derive asymptotic Bahadur representations for the proposed estimator, demonstrating its statistical efficiency comparable to full-data analysis under mild conditions. Empirically, simulations and real-world data analyses confirm the framework’s computational scalability and its ability to maintain inferential accuracy across quantiles.
• Michael Levine (Purdue University)
  Title: Multivariate truncated normal distributions
  Abstract: Our object of interest is an exponential family of truncated multivariate normal distributions with one-sided truncation for some or all coordinates. This distribution is routinely used as a prior distribution on model parameters in a Bayesian shape-constrained regression. It is also commonly used for inference with censored biomedical data and in educational studies. Despite this wide variety of applications, systematical investigation of corresponding exponential families has attracted remarkably little attention. We begin with the observation that, if all components are one-sided truncated, then this family is not full. The family of truncated multivariate normal distributions is extended to a full family, and the extended family is investigated in detail. We identify the canonical parameter space of the extended family and establish that the family is not regular and not even steep. We also consider maximum likelihood estimation for the location vector parameter and the positive definite (symmetric) matrix dispersion parameter of a truncated non-singular multivariate normal distribution. It is shown that if the sample size is sufficiently large then, almost surely, the maximizer of the likelihood function is unique, provided that it exists. It is also shown that each solution to the score equations for the location and dispersion parameters satisfies the method-of-moments equations. Finally, it is observed that similar results arise in the case of an arbitrary number of truncated components.
• Tatjana Miljkovic (Miami University)
  Title: A Novel Framework for Aggregate Loss Modeling in Insurance
  Abstract: The aggregate loss distribution is central to many actuarial applications, offering insights into the risk profile of insurance portfolios. Traditionally, it has been estimated using approximation, recursive, or convolution methods that require discretizing the loss distribution. We propose a direct estimation approach using a mixture model that groups policyholders by claim count, with mixture weights based on a truncated Poisson distribution to capture claim frequency. Closed-form expressions for key risk measures and the net stop-loss premium are derived for Gamma and Lognormal severity distributions. The model also extends existing approximation methods such as Normal, Normal-Power II, Lognormal, Gamma, and Inverse Gaussian by providing explicit formulas for risk measures and premiums. Performance is evaluated using French auto loss data and simulations, showing that the proposed method outperforms traditional techniques, supporting a move away from approximation-based approaches.
• Tomasz J. Kozubowski (University of Nevada, Reno)
  Title: Flexible skew multivariate models via coordinatewise Gaussian mixtures
  Abstract: We propose an extension of the normal mean--variance mixture model that allows for coordinatewise mean--variance mixing within a multivariate Gaussian framework. A particular case based on gamma mixing yields a new class of multivariate distributions with marginal generalized asymmetric Laplace distributions, extending the classical multivariate generalized asymmetric Laplace (GAL) model. The proposed family offers enhanced flexibility for modeling multivariate data exhibiting heterogeneous skewness and tail behavior across dimensions. We present key distributional properties of the new class and discuss parameter estimation via the Expectation--Maximization (EM) algorithm. An application to real data illustrates the practical usefulness of the proposed approach. This is joint work with Amos Natido.
• Shih-Ting Huang (University of Louisville)
  Title: Sample-specific learning of lymphovascular invasion with heterogeneous spatial patterns
  Abstract: Lymphovascular invasion (LVI) represents one of the earliest stages of metastasis. This article is motivated by using the multiplex immunofluorescence imaging data for identification of LVI that can improve diagnosis of early breast cancer prior to metastasis. One unique aspect for this type of data is that individual-level imaging are taken at various locations depending on where the biopsy was located on the breast. Thus, there exists substantial spatial heterogeneity between the images. We present a novel sample-specific learning framework for L1-penalized logistic regression to aid accurate LVI classification in real time. We derive finite sample guarantees for our proposed estimator in this article and present a computationally efficient algorithm for translation. The finite sample performance for the proposed method is assessed via intensive simulation studies. Using images generated from the Stanford cohort, we rigorously assess the LVI classification performance with both internal and external validation and demonstrate high concordance with pathologist labeling.
• Drew Lazar (Ball State University)
  Title: Applications of Dipole-Splitting SVMs to Spatial Survival Analysis
  Abstract: We develop a nonparametric, tree-based framework for spatial survival analysis built on kernel dipole-splitting support vector machines, for settings where survival times depend on both clinical covariates and geographic location. The framework rests on two contributions. First, a three-stage residual procedure in which a linear-kernel survival tree absorbs clinical effects, cumulative hazard residuals carry the censored survival signal forward without invoking proportional hazards, and a Gaussian or polynomial kernel tree recovers residual spatial risk. Second, a kernel orthogonalization that generalizes the Frisch-Waugh-Lovell theorem to reproducing kernel Hilbert spaces, yielding a residualized spatial kernel for a single tree. The choice of kernel places the method on a smooth-to-discrete spectrum that classical geostatistical frailty models (Henderson et al., 2002) cannot span. We illustrate on the LeukSurv acute myeloid leukemia dataset and on simulations with sharp spatial boundaries, where our trees recover discrete risk regions and achieve higher concordance than correlation-based frailty models.
• Chad Schafer (Carnegie Mellon University)
  Title: Motivating and Fitting Models with Mixtures of Conditionally Independent Distributions
  Abstract: The mixture of conditionally independent distributions (MICI) model is a model for multivariate distributions in which the coordinates of a random vector are independent given their membership in one of $k$ groups. When the dimension of the vector is three or larger, the model is nonparametrically identifiable. This talk will describe and motivate a method that was developed within the particle physics community for the purpose of separating signal and background, and how this yields an additional approach to fitting MICI models. This is joint work with Mikael Kuusela, Alex Shen, and Larry Wasserman.
• Saman Muthukumarana (University of Manitoba)
  Title: Bayesian Nonparametric Learning and Clustering for High-Dimensional Imaging Data
  Abstract: We develop two Bayesian nonparametric frameworks for image-based data using transfer-learned features from MobileNetV2 or VGG16. The first model, a Chinese Restaurant Process based Bayesian Deep Clustering framework, extracts high-dimensional image representations and applies a CRP-based Dirichlet Process mixture model to automatically discover disease groups without requiring a predefined number of classes. This approach provides probabilistic cluster assignments, uncertainty quantification, and the ability to identify rare or emerging feature categories. The second model extends this framework through a Dirichlet Process Variational Autoencoder, where feature learning and clustering are performed jointly within a unified Bayesian architecture. By integrating a CRP prior into the latent space, the model simultaneously learns informative image representations, infers the number of clusters, and reconstructs images through a deep generative process. Together, these frameworks offer scalable, uncertainty-aware solutions for pattern discovery and cluster identification in imaging applications.
• Hon Yiu (Henry) So (Oakland University)
  Title: Sequential Estimation for One-shot Device Stockpiles
  Abstract: Stockpile reliability assessment for one-shot devices is challenging due to destructive testing and limited data, where each unit yields only a binary outcome at inspection. In this talk, we propose a sequential estimation framework for assessing stockpile reliability. Units are tested one at a time, and sampling stops based on a data-driven rule that balances estimation accuracy and testing cost, particularly when failures are rare. The approach enables efficient reliability estimation while reducing unnecessary consumption of stockpile units. Its performance is illustrated using a real-world degradation dataset, demonstrating practical value for reliability certification under severe data constraints.
• Jean-Francois Plante (HEC Montreal)
  Title: Wavelet-based strategy to estimate a distribution function on a distributed system
  Abstract: We propose a wavelet-based strategy for estimating a distribution function when the data are stored in a distributed system. The method combines local wavelet estimates computed on each node with an aggregation scheme that preserves the global distributional structure while limiting communication costs. We study the estimator and illustrate its behavior through simulations and a real-data example.
