BEGIN:VCALENDAR PRODID:BedeWork V3.5 VERSION:2.0 METHOD:PUBLISH BEGIN:VTIMEZONE TZID:America/New_York X-LIC-LOCATION:America/New_York BEGIN:DAYLIGHT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 TZNAME:EDT DTSTART:20070311T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0400 TZOFFSETTO:-0500 TZNAME:EST DTSTART:20071104T020000 RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU END:STANDARD BEGIN:STANDARD TZOFFSETFROM:-045602 TZOFFSETTO:-0500 TZNAME:EST DTSTART:18831118T120358 RDATE:18831118T120358 END:STANDARD BEGIN:DAYLIGHT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 TZNAME:EDT DTSTART:19180331T020000 RDATE:19180331T020000 RDATE:19190330T020000 RDATE:19200328T020000 RDATE:19210424T020000 RDATE:19220430T020000 RDATE:19230429T020000 RDATE:19240427T020000 RDATE:19250426T020000 RDATE:19260425T020000 RDATE:19270424T020000 RDATE:19280429T020000 RDATE:19290428T020000 RDATE:19300427T020000 RDATE:19310426T020000 RDATE:19320424T020000 RDATE:19330430T020000 RDATE:19340429T020000 RDATE:19350428T020000 RDATE:19360426T020000 RDATE:19370425T020000 RDATE:19380424T020000 RDATE:19390430T020000 RDATE:19400428T020000 RDATE:19410427T020000 RDATE:19460428T020000 RDATE:19470427T020000 RDATE:19480425T020000 RDATE:19490424T020000 RDATE:19500430T020000 RDATE:19510429T020000 RDATE:19520427T020000 RDATE:19530426T020000 RDATE:19540425T020000 RDATE:19550424T020000 RDATE:19560429T020000 RDATE:19570428T020000 RDATE:19580427T020000 RDATE:19590426T020000 RDATE:19600424T020000 RDATE:19610430T020000 RDATE:19620429T020000 RDATE:19630428T020000 RDATE:19640426T020000 RDATE:19650425T020000 RDATE:19660424T020000 RDATE:19670430T020000 RDATE:19680428T020000 RDATE:19690427T020000 RDATE:19700426T020000 RDATE:19710425T020000 RDATE:19720430T020000 RDATE:19730429T020000 RDATE:19740106T020000 RDATE:19750223T020000 RDATE:19760425T020000 RDATE:19770424T020000 RDATE:19780430T020000 RDATE:19790429T020000 RDATE:19800427T020000 RDATE:19810426T020000 RDATE:19820425T020000 RDATE:19830424T020000 RDATE:19840429T020000 RDATE:19850428T020000 RDATE:19860427T020000 RDATE:19870405T020000 RDATE:19880403T020000 RDATE:19890402T020000 RDATE:19900401T020000 RDATE:19910407T020000 RDATE:19920405T020000 RDATE:19930404T020000 RDATE:19940403T020000 RDATE:19950402T020000 RDATE:19960407T020000 RDATE:19970406T020000 RDATE:19980405T020000 RDATE:19990404T020000 RDATE:20000402T020000 RDATE:20010401T020000 RDATE:20020407T020000 RDATE:20030406T020000 RDATE:20040404T020000 RDATE:20050403T020000 RDATE:20060402T020000 END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0400 TZOFFSETTO:-0500 TZNAME:EST DTSTART:19181027T020000 RDATE:19181027T020000 RDATE:19191026T020000 RDATE:19201031T020000 RDATE:19210925T020000 RDATE:19220924T020000 RDATE:19230930T020000 RDATE:19240928T020000 RDATE:19250927T020000 RDATE:19260926T020000 RDATE:19270925T020000 RDATE:19280930T020000 RDATE:19290929T020000 RDATE:19300928T020000 RDATE:19310927T020000 RDATE:19320925T020000 RDATE:19330924T020000 RDATE:19340930T020000 RDATE:19350929T020000 RDATE:19360927T020000 RDATE:19370926T020000 RDATE:19380925T020000 RDATE:19390924T020000 RDATE:19400929T020000 RDATE:19410928T020000 RDATE:19450930T020000 RDATE:19460929T020000 RDATE:19470928T020000 RDATE:19480926T020000 RDATE:19490925T020000 RDATE:19500924T020000 RDATE:19510930T020000 RDATE:19520928T020000 RDATE:19530927T020000 RDATE:19540926T020000 RDATE:19551030T020000 RDATE:19561028T020000 RDATE:19571027T020000 RDATE:19581026T020000 RDATE:19591025T020000 RDATE:19601030T020000 RDATE:19611029T020000 RDATE:19621028T020000 RDATE:19631027T020000 RDATE:19641025T020000 RDATE:19651031T020000 RDATE:19661030T020000 RDATE:19671029T020000 RDATE:19681027T020000 RDATE:19691026T020000 RDATE:19701025T020000 RDATE:19711031T020000 RDATE:19721029T020000 RDATE:19731028T020000 RDATE:19741027T020000 RDATE:19751026T020000 RDATE:19761031T020000 RDATE:19771030T020000 RDATE:19781029T020000 RDATE:19791028T020000 RDATE:19801026T020000 RDATE:19811025T020000 RDATE:19821031T020000 RDATE:19831030T020000 RDATE:19841028T020000 