Choose
Select an interesting statistical idea, establish why it matters, and make it accessible without sacrificing depth.
STA 701S Duke Department of Statistical Science
Research in progress. Ideas in conversation.
An advanced seminar at the research frontiers of statistical science. Third-year students develop ideas that may grow into research; fourth- and fifth-year students present research in progress or completed work. Every speaker builds an accessible introduction to the problem, area, and central ideas, creating space for students and faculty to ask sharper questions and deepen understanding through discussion.
01 / The seminar
A strong presentation makes an interesting idea accessible.
STA 701S is a working forum for learning across areas and practicing the clear communication of technical ideas. Presenters choose a topic worth understanding, build a thoughtful and engaging talk, and create a conversation that helps speaker and audience think more carefully.
Select an interesting statistical idea, establish why it matters, and make it accessible without sacrificing depth.
Build a focused, accurate, and engaging presentation that gives the audience a clear path into the idea.
Use constructive questions to test understanding, surface assumptions, and connect ideas across statistical science.
02 / Participation
Speakers bring an idea worth understanding; the room helps make sense of it.
Presenters prepare a focused, high-quality talk that gives the audience enough context to engage deeply.
Third-year students should develop and present ideas that may grow into research; fourth- and fifth-year students should present their research in progress or completed work. In every case, the talk should serve as an accessible introduction to the problem, area, and central ideas—not as a dense technical treatise.
Participants listen actively, ask constructive questions, and help the room reach a deeper understanding.
Each presentation is no more than 20 minutes, followed by a 10–15 minute discussion period.
03 / Calendar
The current speaker calendar is shown below. Presentations are grouped by meeting date.
Student submission guideAll seminar meetings
Mondays 11:45 AM–1:00 PM Old Chem 116
Alexander Volfovsky
Statistical models are useful because they let us compare what we observed with what the model says could have happened. This talk introduces posterior predictive checks as a practical way to make that comparison visible, using simple examples to connect simulation, model criticism, and uncertainty. The goal is an accessible introduction to how a model can reveal its own failures, and to what these checks can and cannot tell us about a completed analysis.
Alexander Volfovsky
This second example illustrates a talk title without an uploaded presentation, so students can see how a non-linked schedule entry appears on the website.
Lorenzo Mauri
Factor models are popular approaches for analyzing high-dimensional data to extract low-rank signals and estimate covariances. They decompose the covariance matrix as the sum of low-rank and diagonal components. A key issue is how to choose the latent dimension $k$, which is particularly challenging when the factor model only holds approximately and in low signal-to-noise scenarios. Bayesian overfitted factor models specify an upper bound on $k$ and rely on structured shrinkage priors to effectively remove extra components. Such approaches are popular and effective, but computationally expensive. We propose a much faster EigenBayes approach that provides valid uncertainty quantification, based on spectral estimation of latent factors and adaptive empirical Bayes calibration of key hyperparameters. The resulting posterior distribution factorizes across outcomes and is analytically tractable, bypassing Markov chain Monte Carlo. We show that EigenBayes adapts to the signal-to-noise ratio of each outcome and latent dimension, while shrinking superfluous latent components to zero. We establish favorable asymptotic properties and demonstrate strong empirical performance in numerical experiments and a genomics application, where EigenBayes outperforms state-of-the-art alternatives.
Leah Johnson
Treatment effects rarely apply uniformly across patients, and average effects can obscure clinically important heterogeneity. We consider global testing of treatment-effect heterogeneity in randomized trials, where flexible models like XGBoost capture nonlinear covariate interactions but complicate inference, since the null distribution of any resulting test statistic is intractable analytically. We address this with permutation testing, which calibrates the full model-fitting pipeline by refitting under each permutation. We develop two approaches: an general A-learning framework that uses SHAP-based feature importance for clinical interpretability, and a pseudo-outcome approach that directly targets a chosen treatment-effect contrast at the cost of outcome-specific construction. We compare both methods across continuous, binary, and time-to-event outcomes in simulation, and discuss their trade-offs for exploratory precision-medicine analyses in pharmaceutical development.
Private slide delivery; no public PDF is available.