• Ash Abebe (Auburn University)
  Title: A Two-Stage Bayesian Framework for Joint Modeling of Clustered Proportional Outcomes with Application to Social Determinants of HIV
  Abstract: Community-level survey data often yield multiple outcomes that are proportions. These include disease prevalence, behavioral rates, and attainment indicators among many others. These follow a Binomial distribution at the cluster level. Characterizing the joint distributional relationships among such outcomes is complicated by nonlinearity, zero-inflation, and spatial dependence. We propose a two-stage Bayesian framework that first characterizes empirical association patterns via a Dirichlet process infinite mixture model, then quantifies theory-motivated nonlinear relationships using shape-constrained piecewise spline functions. The first stage clusters communities by their joint profiles without imposing functional constraints, enabling data-driven identification of association shapes. The second stage encodes substantive hypotheses as piecewise cubic spline structures anchored at interpretable boundary conditions, with effect sizes estimated from full posterior distributions. We apply this framework to Demographic and Health Survey data from Ethiopia, Kenya, Rwanda, and Tanzania (2003–2016) to evaluate structural pathways linking educational attainment, HIV-related stigma, and HIV prevalence across rural and urban communities. The method extends naturally to any multivariate setting with bounded, clustered outcomes and offers a principled approach to combining nonparametric exploration with parametric inference.
• Samuel Morrissette (University of Manitoba)
  Title: Variational Bayesian Multidimensional Scaling
  Abstract: Multidimensional Scaling (MDS) is a statistical method often used for visualization or as a preprocessing step for subsequent analysis. MDS estimates a configuration of points in Euclidean space such that the distances between them closely approximate the observed dissimilarities between a set of items. Bayesian Multidimensional Scaling (BMDS) extends classical MDS by accounting for uncertainty in the configuration and measurement error in the dissimilarities. However, fitting BMDS using Markov Chain Monte Carlo is computationally intractable when the embedding dimensionality and/or sample size is large. Therefore, we propose a variational algorithm to accelerate inference under the BMDS model. Using modern techniques such as mini-batching and stochastic gradient descent, our approach substantially reduces computational burden and enables the application of BMDS to large-scale datasets, thus improving on both MDS and BMDS.
• Tung-Lung Wu (Mississippi State University)
  Title: Scan Statistics for Nonhomogeneous Poisson Processes with Extreme-Value Calibration and Application to CNV Detection
  Abstract: We develop a scan statistic method for detecting local clusters in a two-sample nonhomogeneous Poisson process (NHPP) framework, motivated by copy number variation (CNV) analysis in next-generation sequencing data. We approximate the distribution of the scan statistic by a generalized extreme value (GEV) distribution with an extremal index that captures the dependence induced by overlapping windows. The GEV parameters and extremal index are estimated using maximum likelihood and exceedance clustering methods, providing an asymptotic calibration of the test. A permutation procedure is also developed to provide a nonparametric alternative. Simulation studies demonstrate that the proposed methods maintain accurate type-I error and perform well compared with the competing continuous testing method under heterogeneous baseline intensities. An application to sequencing data illustrates the effectiveness of the proposed approach for detecting CNV regions.
• Nick Wintz (Radford University)
  Title: The Conformable Kalman Filter
  Abstract: In this project, we formulate the Kalman-Bucy filter for linear, continuous conformable control systems corrupted by white noise. Here, we use the system first introduced by Khalil, et al. We obtain a state transition matrix via a Peano-Baker expansion that allows us to calculate our error propagation through a Riccati equation. In addition, we show the duality between the conformable Kalman filter and its associated conformable linear quadratic regulator (CLQR) problem is preserved. Finally, we provide numerical simulations for relevant applications.
• Kevin Lee (Western Michigan University)
  Title: Nonparametric Finite Mixture of Ising Graphical Models
  Abstract: Graphical models provide a general methodology for approaching statistical applications in many different fields where thousands of random variables are linked in complex ways. In current literature, high-dimensional data are more assumed to come from one population, whereas in real-world applications, it is more common that the observed data come from multiple populations. Hence, it is crucial to explore the presence of subpopulations within an overall population and estimate the heterogeneous dependencies in the graph structures. In this work, we focus on modeling the binary-valued data where most of the existing work assumes that the relational structure is fixed over time, which might not be always the case. Therefore, this study focuses on learning a nonparametric finite mixture of Ising graphical models to estimate functional parameters and recover functional heterogeneous graphical edge structure. The performance of the proposed method is demonstrated via simulation studies and a real-world application.
• Surani Matharaarachchi (University of Victoria)
  Title: A Dirichlet–Multinomial Framework for Sequential Bayesian Estimation of the F1 Score
  Abstract: The F1 score is widely used to evaluate classification models, especially when the classes are imbalanced. However, traditional evaluations report only a point estimate and therefore do not reflect the uncertainty arising from finite samples, particularly when minority-class observations are scarce. We propose a Bayesian framework based on the Dirichlet–Multinomial model to estimate the posterior distribution of the F1 score through sequential updating. The method provides continuous and interpretable uncertainty quantification as new observations become available, making it suitable for streaming data, online learning, and real-time model monitoring. Its performance is evaluated through simulation studies covering different sample sizes and levels of class imbalance. Applications to benchmark classification datasets further demonstrate its practical value. The proposed framework provides a dynamic assessment of model performance, allowing changes in both the estimated F1 score and its uncertainty to be monitored over time and supporting more reliable model evaluation and decision-making.
• Zhu Wang (University of Tennessee Health Science Center)
  Title: Unified Robust Estimation
  Abstract: Robust estimation is primarily concerned with providing reliable parameter estimates in the presence of outliers. Numerous robust loss functions have been proposed in regression and classification, along with various computing algorithms. In modern penalised generalised linear models (GLMs), however, there is limited research on robust estimation that can provide weights to determine the outlier status of the observations. This article proposes a unified framework based on a large family of loss functions, a composite of concave and convex functions (CC-family). Properties of the CC-family are investigated, and estimation is innovatively conducted via the iteratively reweighted convex optimisation (IRCO), which is a generalisation of the iteratively reweighted least squares in robust linear regression. For robust GLM, the IRCO becomes the iteratively reweighted GLM. The unified framework contains penalised estimation and robust support vector machine (SVM) and is demonstrated with a variety of data applications.
• Golshid Aflaki / Jean-Francois Plante (HEC Montreal)
  Title: Enhancing Sawmills Statistical Process Control System Using Mixture Models
  Abstract: Process control is crucial for detecting and addressing issues early in the production cycle, reducing the risk of costly recalls and rework. In the wood industry, it is important to control the quality of lumber by monitoring its dimensions, especially thickness and width. Often used traditional quality control models usually assume such measures follow a single normal distribution which is not necessarily correct, due to the presence of multiple sources of variability in board production. Instead, Gaussian Mixture models account for the possibility that the data may be generated from multiple underlying distributions rather than a single one. This research aims to design an advanced, data-driven quality control framework tailored to the specific needs of sawmills by minimizing false alerts and ensuring that the system generates only accurate, actionable warnings by adopting Gaussian mixture models.
• H. Frazier Bindele (University of South Alabama)
  Title: Robust Empirical likelihood variable selection for the high dimensional single-index regression model
  Abstract: In this talk, a single-index regression model is considered, from which a robust and efficient inference about the model parameters is proposed. From a local linear approximation of the unknown regression function, such a function is estimated using the generalized signed-rank approach. Next considering the estimated function together with the estimating equation obtained from the generalized sign-rank objective function, a penalized empirical likelihood objective function of the index parameter is defined, from which its asymptotic distribution is established under mild regularity conditions. The performance of the proposed method is demonstrated via extensive Monte Carlo simulation experiments. The obtained simulation results are compared with those obtained from a normal approximation alternative and those obtained based on the least squares and least absolute deviations approaches. Finally, a real data example is given to illustrate the proposed methodology.
• Kevin McGregor (University of Manitoba)
  Title: Revisiting the alpha-folding multivariate normal distribution
  Abstract: We revisit the alpha-folding multivariate normal distribution for compositional data analysis under two inferential frameworks. We express the model as a log-sum-exp representation, which highlights how the two branches arise from a many-to-one folding mechanism and their Jacobian terms, without a finite-mixture interpretation. For frequentist inference, we propose a practical maximum-likelihood method based on fixed-point estimation and profile likelihood, and construct likelihood-ratio confidence intervals for alpha. We additionally propose a Bayesian framework that targets the marginal posterior of alpha using a Metropolis-within-Gibbs sampler that alternates between updates of alpha and the Gaussian parameters, motivated by the strong coupling between them and the potential sensitivity of profile optimization to initialization. Through simulation studies and real-data analysis, we compare the two approaches in terms of estimation accuracy, numerical stability, and uncertainty quantification. We also discuss a variational inference approach to this model.