RDATE:19851027T020000 RDATE:19861026T020000 RDATE:19871025T020000 RDATE:19881030T020000 RDATE:19891029T020000 RDATE:19901028T020000 RDATE:19911027T020000 RDATE:19921025T020000 RDATE:19931031T020000 RDATE:19941030T020000 RDATE:19951029T020000 RDATE:19961027T020000 RDATE:19971026T020000 RDATE:19981025T020000 RDATE:19991031T020000 RDATE:20001029T020000 RDATE:20011028T020000 RDATE:20021027T020000 RDATE:20031026T020000 RDATE:20041031T020000 RDATE:20051030T020000 RDATE:20061029T020000 END:STANDARD BEGIN:STANDARD TZOFFSETFROM:-0500 TZOFFSETTO:-0500 TZNAME:EST DTSTART:19200101T000000 RDATE:19200101T000000 RDATE:19420101T000000 RDATE:19460101T000000 RDATE:19670101T000000 END:STANDARD BEGIN:DAYLIGHT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 TZNAME:EWT DTSTART:19420209T020000 RDATE:19420209T020000 END:DAYLIGHT BEGIN:DAYLIGHT TZOFFSETFROM:-0400 TZOFFSETTO:-0400 TZNAME:EPT DTSTART:19450814T190000 RDATE:19450814T190000 END:DAYLIGHT END:VTIMEZONE BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20240919T145220Z DESCRIPTION:The frequentist variability of Bayesian posterior expectations can provide meaningful measures of uncertainty even when models are miss pecified. Classical methods to asymptotically approximate the frequentist covariance of Bayesian estimators such as the Laplace approximation and the nonparametric bootstrap can be practically inconvenient\, since the L aplace approximation may require an intractable integral to compute the m arginal log posterior\, and the bootstrap requires computing the posterio r for many different bootstrap datasets. We develop and explore the infin itesimal jackknife (IJ)\, an alternative method for computing asymptotic frequentist covariance of smooth functionals of exchangeable data\, which is based on the "influence function" of robust statistics. We show that the influence function for posterior expectations has the form of a simpl e posterior covariance\, and that the IJ covariance estimate is\, in turn \, easily computed from a single set of posterior samples. Under conditio ns similar to those required for a Bayesian central limit theorem to appl y\, we prove that the corresponding IJ covariance estimate is asymptotica lly equivalent to the Laplace approximation and the bootstrap. In the pre sence of nuisance parameters that may not obey a central limit theorem\, we argue using a von Mises expansion that the IJ covariance is inconsiste nt\, but can remain a good approximation to the limiting frequentist vari ance. We demonstrate the accuracy and computational benefits of the IJ co variance estimates with simulated and real-world experiments. DURATION:PT1H DTSTAMP:20240930T200850Z DTSTART;TZID=America/New_York:20241004T153000 LAST-MODIFIED:20240930T200850Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:The Bayesian Infinitesimal Jackknife for Variance UID:CAL-8a00048d-91324965-0192-0ac514c4-00004350demobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=karen.whitesell@duke.edu :Karen Whitesell X-BEDEWORK-SPEAKER:Ryan Giordano\, Berkeley X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar X-BEDEWORK-SUBMITTEDBY:kherndon for Statistical Science (agrp_StatisticalS cience) X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20240907T145742Z DESCRIPTION:Across the physical sciences\, there has been a shift in parad igm from a theory-driven to a data-driven era. In this new regime\, we le t the data speak for themselves by using modern machine learning tools un imaginable prior to the deep learning revolution of the last decade. At t he same time\, the physical sciences face unique challenges that require dedicated solutions to maximize the potential for discovery. Now\, more t han ever\, we need a new kind of researcher - a phystatistician (like bio statistician) or a data physicist (like data scientist). In this talk\, I 'll describe unique challenges faced by phystatisticians and how innovati ve\, reproducible\, and scalable methodologies and scientific software ar e enabling researchers to harness the