Piotr Suder
Bayesian optimization (BO) is widely used for optimizing expensive black-box systems, but its performance often degrades when relying on simulators/digital twins that exhibit model mismatch due to unmodeled dynamics, noise, or physical effects. This sim-to-real gap can lead to poor real-world performance despite strong results in simulation. We propose an online digital twin calibration approach for BO that adaptively aligns simulations with real-world feedback. The true objective and sim-to-real gap / discrepancy are modeled independently using Gaussian process (GP) surrogates, yielding a principled multi-fidelity surrogate that combines cheap, biased simulations with intermittent, unbiased high-fidelity evaluations. We formulate a deep GP surrogate which allows us to flexibly model the lengthscales of the discrepancy function and search specifically for the simulator calibration most useful for the underlying optimization task. We introduce the expected improvement variance reduction (EIVAR) acquisition function which targets reducing posterior uncertainty of the true objective in promising regions by jointly selecting design variables and calibration parameters. We provide an efficient approximation algorithm for computing and optimizing EIVAR. Our method improves sample efficiency and outperforms standard multi-fidelity and offline calibration baselines.
Private slide delivery; no public PDF is available.
Yen-Chun Liu
The need to explore and/or optimize expensive simulators with many qualitative factors arises in broad scientific and engineering problems. Our motivating application lies in path planning – the exploration of feasible paths for navigation – which plays an important role in robotics and assembly planning. For complex settings, the parameter space for path exploration can be discrete and high-dimensional, and the evaluation of path feasibility requires expensive virtual simulations. A carefully selected experimental design is thus essential for timely decision-making. We propose here a novel framework called QuIP, for experimental design of a Gaussian process (GP) surrogate with Qualitative factors via Integer Programming. QuIP leverages a GP surrogate with an exchangeable covariance function. For initial design, we show that its maximin design can be formulated as an assignment problem from operations research, which can be efficiently and globally optimized via state-of-the-art integer programming solvers. For sequential design (specifically, for active learning or black-box optimization), we show that its design problem can similarly be formulated as an assignment problem, which facilitates efficient and reliable optimization with state-of-the-art solvers. We demonstrate the effectiveness of QuIP over existing design methods in a suite of path planning experiments and an application to rover trajectory optimization.
Gwen Jacobson
In completely randomized experiments, regression adjustment for pre-treatment covariates can effect greater efficiency in average treatment effect (ATE) estimation compared to using the unadjusted difference-in-means estimator. However, analysts sometimes contend with covariate missingness, complicating the adjustment approach. To overcome this missingness without discarding data, analysts use multiple imputation (MI) to generate multiple completed data sets. However, another question arises: What happens if the imputation model is mis-specified? Use of a mis-specified imputation model may, or may not, still improve the inferences. In this research, we investigate this question with the goal of providing guidance to analysts on when they might benefit from using multiple imputation with missing covariates in randomized experiments. We present preliminary results from a simulation study where the mis-specified model corresponds to an unintentional omission of an important interaction, as might happen in default applications of popular routines like multiple imputation by chained equations. We examine the relationship between estimation efficiency and interaction strength. We also preview ongoing work on additional experimental settings, future methodological comparisons, and potential diagnostics to provide guidance on the use of MI in ATE estimation.
Caitrin Murphy
To remain competitive and maximize revenue, online retailers must respond to customer search queries with personalized lists of items that are both relevant to the query and appealing to the customer’s individual preferences. Discrete choice models are commonly used to model purchase behavior in this setting. However, such approaches typically rely on the assumptions that customers examine every displayed item before making a purchase decision, and that utilities are determined entirely by observed information. In practice, customers may lose interest before reaching the end of a list, making the composition and ordering of the display important decisions for retailers. Additionally, a search query may reveal only part of a customer’s preferences, creating uncertainty about the utility assigned to each item. Failure to account for limited attention or unobserved preferences can lead to suboptimal recommendations and reduced revenue. We propose two extensions to the multinomial logit model that address these limitations. First, each customer is associated with a latent preference vector that represents tastes not revealed by the query. Second, the number of items examined follows a discrete-time survival process, so position effects arise from browsing behavior rather than from exogenous weights. We formulate the construction of personalized displays as a contextual bandit problem and propose a sequential recommendation policy using the SquareCB algorithm. We show that, with high probability, the regret is sublinear. Simulation experiments quantify the value of modeling latent preferences and limited attention.