• Yan Zhuang (Connecticut College)
  Title: Fixed-Accuracy Inference on Weibull Shape Parameters Using Records Data
  Abstract: The Weibull distribution is a widely used model for lifetime data due to its flexibility in capturing various failure behaviors. The shape parameter, in particular, plays a crucial role in governing the tail behavior and reliability characteristics of the Weibull distribution. In many life testing scenarios, such as fatigue and strength testing, only the highest observed values (upper records) drive failure. Analyzing such record data can be sufficient to yield more efficient inference for tail-behavior parameters. In this paper, we focus on statistical inferences for the shape parameters of Weibull distributions using record values. We first construct fixed-accuracy confidence intervals (FACIs) for the shape parameter of a single Weibull distribution. We then investigate the ratio of shape parameters from two independent Weibull distributions. We also discuss the FACIs for the reliability parameter, P(X<Y), which quantifies the probability that one system (Y) outlasts the other (X). Extensive Monte Carlo simulation studies demonstrate the appealing properties of our proposed methods in terms of accuracy and efficiency. The methods are further illustrated through real-world data on carbon fiber strength with different gauge lengths.
• Moh Khalid Hasan (James Madison University)
  Title: Learning-Driven Predictive Beamforming for Dynamic MIMO-UAV Networks
  Abstract: Mobile MIMO-UAV networks operate over highly non-stationary wireless channels due to three-dimensional UAV mobility, Doppler variation, link-geometry changes, and multipath fading. These effects make channel state information prediction and beamforming design challenging, particularly when outdated CSI reduces array gain and spectral efficiency. This work studies a learning-enhanced predictive beamforming approach for autonomous MIMO-UAV communications under mobility-induced channel uncertainty. The proposed design exploits temporal channel correlation through data-driven sequence modeling and incorporates prediction-aware beamforming to improve downlink transmission performance. A dynamic UAV communication setting is considered, where channel evolution is governed by stochastic mobility and Rician propagation characteristics. Simulation results indicate that the learning-assisted mechanism can improve prediction accuracy and provide a higher achievable sum rate.
• Rida Benhaddou (Ohio University)
  Title: Minimax Estimation for Varying Coefficient Model via LaGuerre Series
  Abstract: We delve into the estimation of the functional coefficients and inference for varying coefficient model. Applying Laguerre series, we develop an estimator for the vector of functional coefficients that attains asymptotically optimal convergence rates in the minimax sense. These rates are derived for functional coefficients that belong to Laguerre-Sobolev space. The method is based on approximating the functional coefficients using truncated Laguerre series and choosing empirical Laguerre coefficients that minimize the least squares criterion. In addition, we establish the asymptotic normality of the estimator for the functional coefficients, construct their confidence intervals, and establish point-wise hypothesis tests about their true values. A simulations study is carried out to examine the finite-sample properties of the proposed methodology and to compare the performance to some of the existing methods. A real data set is considered as well, and results based on the proposed methodology are compared to those based on selected existing approaches.
• Tom Cuchta (Marshall University)
  Title: Probability distributions on time scales
  Abstract: Time scales calculus unifies and extends continuous and discrete calculus by defining a derivative and integral on a so-called time scale, which is any closed subset of R. In this talk, we will overview the basic definitions from time scales, discuss various probability distributions defined on time scales, and see how this framework can reveal surprising connections between seemingly unrelated probability distributions.
• Lei Fang (Miami University)
  Title: Multi-ancestry Proteome-Wide Association Study for Alzheimer’s Disease
  Abstract: Proteome-Wide Association Studies (PWAS) have emerged as a powerful framework for identifying protein-trait associations by integrating genetic and proteomic data. However, most existing PWAS analyses are conducted in populations of European ancestry, limiting the generalizability and transferability of findings across diverse populations. In contrast, sample sizes for other ancestries, such as African populations, are often substantially smaller, resulting in reduced statistical power and limiting the discovery of meaningful associations. In this study, we propose a multi-ancestry PWAS method, MA-2SLS, designed to improve statistical power and estimation precision in underrepresented target populations. Based on the two-stage least squares (2SLS) framework, MA-2SLS leverages similarity in cross-ancestry genetic effect sizes to borrow information from an external population with large-scale GWAS summary data. Through simulation studies, we demonstrate that MA-2SLS increased statistical power while maintaining well-controlled Type I error rates. The real data application also highlights the potential of MA-2SLS for identifying additional proteins associated with Alzheimer’s Disease in African population.
• Shuoyang Wang (University of Louisville)
  Title: Comprehensive neural network methods for heteroscedastic partial linear quantile regression in ultra-high dimensions: deep and shallow architectures
  Abstract: We study heteroscedastic partial linear quantile regression in ultra-high dimensions allowing an exponentially growing dimension of the high-dimensional linear covariates. Recent advances show that deep neural networks (DNNs) can estimate the unknown nonlinear component beyond the limits of classical nonparametric methods. However, two fundamental challenges remain: the additional sparsity tuning commonly required by sparse neural networks substantially increases computational cost, and the trade-off among predictive accuracy, variable-selection consistency, and model complexity between deep and shallow architectures is not well understood. To address these challenges, we develop a unified neural framework with complementary theoretical guarantees. When the true nonlinear component lies in a hierarchically compositional function space, we establish theory for fully connected deep network estimators; when it lies in a variation space, we provide parallel theory for fully connected shallow network estimators. Our results establish sharp-rate estimation consistency, asymptotic normality, asymptotically valid inference for the key linear covariates, and consistent variable selection for the sparse linear coefficient vector at the target quantile level. Building on this theoretical characterization, we propose a data-adaptive depth-and-width selection procedure that achieves substantial computational savings while preserving statistical guarantees. Extensive simulations demonstrate state-of-the-art predictive performance, stable recovery of the active set, and marked computational gains of shallow architectures when the underlying structure favors low-order variation. An application to high-dimensional metabolomic data demonstrates the practical advantages of the proposed method in identifying biologically meaningful metabolites while maintaining strong predictive performance
• Derek Young (University of Kentucky)
  Title: Characterizing Subgroups of Count Data with Different Degrees of Dispersion via Finite Mixtures of Mean-Parameterized Conway-Maxwell-Poisson (CMP) Regressions
  Abstract: The Conway–Maxwell–Poisson (CMP) distribution is a widely used generalization of the Poisson distribution and often provides a competitive alternative to other discrete models because of its ability to explicitly accommodate both over- and under-dispersion. The mean-parameterized CMP has attracted increasing attention in the literature, as it allows for direct modeling of the data mean -- particularly the conditional mean in regression settings -- and can be naturally embedded within a generalized linear modeling framework. We propose a mixture of mean-parameterized CMP regression models for count data that may arise from heterogeneous subpopulations with distinct conditional means and varying dispersion levels. Additional flexibility is introduced by allowing covariates to enter the model as concomitant variables governing the component weights. Simulation studies and a real-data application illustrate the strong performance of the proposed approach relative to traditional mixtures of count regression models.
• Philip Yates (DePaul University)
  Title: Robust Local Likelihood Estimation for Non-stationary Flood Frequency Analysis
  Abstract: As changes to the environment, both human and non-human made, continue to occur, the assumption of stationarity in flood frequency analysis is seldom met. A proposed method for estimating the 1% chance flood, i.e. Q_100 of a flood gauge’s annual peak streamflows, using robust local likelihood will be discussed. When a flood series seems to be from a more mixed population of values, often due to extreme snowmelt or tropical storms in a given flood year, robust local likelihood estimation at estimating the 1% chance flood than local likelihood estimation. Discussion will include a case study as well as possible extensions of this work.
• Hossein Moradi Rekabdarkolaee (Bowling Green State University)
  Title: Addressing Imbalance in data Using Generative AI
  Abstract: Bovine pinkeye is a contagious ocular disease that significantly impacts cattle health and agricultural productivity. Conventional diagnostic methods rely on clinical observation, which can be subjective, time-consuming, and impractical for large herds or remote locations. To address these limitations, this study explores the application of deep learning for automated pinkeye detection using Deep Learning. Furthermore, we employed different generative AI algorithms to address the imbalance data. Our findings shows the effectiveness of the proposed approach.
• Yuting Chen (Eastern Kentucky University)
  Title: Empirical best prediction of poverty indicators via nested error regression with high dimensional parameters
  Abstract: The Nested Error Regression Model with High-Dimensional Parameters (NERHDP) is extended to address challenges in small area poverty estimation. A robust and flexible framework is proposed to derive empirical best predictors (EBPs) of small area poverty indicators while accommodating heterogeneity in regression coefficients and sampling variances across areas. To mitigate the computational limitations of the existing algorithm, an efficient estimation procedure is introduced, substantially reducing computation time and enhancing scalability for large datasets. A novel approach for generating area-specific poverty estimates in out-of-sample areas is also developed, improving the reliability of synthetic estimates. Uncertainty is quantified through a parametric bootstrap method specifically tailored to the extended model. Under heterogeneous data-generating scenarios, the proposed method yields lower relative bias and relative root mean squared prediction error than a range of existing approaches. The methodology is further illustrated using data from the 2002 Albania Living Standards Measurement Survey, combined with auxiliary information from the 2001 census, to estimate poverty indicators for 374 municipalities.
• Olcay Arslan (Ankara University)
  Title: Joint Location–Scale Modeling for Non-Gaussian Data Using the Generalized Asymmetric Least Informative Distribution
  Abstract: Classical linear regression models often rely on homoscedasticity and Gaussian error assumptions, which are frequently violated in practice due to heterogeneity, skewness, heavy tails, and outlying observations. Moreover, explanatory variables may affect not only the mean of a response but also its variability, motivating the use of joint location–scale models. This work proposes a robust joint location–scale modeling framework for non-Gaussian data based on the Generalized Asymmetric Least Informative Distribution (GALID). The GALID, which is an asymmetric generalization of the Huber's Least Informative Distribution, accommodates asymmetric and heavy-tailed responses and provides a flexible alternative to conventional Gaussian-based models by jointly modeling the conditional location and scale of the response variable. The relationship between the GALID likelihood and the asymmetric Huber loss provides a basis for M-quantile estimation. Considering this connection, we develop an M-quantile approach for estimating the parameters of the JLSM. Furthermore, we propose a penalized M-quantile estimation procedure to perform simultaneous variable selection in the location and scale submodels. The performance of the proposed method is evaluated through simulation studies under various non-Gaussian settings. Joint work with Yeşim Guney and Yetkin Tuaç.
• Scott Smith (University of the Incarnate Word)
  Title: Comparison of Estimation and Inferential Properties of Competing Models for Likert Scale Variables
  Abstract: We discuss basic properties of three novel models and two models previously discussed in the literature. One of these is the Normal model commonly assumed in scientific research. The five models will be compared in terms of robustness in model selection, error rates in inference, and coverage and margin of error in interval estimation. Some comparisons originate from theoretical results and some from simulation study. Both maximum likelihood and parametric bootstrap estimation are considered in the study. We will conclude with some general comments summarizing the findings.
• Mahsa Ashouri (Miami University)
  Title: Generalized Bayesian Additive Regression Trees for Dynamic Inference on the Restricted Residual Life
  Abstract: Dynamic survival models that seamlessly integrate new information as it becomes available to make updated predictions and uncertainty assessments are highly useful for accurate evaluation of a patient’s current risk and for improving treatment decision making. To improve the effectiveness and robustness of Bayesian dynamic survival models, we propose a model that mainly targets inference for the expected remaining restricted life at a series of landmark times. In our approach, at each landmark time, a new Bayesian additive regression tree (BART) is incorporated into the overall risk model with new predictions at the landmark times determined from a vector autoregressive model that enables the collection of BART models to perform smoothing for the full risk model across time. We adopt a generalized Bayes framework which allows us to generate uncertainty intervals for expected remaining life without requiring us to specify a full probability model for outcomes, which improves overall robustness.
• Apsara Pitigalaarachchi (University of Louisville)
  Title: Doubly Robust Angle based Direct Learning for Optimal Individualized Treatment Rules using AIPW Pseudo-Outcomes with Super Learner Estimation
  Abstract: Since every patient reacts differently, treatment effect heterogeneity (TEH), which describes variations in treatment responses between individuals, are becoming increasingly crucial in clinical research, because it sheds light on developing precision medicine. Conditional Average Treatment Effects (CATE), the difference in the conditional mean outcome between treatments given covariates, is crucial to study TEH and identify individualized treatment rules (ITR) to maximize patient outcomes. In practice, estimating an optimal ITR that maximizes a patient’s anticipated clinical outcome based on the patient’s characteristics is a significant challenge. The existing methods are often computationally expensive and/or not very accurate. In this work, we propose a technique combining an angle-based strategy for multiple treatments with doubly robust pseudo-outcomes to accurately predict the optimal ITR. We employ the SuperLearner to implement an ensemble of models to prevent model misspecification and achieve accuracy, while simultaneously estimating all possible pairwise CATE to improve efficiency. We conduct numerical studies to validate the merits of the proposed method.
• Yaqin Feng (Ohio University)
  Title: Initial Margin and Margin Valuation Adjustment for Cliquet Options under Jump–Diffusion Dynamics
  Abstract: We study the initial margin (IM) and margin valuation adjustment (MVA) of discretely monitored cliquet options, where MVA reflects the cost of funding collateral requirements under jump–diffusion dynamics. Mark-to-market dynamics and forward-looking margin profiles are evaluated using a Fourier–cosine (COS) expansion that accommodates both path dependence and jump discontinuities. Our results highlight a structural interaction between jump-induced tail risk and payoff truncation. While higher volatility and jump intensity increase margin and funding costs for loosely truncated cliquets, stronger payoff clipping compresses tail exposure and dampens curvature, leading to more moderate IM and MVA responses. In certain regimes, margin sensitivity to jump activity may become non-monotonic when contractual truncation dominates tail amplification. These findings show that margin and funding risk depend not only on underlying tail dynamics, but also critically on payoff design, with important implications for capital efficiency, risk management, and product structuring.
• Shixuan Wang (University of Cincinnati )
  Title: Bayesian Predictive Congruence for Incorporating Historical Information
  Abstract: Recent FDA draft guidance on the Bayesian methodology in clinical trials marks an important shift toward the routine use of Bayesian inference for regulatory decision-making, particularly in settings such as pediatric rare diseases where randomized controlled trial is infeasible and external or historical data may be leveraged to improve efficiency. In these settings, a key challenge is determining an appropriate level of borrowing that reflects the similarity between historical and current trial data. In this study, we devote ourselves to the idea of "checking if one trial data could successfully replicate the other trial data", and propose a new congruence measure pCM. pCM quantifies the heterogeneity between two datasets following the idea of the marginal posterior predictive p-value, and its asymptotic properties were derived. Building upon pCM, we develop the pointwise predictive density calibrated-power prior (PPD-CPP) to dynamically leverage historical information. Simulation studies were conducted to demonstrate the performance of these methods and the methodology was illustrated using the Mother's Gift vaccine study.
• Shih-Ting Huang (University of Louisville)
  Title: Targeted deep learning: Framework, methods, and applications
  Abstract: Deep learning systems are typically designed to perform for a wide range of test inputs. For example, deep learning systems in autonomous cars are supposed to deal with traffic situations for which they were not specifically trained. In general, the ability to cope with a broad spectrum of unseen test inputs is called generalization. Generalization is definitely important in applications where the possible test inputs are known but plentiful or simply unknown, but there are also cases where the possible inputs are few and unlabeled but known beforehand. For example, medicine is currently interested in targeting treatments to individual patients; the number of patients at any given time is usually small (typically one), their diagnoses/responses are still unknown, but their general characteristics (such as genome information, protein levels in the blood, and so forth) are known before the treatment. We propose to call deep learning in such applications In this paper, we introduce a framework for targeted deep learning, and we devise and test an approach for adapting standard pipelines to the requirements of targeted deep learning. The approach is very general yet easy to use: It can be implemented as a simple data-preprocessing step. We demonstrate on a variety of real-world data that our approach can indeed render standard deep learning faster and more accurate when the test inputs are known beforehand.