Private slide delivery; no public PDF is available.
Vivek Singh
Testing independence between random vectors requires detecting dependence without specifying their marginal distributions. Multivariate ranks based on optimal transport address this problem by matching observations to a fixed reference set, extending the distribution-free properties of univariate ranks. We introduce an alternative construction of multivariate ranks based on stable matching. Instead of minimizing a global transport cost, the assignment satisfies a local stability condition. Under standard regularity conditions, the stable assignment is unique and the resulting rank vector is distribution-free. At the population level, measure transportation, preservation of dependence, and convergence of empirical ranks provide the basis for consistent independence tests using test statistics such as distance covariance. Stable matching also permits different choices of utility, producing rank maps that can lead to different power against dependence alternatives. We compare stable matching with optimal transport in terms of computation, the resulting maps, and testing power, with particular interest in heavy-tailed settings.
Private slide delivery; no public PDF is available.
Kateryna “Kat” Husar
Randomized controlled trials typically assume that prognostic covariates are known and available at no cost. In practice, obtaining high-dimensional pretreatment data is costly, forcing a trade-off between covariate-adaptive precision and a measurement budget. We introduce Dynamic Adaptive Rerandomization via Thompson Sampling (DARTS), which treats covariate acquisition as a sequential optimization problem embedded within a design-based causal inference task. A budgeted combinatorial Thompson sampler learns which covariates are most prognostic across successive batches; selected covariates then drive rerandomization and regression adjustment to reduce batch-level average treatment effect variance. Our primary theoretical contribution is a decoupling result: adaptive covariate selection based on past batches preserves batch-level randomization validity, and the cumulative inverse-variance weighted estimator achieves at least nominal asymptotic coverage. We further derive a Bayes risk bound for the acquisition layer that matches the minimax lower bound up to logarithmic factors. Empirically, DARTS systematically concentrates the budget on informative features, significantly closing the efficiency gap to oracle designs while maintaining strict inferential validity.
DongKyu “Derek” Cho
Proximal Policy Optimization (PPO) stabilizes policy updates by clipping the importance ratio between the new policy and the data-collecting policy, helping prevent overly large updates during training. The standard rule uses one fixed clipping range for all samples. While this choice offers simplicity and effectiveness, it fails to distinguish statistically harmful extreme ratio values from large, yet tolerable changes. We study how to adapt the clipping radius to the reliability of the importance ratio distribution. Our case study demonstrates that reliable updates require controlling these extreme ratios, rather than merely verifying that individual ratios fall within a fixed interval. Motivated by this observation, we propose Ordered Policy Optimization (OPO). In each minibatch, OPO sorts the importance ratios, selects the extreme ratios that potentially have the largest effect on the update, and assigns them adaptive sample-wise radii instead of using the fixed clipping radius. These radii are derived from a prescribed pattern for the tails of the ratio distribution, which specifies desired target values for the selected extreme ratios. The resulting objective remains computationally simple and requires only minibatch sorting and a modified policy loss. Experiments conducted on MuJoCo and Atari benchmarks demonstrate that this simple modification achieves strong performance relative to established baselines.
Private slide delivery; no public PDF is available.
Suchismita Roy
Accounting for the wide heterogeneity of individuals remains a challenge in epidemic modeling. In this paper, we incorporate individual-level covariate information to improve inference on population-level epidemic parameters. We develop a closed-form likelihood via the Dynamical Survival Analysis (DSA) method, which can be used for both frequentist and Bayesian inference. Our method jointly estimates the infection rate, the recovery rate, and the covariate effects. It thereby identifies which covariates most strongly affect individual susceptibility to the disease and allows individuals to be ranked by their susceptibility. The method also accommodates under-reported case counts. We illustrate the approach using the SIR (Susceptible-Infected-Recovered) model with and without network structure; it can be extended similarly to the SEIR (Susceptible-Exposed-Infected-Recovered) model. We demonstrate our method on simulated data where the covariates are different biomarker levels.
Private slide delivery; no public PDF is available.