• Pratyaydipta Rudra (Oklahoma State University)
  Title: Diagnosing the Diagnostic: Assessing and Calibrating Normality Pre-Tests
  Abstract: Checking whether data appear normal before selecting a downstream test is common in applied statistics, yet part of the literature concludes the practice is not worthwhile and recommends against it. We argue that a single recommendation is hard to justify, because the effect of a pre-test depends on the distribution, the sample size, and the downstream procedure. We show that the distortion introduced by pre-testing has two distinct sources: a selective inference effect, from reusing the data for both the pre-test and the main analysis, and a misrouting effect, from a low-power pre-test directing data to a downstream test whose assumptions fail. Typical corrections such as the ones using sample splitting addresses only the first and can worsen the total distortion. We describe how to calibrate a pre-testing threshold that maintains the nominal Type I error rate while retaining power, within a simulation-based framework for assessing when pre-testing is useful.
• David Collins (Binghamton University)
  Title: Subdata selection for high-dimensional big data
  Abstract: Generalized Linear Models (GLMs) provide a flexible framework for modelling relationships between a response variable and a set of covariates, extending linear regression to accommodate various response distributions. Despite advancements in both hardware and software, efficiently fitting GLMs with a large number of observations and covariates remains computationally demanding. We address this challenge by selecting subdata: an informative, small, and efficiently obtained sample of the big dataset. Our subdata selection procedure first identifies the important covariates by combining multiple penalized likelihood estimators applied to small subsets of the data. Based on the type of response, an appropriately designed sample is then selected using only the identified covariates. This method adapts and synthesizes the corresponding method of Singh and Stufken [The New England Journal of Statistics in Data Science, 1(3), 426--438 (2023)] for linear models to the generalized linear model setting, with variable selection performed via repeated applications of Lasso. As a foundational contribution, we theoretically establish the model selection consistency of the CLASS-type variable selection procedure, which had not been previously analyzed. For linear, logistic, and Poisson regression, we then investigate the theoretical properties of the generalized linear model estimator fitted on the subdata obtained using the truly active covariates. Through extensive simulations, we demonstrate that our method has a superior screening performance, lower estimation error for active coefficients, and is computationally cheaper than existing methods.
• Qi Zhang (University of New Hampshire)
  Title: High dimensional mediation analysis with non-gaussian outcomes
  Abstract: High-dimensional mediation analysis has attracted increasing attention in recent years, motivated by applications in genomics and biomedical imaging where both exposures and mediators may be high-dimensional. Existing methods largely focus on mediator selection or rely on linear mediation models with continuous outcomes, which limits applicability in high-dimensional multi-omics studies and settings with non-Gaussian responses. Recent work by Zhang et al. (2024) introduced Mediation via Difference-in-Coefficients (MedDiC) framework for estimating causal effects in high-dimensional linear mediation models using debiased LASSO techniques. In this project, we extend MedDiC to generalized linear models.
• Yiran Jiang (University of Kentucky)
  Title: The typicality principle and its implications for statistics and data science
  Abstract: A central focus of data science is the transformation of empirical evidence into knowledge. By “knowledge,” we mean claims that are (i) supported by data through an explicit inferential procedure and (ii) accompanied by calibrated measures of uncertainty. As such, the scientific insights and attitudes of deep thinkers like Ronald A. Fisher, Karl R. Popper, and John W. Tukey are expected to inspire exciting new advances in machine learning and artificial intelligence in years to come. Along these lines, the present paper advances a novel typicality principle which states, roughly, that if the observed data is sufficiently “atypical” in a certain sense relative to a posited theory, then that theory is unwarranted. This emphasis on typicality brings familiar but often overlooked background notions like model-checking to the inferential foreground. One instantiation of the typicality principle is in the context of parameter estimation, where we propose a new typicality-based regularization strategy that leans heavily on goodness-of-fit testing. The effectiveness of this new regularization strategy is illustrated in three non-trivial examples where ordinary maximum likelihood estimation fails miserably. We also demonstrate how the typicality principle fits within a bigger picture of reliable and efficient uncertainty quantification.
• Sher Chhetri (University of South Carolina Sumter)
  Title: Exponentiated Odd Lindley-X Power Series Class of Distributions with Applications
  Abstract: Compounding and generalizing probability distributions is an old practice, but it has become more popular recently. This is because flexible models are needed to better fit different kinds of data and characterize their probabilistic structures accurately. In this work, we introduce a new family of probability distributions, the Exponentiated Odd Lindley-X Power Series class, which is derived by integrating the power series distribution with the exponentiated Odd Lindley-X family. As a special case of the generalized model, we propose a new class of distributions called the exponentiated odd Lindley-X power series (EOL-XPS) family of distributions. We derive expressions for various statistical properties, including the quantile function, moments, moment generating function, mean deviation, hazard rate function, order statistics, and entropy. As a special case, we consider the exponentiated odd Lindley-Weibull Poisson (EOLWP) distribution and conduct a simulation study to assess the validity of the model parameters. We demonstrate the usefulness and versatility of the proposed model by applying it to a COVID-19 dataset from Canada and failure time data (in hours) of Kevlar 49/epoxy strands under 90% pressure. Using these real data sets, model performances was assessed using various goodness-of-fit statistics which include; -2log-likelihood statistics, Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Consistent Akaike Information Criterion, Cramér-von Mises, Anderson-Darling, Kolmogorov-Smirnov statistic, as well as its associated p-value. The Exponentiated Odd Lindley-X Power Series (EOL-XPS) class of distributions provides a flexible framework with promising future applications and developments. It can be extended to handle multivariate datasets, making it suitable for complex domains such as finance, engineering, and environmental sciences. Integrating machine learning techniques with the model could enhance predictive accuracy in high-dimensional or time-series data.
• Bruce Phillips (Binghamton University)
  Title: Subdata Selection for Principal Component Analysis
  Abstract: Principal component analysis (PCA) is a powerful statistical tool for dimension reduction. Performing PCA requires computing the eigen-decomposition of the sample covariance matrix, or the singular value decomposition of the data matrix, which is often too computationally expensive for datasets with many observations and high dimensionality. in this talk, I will introduce our proposed subdata selection method for PCA motivated by the D-optimal design criterion. We will show that our method is computationally efficient compared to full data PCA and achieves lower eigenspace estimation error than other relevant methods.
• Abdul-Nasah Soale (Case Western Reserve University)
  Title: Doubly-robust sufficient variable selection in single-index models with outlier contamination
  Abstract: Case-influence diagnostics are central to reliable regression analysis, yet they are rarely integrated into modern variable-selection techniques, particularly in high-dimensional settings where penalized estimators can be highly sensitive to a small number of atypical observations. We propose a sparse outer product of gradients of sparse median regression called smOPG-LASSO for variable selection in single-index models with influential observations. In the first step, $\ell_1$-penalized local median regression coefficients are estimated to eliminate irrelevant predictors at each point. Then the final active predictor set is estimated based on the best rank-1 sparse loadings of the outer-product of the median coefficients via regularized singular value decomposition. By coupling median regression with local weights that downweight leverage points, smOPG-LASSO achieves double robustness to outliers in both the predictor and the response. An extensive simulation study spanning low- and ultra-high-dimensional regimes with multiple contamination mechanisms demonstrate the strong performance of smOPG-LASSO compared to existing variable selection techniques. Two real applications on air pollution and gene selection are provided to illustrate the practical utility of the proposed method. The theoretical properties of the smOPG-LASSO estimator are also included.
• Pang Lingyou (Johns Hopkins University)
  Title: Spectral Geometry of Synthetic-Data-Induced Model Collapse
  Abstract: Synthetic data has become a training substrate for large language models, but its risks are not fully visible from provenance labels or surface fluency: a model may remain fluent while its response distribution loses semantic volume. We introduce SIGMA, a post hoc and model-agnostic framework for measuring synthetic-data-induced model collapse through response spectral geometry. Given a fixed prompt bank, SIGMA embeds model responses in a frozen semantic space and summarizes them by a response Gram matrix. Collapse is quantified by a baseline-subtracted regularized log-determinant, a response-volume functional sensitive to both isotropic shrinkage and anisotropic loss of effective rank. Under a covariance-mixture idealization, the synthetic-ratio response curve is concave and yields interpretable summaries including scalar fidelity, marginal alpha-rates, temporal decay, and geometry-safe synthetic ratios. We further develop scalable SIGMA-lite estimators based on deterministic spectral envelopes and stochastic observed-block scaling, with uncertainty assessed through paired prompt-cluster and crossed model--domain resampling.
• Keshav Pokhrel (University of Michigan - Dearborn)
  Title: Bayesian Method for Estimating Kumaraswamy Weibull Parameters
  Abstract: In this report, we investigate Bayesian estimation for the Kumaraswamy–Weibull (KW) distribution under informative, non-informative, and weakly informative priors. We derive the corresponding posterior distributions and evaluate the performance of the Bayesian approach using both complete (uncensored) and partially right-censored data. Parameter estimation is performed via Markov Chain Monte Carlo (MCMC) using the No-U-Turn Sampler (NUTS) algorithm. The Bayesian estimation procedure for the KW distribution yielded a better fit to both uncensored and right-censored Cancer Survival data.
• Pratik Misra (Binghamton University)
  Room: Ohio Room
  Title: Structural identifiability in Gaussian graphical models
  Abstract: Algebraic Statistics is an emerging field of research that uses techniques from Algebraic Geometry, Combinatorics and Commutative Algebra to enhance our understanding of statistical and causal inference problems. A key area of research in this field is the Gaussian graphical models, where the dependence structure between jointly normal random variables is determined by a graph. In this talk, I will present the problem of structural identifiability in Gaussian graphical models. I will show how imposing symmetry conditions on model parameters can guarantee identifiability. Building on prior work by Peters and Bühlmann and by Wu and Drton, we establish structural identifiability under homogeneous structural coefficients, as well as for a broader class of models with partially homogeneous structural coefficients called BPEC-DAGs. In the end, I will present an analogue of the GES algorithm for learning BPEC-DAGs and evaluate its performance on real datasets.
• Pei Wang (Bowling Green State University)
  Room: Pennsylvania Room
  Title: Multitask Sufficient Dimension Reduction for Complex High-Dimensional Data
  Abstract: Sufficient dimension reduction is a powerful statistical learning framework for identifying low-dimensional structures in high-dimensional data while preserving the information relevant to prediction. Multitask learning, meanwhile, improves predictive performance by leveraging shared information across multiple related tasks. In this work, we propose a novel multitask sufficient dimension reduction approach that integrates these two frameworks to enable efficient learning from complex high-dimensional data. The proposed method seeks low-dimensional representations that retain essential information for simultaneously modeling multiple tasks, thereby improving both predictive accuracy and interpretability. Extensive simulation studies and real data applications are conducted to evaluate the performance of the proposed approach and demonstrate its effectiveness in multitask high-dimensional learning settings.
• Hyeong Hyun (University of Cincinnati)
  Room: Maryland Room
  Title: Transport-induced Amortized Bayesian Computation
  Abstract: Simulation-based inference (SBI) enables Bayesian inference in complex models with intractable likelihoods but accessible simulators. Approximate Bayesian computation (ABC) offers a principled solution with theoretical guarantees, but its effectiveness collapses in high dimensions due to the curse of parameter dimensionality. Modern machine learning (ML) based SBI methods substantially alleviate this issue, but typically sacrifice theoretical guarantees and incur substantial training costs, raising concerns on accuracy and scalability. We propose Transport-induced Amortized Bayesian Computation (NABC), a new framework that reconciles the statistical guarantees of ABC with the performance and amortization benefits of modern SBI methods. TABC learns generative conditional transport maps from a baseline distribution to the target posterior and decouples training from calibration, enabling inference to exploit an effectively unbounded supply of simulated data. We establish that TABC preserves the asymptotic exactness of ABC while substantially mitigating the curse of dimensionality of ABC. TABC not only outperforms state-of-the-art ML-based SBI methods but also achieves these improvements at substantially lower computational cost. We demonstrate its effectiveness on a range of benchmark studies reflecting practical scenarios and further extend its use to the challenging Bayesian inference for differentially privatized data.
• Netra Khanal (The University of Tampa)
  Room: Virginia Room
  Title: Probability Distributions for Modeling Stock Market Returns
  Abstract: Investing in stocks and shares is a common strategy to pursue potential gains while considering future financial needs, such as retirement and children’s education. Effectively managing investment risk requires thoroughly analyzing stock market returns and making informed predictions. Traditional models often utilize normal variance distributions to describe these returns. However, stock market returns often deviate from normality, exhibiting skewness, higher kurtosis, heavier tails, and a more pronounced center. This paper investigates the Laplace distribution and its generalized forms, including asymmetric Laplace, skewed Laplace, and the Kumaraswamy Laplace distribution, for modeling stock market returns. Our analysis involves a comparative study with the widely-used Variance-Gamma distribution, assessing their fit with the weekly returns of the S&P 500 Index and its eleven business sectors, drawing parallel inferences from international stock market indices like IBOVESPA and KOSPI for emerging and developed economies, as well as the 20+ Years Treasury Bond ETFs and individual stocks across varied time horizons. The empirical findings indicate the superior performance of the Kumaraswamy Laplace distribution, which establishes it as a robust alternative for precise return predictions and efficient risk mitigation in investments.
• Minjie Wang (Binghamton University)
  Title: High-Dimensional Variable Selection with Diffusion-Generated Synthetic Data
  Abstract: Recent advances in generative models and generative artificial intelligence (AI) have prompted the advent of synthetic data generation, which challenges the traditional statistical practice of focusing solely on the raw data and shifts toward a more synthetic data-centric approach. On the other hand, variable selection for high-dimensional, highly correlated data has long been a challenging problem, often yielding unstable and unreliable models. To enhance variable selection, we present a resample-aggregate framework that exploits diffusion models' ability to generate high-fidelity synthetic data. Specifically, we draw multiple pseudo-data sets from a diffusion model fitted to the original data, apply any off-the-shelf selector (e.g., lasso or SCAD), and store the resulting inclusion indicators and coefficients. Aggregating across replicas produces a stable subset of predictors with calibrated stability scores for variable selection. Theoretically, we show that the proposed method is selection consistent under mild assumptions. Extensive simulations show consistent gains over the lasso, stability selection, and knockoff baselines, especially when predictors are strongly correlated, achieving higher true-positive rates and lower false-discovery proportions. By coupling diffusion-based data augmentation with principled aggregation, our method advances variable selection methodology and broadens the toolkit for interpretable, statistically rigorous analysis in complex scientific applications.
• Zi Ye (Lehigh University)
  Title: An adaptive nonparameteric rank-based two-sample test for high-dimensional data
  Abstract: In this talk, we propose a novel adaptive, nonparametric, rank-based procedure for comparing two high-dimensional populations. By extending adaptive testing ideas to a nonparametric framework, the method enables robust inference without imposing restrictive distributional assumptions. The procedure aggregates information across a family of test statistics constructed from sum-of-powers and supremum-type asymptotic relative effects. We derive the joint asymptotic null distribution of these statistics and develop an adaptive approach for computing P-values. Theoretical analysis shows that the proposed test adapts to a wide range of alternatives, ranging from dense to sparse, while requiring substantially weaker assumptions than existing methods. Extensive simulations demonstrate that the proposed test accurately controls Type I error while providing superior power and stability, making it well-suited for complex highdimensional analyses. An application comparing gene expression profiles between Inflammatory Breast Cancer (IBC) and non-IBC cases is provided to illustrate the practical utility of the proposed test.
• Jiawei Zhang (University of Kentucky)
  Title: Additive-Effect Assisted Learning
  Abstract: It is quite popular nowadays for researchers and data analysts holding different datasets to seek assistance from each other to enhance their modelling performance. We consider a scenario where different learners hold datasets with potentially distinct variables, and their observations can be aligned by a nonprivate identifier. Their collaboration faces the following difficulties: first, learners may need to keep data values or even variable names undisclosed due to, e.g. commercial interest or privacy regulations; second, there are restrictions on the number of transmission rounds between them due to e.g. communication costs. To address these challenges, we develop a two-stage assisted learning architecture for an agent, Alice, to seek assistance from another agent, Bob. In the first stage, we propose a privacy-aware hypothesis testing-based screening method for Alice to decide on the usefulness of the data from Bob, in a way that only requires Bob to transmit sketchy data. Once Alice recognizes Bob’s usefulness, Alice and Bob move to the second stage, where they jointly apply a synergistic iterative model training procedure. With limited transmissions of summary statistics, we show that Alice can achieve the oracle performance as if the training were from centralized data, both theoretically and numerically.