Shuo Wang
The Bayesian optimal interval (BOIN) design is widely used in early-phase trials for its prespecified dose-escalation rules. However, its interim decisions use only current-dose data and ignore information from other doses. We propose a cross-dose borrowing extension that preserves BOIN’s three-state decision structure while allowing data-adaptive boundaries. At each eligible decision, a leave-one-dose-out dose–toxicity model predicts current-dose toxicity using other-dose data. These predictions provide cross-dose evidence for the underdosing, target and overdosing hypotheses, which is tempered by agreement with current-dose data. The tempered evidence updates prior probabilities of the three BOIN hypotheses while keeping their locations fixed. Underdosing evidence raises the escalation boundary, while overdosing evidence lowers the de-escalation boundary. We also provide a rule for crossing decision boundaries. Simulations across dose–toxicity scenarios and toxicity targets compare the proposed design with fixed BOIN and other adaptive interval designs. Results show that cross-dose borrowing improves dose targeting while preserving an explicit BOIN-based allocation rule.
Private slide delivery; no public PDF is available.
Gyeonghun “Hun” Kang
Prior-data fitted networks (PFNs) have recently emerged as a powerful approach for Bayesian prediction tasks, approximating the posterior predictive distribution (PPD) through in-context learning. Despite their strong empirical performance and ability to go beyond point predictions, theoretical understandings of the algorithmic capability of transformers to learn distributions in context are still lacking. Focusing on Gaussian process regression problems, we show by construction that transformers can implement a gradient descent algorithm targeting the posterior predictive mean and variance, followed by nonlinear mappings that yield binned probabilities of PPD. We study the error bounds of the approximated PPD in terms of attention depth and bin resolution. Based on these results, we further demonstrate the key role of normalization and the choice of attention depth in enabling the extrapolation abilities of transformers beyond the pretraining sample size range. We conduct simulations that corroborate our findings, providing insight into the expressivity of PFNs targeting PPDs and how architectural choices may influence generalization capabilities.
Marie Neubrander
In many applications, researchers may aim to understand how some intervention changes writing, from students’ essays to social media posts. Standard approaches in causal inference score each observed text, such as with a letter grade or sentiment rating, and estimate treatment effects on those scores. However, human evaluation is costly and automated models may be difficult to develop or not generalize across settings. Moreover, we may be interested in how individuals would perform on tasks beyond their observed writing. We propose using large language models as “thermometers” for text outcomes. We adapt copies of a common model to treatment and control writing, then compare their performance on fixed external benchmarks. Our primary target is the average benchmark effect we would obtain with abundant writing from each person. When few texts per person are available, we compare pooled model training with training one model per individual. We bound estimation error in terms of within-person variability, between-person heterogeneity, and benchmark curvature. Simulations examine how performance depends on the learning procedure; an application to a randomized study of AI-assisted practice examines the approach on real writing.
Private slide delivery; no public PDF is available.
Patrick Woitschig
In high-dimensional monitoring applications, changes often affect only a small and unknown subset of data streams. We develop a Bayesian framework for joint changepoint detection and localization that combines likelihood evidence across possible changepoints and affected subsets, incorporating structured priors to represent plausible change patterns. Our approach provides false-alarm guarantees and posterior inference on when and where changes occur. We establish theoretical results characterizing the benefits of pooling information across affected streams and the influence of prior specification on detection delay. We illustrate the methodology through numerical experiments with independent and spatially dependent observations, motivated by fault detection in CERN’s electromagnetic calorimeter.
Private slide delivery; no public PDF is available.
Li Fan
Radiological mapping aims to reconstruct spatial radiation activity from noisy measurements collected by moving detectors, such as drones. However, limited measurement budgets and increasingly high-resolution maps make accurate reconstruction and uncertainty quantification computationally challenging. In this talk, we propose BayeSTAR (Bayesian Structure-Aware Radiological mapping), a framework that combines low-rank matrix modeling, hierarchical shrinkage, and spatial smoothing to capture important features of radiation activity maps. The model is fitted using low-rank variational inference (VI), enabling efficient estimation and uncertainty quantification. Through synthetic experiments, we demonstrate that our approach achieves improved reconstruction accuracy and competitive probabilistic predictions compared with established benchmark methods, while showing promise for scaling to higher-resolution maps.
Private slide delivery; no public PDF is available.