• Chudamani Poudyal (University of Central Florida)
  Title: Learning Flexible Discrete Distributions via Transmutation of Continuous Distributions
  Abstract: This talk presents a discrete analogue of the transmuted Weibull distribution constructed via discretization. The model introduces an additional transmutation parameter that controls tail behavior and probability mass allocation while retaining tractable analytical properties. Key results include the probability mass function, cumulative distribution function, and moments, with theoretical establishment of finiteness of moments of any positive order and their numerical evaluation. The shape behavior is examined, identifying conditions under which the model is decreasing or unimodal. Statistical inference is developed using the score function and Fisher information matrix, with computation based on finite summation and tail approximation for parameter estimation and uncertainty quantification. A simulation study evaluates performance under different parameter settings. A real data analysis is included to demonstrate the applicability of the model and its flexibility compared to standard discrete distributions.
• Xuexia Wang (Florida International University)
  Title: Correlation-Aware Inference of Three-Dimensional Genome Architecture Reveals Subtype-Specific Regulatory Mechanisms in Medulloblastoma
  Abstract: High-throughput Hi-C technologies enable systematic investigation of three-dimensional genome organization, but identifying significant chromatin interactions remains challenging due to large-scale multiple testing, spatial dependence, and limited biological replicates. We developed a correlation-aware framework integrating FitHiC2, LAWS-HiC, clusterProfiler to detect significant chromatin interactions, account for local spatial correlation, and interpret enriched biological pathways. We applied this framework to pediatric medulloblastoma Hi-C data, including a discovery dataset (GSE246125; n = 21: 8 Group 3 [G3], 13 Group 4 [G4]) and two replication datasets (GSE267471, n = 2; GSE240410, n = 4). In discovery analysis, we identified 2.3 million and 2.2 million significant chromatin interactions (FDR adjusted p<0.01) in G3 and G4, respectively, including 42,226 G3-specific and 18,074 G4-specific interactions. Findings were validated using independent datasets GSE267471 and GSE240410 for G3, confirming the robustness of subtype-specific regulatory landscapes. G3 showed enriched enhancer–promoter contacts and pathways related to chromatin remodeling, nucleosome assembly, and telomere organization. Identified telomere-related pathway (p_adj=0.002 in GSE246125; p=0.0004 in GSE267471 and p=9.5x10-6 in GSE240410) highlights subtype-specific regulatory architecture in medulloblastoma.
• Duval Zephirin (Florida SouthWestern State College)
  Title: Portfolio Maximization for Investors in Fads Models Driven by Lévy Processes
  Abstract: In recent years, there has been a growing interest in the use of Lévy processes to model stock market behavior. Empirical studies of stock prices indicate distributions with heavy tails, which are incompatible with a Gaussian distribution, and a more realistic approach is to allow small jumps in small time intervals. Lévy processes are used to model stock prices for informed and uninformed investors. We derive the optimal portfolio for the informed investor in a Fads model driven by Lévy processes under asymmetric information and stochastic volatility of the stock price. We employ methods of stochastic calculus, namely the Hamilton-Jacobi-Bellman equation, instantaneous centralized moments of returns, and three-level Crank-Nicolson method. We solve numerically the partial differential equation associated with the optimal portfolio. Under the power utility function, analogous results to those obtained in the jump-diffusion model under logarithmic utility function and deterministic volatility are obtained.
• Sanjib Basu (University of Illinois Chicago)
  Title: Cure rate: Models, Identifiability, and Bayesian Inference
  Abstract: Cure rate or limited failure models postulate a fraction of subjects or units to be eventually failure free. These models can be formulated in different ways, such as a mixture cure model or the alternative bounded cumulative hazard model. The issue of identifiability has been a topic of substantial research both in the cure rate model as well as in the competing risks literature. We find that identifiability may be of concern in general model formulation, but it can often be regained with modeling restrictions. We also consider a class of cure rate models indexed by an adaptive activation parameter and a function. We establish that the class is stochastically ordered in the activation parameter and also establish two identifiability results for this class. The first and last-activation models are members of this class, whereas many cure rate models proposed in the literature are also part of this class. We apply the proposed approach in analysis of colon cancer and triple-negative breast cancer survival data.
• Jung-Ying Tzeng (North Carolina State University)
  Title: A Federated Meta-Analysis Framework for Rare CNV Association Testing
  Abstract: Copy number variants (CNVs) are genomic gains or losses of DNA segments, typically >50 basepairs. Rare CNV association studies have gained increasing attention due to advances in rare CNVdetection and growing evidence of their roles in disease etiology. Because rare CNV analysis requireslarge sample sizes to achieve adequate power, approaches based on summary statistics, such asmeta-analysis and federated analysis, provide efficient and privacy-preserving strategies foraggregating data across studies. Meta-analysis can naturally accommodate heterogeneous covariatestructures (e.g., different covariate sets and/or different covariate effects across studies), a scenariocommonly encountered in multi-ethnic analyses. However, unlike SNPs, CNV meta-analysis facesunique challenges that complicate cross-study integration, including inconsistent CNV locus definitionsacross studies and lack of standardized association summary outputs. To address these challenges,we propose fedCONCUR, a federated meta-analysis framework for rare CNV association testingwhere a coordinating site collects relevant study-level summary statistics and CNV genomicinformation from participating studies. We develop a federated testing algorithm that harmonizes CNVlocus definitions across studies, computes the collapsing test statistics and p-values based on therare-CNV score test CONCUR, and reduces communication to a single round. In contrast toconventional federated analysis that approximates a pooled analysis using globally shared parameterestimates, fedCONCUR preserves study-specific regression parameter estimates in the score statisticconstruction, which improves robustness to cross-study heterogeneity and avoids inflation of Type Ierror rate. We evaluate the proposed framework using simulations and application to the Alzheimer’sDisease Sequencing Project (ADSP) and demonstrate its utility and scalability for large-scale rareCNV studies.
• Gokarna Aryal (Purdue University Northwest)
  Title: L-estimation of Location-Scale Distributions Weighted by Cubic Rank Transmuted Kumaraswamy Distribution
  Abstract: Identifying the underlying distribution from a sample dataset is always a critical task in statistical inference. For multimodal datasets, mixture models are commonly used and typically require iterative EM algorithms for parameter estimation-an approach that can be both computationally intensive and sensitive to initialization. To address these issues, we introduce a method in modeling standalone loss distributions by integrating L-estimation with a cubic rank transmuted Kumaraswamy (CRTK) probability density with unit support - as the weight-generating function. The CRTK-weighted L-estimation framework is particularly effective for multimodal data, enabling flexible weighing schemes that capture complex distributional shapes. Through simulation studies and real-world bimodal dataset analysis, we show that CRTK- weighted models offer competitive or superior performance compared to finite mixture models in terms of flexibility, efficiency, and parsimony.
• Henry So (Oakland University)
  Title: Flexible Cure Rate Modeling with Proportional Odds, COM-Poisson Competing Risks, and Gamma Frailty
  Abstract: Cure rate models are increasingly used for time-to-event data with a meaningful fraction of long-term survivors. We present a flexible framework that combines: (i) a competing risks formulation for the latent number of causes modeled by the COM-Poisson distribution (allowing under- and over-dispersion), (ii) a proportional odds model for the susceptible survival component, and (iii) gamma frailty to capture unobserved heterogeneity. This yields a unified model class that includes several familiar cure rate models as special cases and supports interpretable covariate effects on the proportional-odds scale. We describe likelihood-based estimation for right-censored data under the standard non-informative censoring assumption and discuss practical implementation, including EM-type computation and likelihood/information-based model discrimination across related submodels. We also note how the approach can be paired with common baseline distributions (e.g., Weibull or log-logistic) depending on modeling goals. Finally, we outline future directions that relax non-informative censoring, including informative censoring structures linked to latent susceptibility or frailty, framed as open problems and a roadmap for ongoing work.
• Xiaofeng Zhu (Case Western Reserve University)
  Title: Multi-ancestry multivariable Mendelian Randomization with Transfer Learning
  Abstract: Multivariable Mendelian randomization (MVMR) is a widely used approach to estimate the causal effects of exposures on disease outcomes. However, its applications are predominantly limited to European populations due to the much larger sample sizes available in European GWAS, which provide greater statistical power. Joint modeling of multiple ancestral populations has been suggested to improve causal inference in MR, but this approach has so far been limited to univariable MR. Here, we present MRBEE-TL, a novel MVMR that integrates transfer learning with bias-corrected estimating equations, to enhance causal inference in underpowered populations and to assess causal effect heterogeneity across populations. In simulations, MRBEE-TL consistently outperformed MR methods that relied solely on population-specific GWAS data, demonstrating superior estimation accuracy, statistical power, and type-I error control. In real data analysis, we will present exemplary results of causal effect estimation for multiple risk factors on five stroke subtypes in both Europeans and Asians. MRBEE-TL offers a valuable tool to jointly perform MR in multiple populations simultaneously. This is a joint work with Yihe Yang.
• Nirajan Budhathoki (Henry Ford Health & Michigan State University Health Sciences )
  Title: Modeling COVID-19 Outcomes Using a New Extended Inverse Burr Distribution
  Abstract: A new extension of the inverse Burr distribution, namely the extended odd-Weibull inverse Burr distribution (ExOW-IB), is proposed and defined. This new distribution is a member of the extended odd Weibull-G family. Statistical properties, including linear representations, quantile functions, moment-generating functions, and others, are derived. The method of maximum likelihood is discussed to estimate parameters of the proposed distribution. A simulation study is performed to assess the performance of the estimation method. The proposed distribution offers greater flexibility in modeling COVID-19 mortality data in comparison to other existing distributions.
• Indranil Ghosh (University of North Carolina Wilmington)
  Title: Copula-based mutual information measures and mutual entropy: A brief survey
  Abstract: The notion of divergence is often used to measure the separation between two parametric densities with possible applications of differential geometry to problems of parametric inference among others. Among them, in this paper, we discuss several popular divergence (alias pseudo-distance) measures such as the Power divergence, Kullback-Leibler information, J-divergence, Hellinger distance, Bhattacharya divergence, alpha-Divergence and so on. A wide range of discussion in terms of properties, relationship among these measures, and results related to distance between two probability distributions are derived via copula functions. In addition, several dependency concepts such as weak negative dependence etc. are visited in terms of copula-based divergence measures and important inequalities are obtained.
• Shili Lin (Ohio State University)
  Title: A Divid and Conquer Strategy for Recapitulating Whole-Genome 3D Structure Using Hi-C Data
  Abstract: The three-dimensional (3D) spatial organization of the genome is closely linked to biological function and can be captured by Hi-C assays through interrogating genome-wide chromatin interactions. Methodologies for inferring 3D structures from Hi-C data, summarized as a two-dimensional (2D) contact matrix, include sampling-based methods. These are probabilistic model-based approaches that can account for features inherent in Hi-C data, including dependency, heterogeneity, over-dispersion, and sparsity. However, whole-genome 3D structure recapitulation is too computationally expensive for current sampling-based methods. While a chromosome-by-chromosome strategy is an option, it ignores important information from inter-chromosomal contacts. To address these issues, we propose the truncated Random effect EXpression-cut and paste (tREX-cap) method, which applies the tREX model alongside a isometric-transformation pasting strategy. This method inherits the data-feature-cognizant properties of tREX while efficiently inferring the whole-genome 3D structure. We demonstrate the performance of tREX-cap through an extensive simulation study and an analysis of a Hi-C lymphoblastoid dataset.
• Durga Kutal (Augusta University)
  Title: Bayesian Mixture and Non-Mixture Cure Models Based on the Fréchet Distribution for Right-Censored Survival Data
  Abstract: Survival studies often include long-term survivors who never experience the event of interest, making cure rate models important for proper analysis. This study proposes Bayesian mixture and non-mixture cure models for right-censored survival data using the Fréchet distribution to capture heavy-tailed survival behavior and heterogeneity in biomedical data. Bayesian inference is performed through MCMC methods implemented in OpenBUGS. Simulation studies under varying sample sizes, censoring levels, and cure fractions demonstrate improved estimator performance with increasing sample size. The models are applied to leukemia bone marrow transplantation data and melanoma data from the ECOG Phase III clinical trial E1684 with covariate effects. Comparisons using Bayesian model selection criteria show that the proposed Fréchet-based cure models outperform Weibull and exponentiated exponential alternatives, providing a flexible and effective framework for survival data with long-term survivors.
• Suvra Pal (University of Texas at Arlington)
  Title: A Mixture Cure Regression Model with Competing Risks
  Abstract: Cancer remains the second leading cause of death in the United States, accounting for 605,213 deaths in 2021 and surpassing COVID-19 fatalities. While the cancer mortality rate has declined by 33% since 1991, driven largely by advances in treatment and increased rates of remission, patients now face not only the risk of primary cancer but also competing risks from other cancers or unrelated diseases. Although survival analysis under competing risks and cure models has been well studied separately, limited work exists that integrates the possibility of cure from one cause in the presence of competing risks. We propose a novel unified competing risks cure model that incorporates covariates within a cause-specific hazard framework. The model accommodates cure from the primary cancer while ensuring that the overall survival probability asymptotically approaches zero, aligning with the clinical expectation of eventual failure. We establish conditions for model identifiability under regression and develop a computationally efficient Expectation-Maximization (EM) algorithm for parameter estimation. Simulation studies demonstrate the model’s robustness across various scenarios. The practical utility of the method is illustrated through an application to breast cancer data from the SEER database.

Index of names, alphabetical by first name, with links to position in sessions (back to top)

• Abdul-Nasah Soale
• Ash Abebe
• Amos Natido
• Anna Panorska
• Apsara Pitigalaarachchi
• Ayaka Yagi
• Aye Aye Maung
• Bruce Phillips
• Chudamani Poudyal
• David Collins
• Derek Young
• Drew Lazar
• Durga Kutal
• Duval Zephirin
• Fatemeh Salboukh
• George Yanev
• Gokarna Aryal
• Golshid Aflaki
• Gozde Sert
• H. Frazier Bindele
• Hamdan Hasan
• Hapuhinna Nelum Kasturiratna Dhanuja
• Henry So
• Hideki Nagatsuka
• Hon Yiu (Henry) So
• Hossein Moradi Rekabdarkolaee
• Hyeong Hyun
• Indranil Ghosh
• Jeonghwa Lee
• Jiawei Zhang
• John Nolan
• Jongwook Kim
• Jung-Ying Tzeng
• Keshav Pokhrel
• Kevin Lee
• Kevin McGregor
• Kun Gou
• Lei Fang
• Mahsa Ashouri
• Michael Levine
• Minjie Wang
• Moh Khalid Hasan
• Mozhdeh Forghaniarani
• Munni Begum
• Nadeesha Jayaweera
• Netra Khanal
• Nirajan Budhathoki
• Olcay Arslan
• Olusegun Michael Otunuga
• Pang Lingyou
• Pei Wang
• Philip Yates
• Pratik Misra
• Pratyaydipta Rudra
• Qi Zhang
• Qi Zheng
• Rida Benhaddou
• Saman Muthukumarana
• Samuel Morrissette
• Sanjib Basu
• Chad Schafer
• Scott Smith
• Sher Chhetri
• Shih-Ting Huang
• Shih-Ting Huang
• Shili Lin
• Shimeng Huang
• Shixuan Wang
• Shuoyang Wang
• Souparno Ghosh
• Surani Matharaarachchi
• Suvra Pal
• Tatjana Miljkovic
• Tomasz J. Kozubowski
• Tomoaki Imoto
• Tung-Lung Wu
• Xiaofeng Zhu
• Xuexia Wang
• Yan Zhuang
• Yaqin Feng
• Yifan Hsu
• Yiran Jiang
• Yuting Chen
• Zheng Wei
• Zhu Wang
• Zi Ye