power of modern machine learning fo r discoveries in the physical sciences. DURATION:PT1H DTSTAMP:20240907T145742Z DTSTART;TZID=America/New_York:20241018T153000 LAST-MODIFIED:20240907T145742Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Phystatistics: The Rise of the Data Physicist UID:CAL-8a00048d-91324965-0191-ccfdad8a-00004a7fdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-SPEAKER:Benjamin Nachman\, Staff Scientist\, Lawrence Berkeley National Laboratory X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar Series X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=Karen.whitesell@gmail.co m:Karen Whitesell X-BEDEWORK-SUBMITTEDBY:kherndon for Statistical Science (agrp_StatisticalS cience) END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20240930T200739Z DESCRIPTION:Mixtures of experts (MoEs)\, a class of statistical machine le arning models that combine multiple models\, known as experts\, to form m ore complex and accurate models\, have been combined into deep learning a rchitectures to improve the ability of these architectures and AI models to capture the heterogeneity of the data and to scale up these architectu res without increasing the computational cost. In mixtures of experts\, e ach expert specializes in a different aspect of the data\, which is then combined with a gating function to produce the final output. Therefore\, parameter and expert estimates play a crucial role by enabling statistici ans and data scientists to articulate and make sense of the diverse patte rns present in the data. However\, the statistical behaviors of parameter s and experts in a mixture of experts have remained unsolved\, which is d ue to the complex interaction between gating function and expert paramete rs.\n \nIn the first part of the talk\, we investigate the performance of the least squares estimators (LSE) under a deterministic MoEs model wher e the data are sampled according to a regression model\, a setting that h as remained largely unexplored. We establish a condition called strong id entifiability to characterize the convergence behavior of various types o f expert functions. We demonstrate that the rates for estimating strongly identifiable experts\, namely the widely used feed-forward networks with activation functions sigmoid(·) and tanh(·)\, are substantially faster t han those of polynomial experts\, which we show to exhibit a surprising s low estimation rate.\n \nIn the second part of the talk\, we show that th e insights from theories shed light into understanding and improving impo rtant practical applications in machine learning and artificial intellige nce (AI)\, including effectively scaling up massive AI models with severa l billion parameters\, efficiently finetuning large-scale AI models for d ownstream tasks\, and enhancing the performance of Transformer model\, st ate-of-the-art deep learning architecture\, with a novel self-attention m echanism. DURATION:PT1H DTSTAMP:20240930T200739Z DTSTART;TZID=America/New_York:20241025T153000 LAST-MODIFIED:20240930T200739Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:On Mixture of Experts in Large-Scale Statistical Machine Learning Applications UID:CAL-8a00048d-91324965-0192-448bb5e5-000066c0demobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-SPEAKER:Nhat Ho\, Assistant Professor\, The University of Texa s at Austin X-BEDEWORK-DUKE-SERIES:Statistical Science X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=karen.whitesell@duke.edu :Karen Whitesell X-BEDEWORK-SUBMITTEDBY:kherndon for Statistical Science (agrp_StatisticalS cience) END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20241029T194741Z DESCRIPTION:Regularization plays a pivotal role in tackling challenging il l-posed problems by guiding solutions towards desired properties. This ta lk covers three key problems---sparse signal recovery\, low-rank matrix c ompletion\, and robust principle component analysis (RPCA) for tensors--- all of which are linked to the $L_0$ regularization. As the $L_0$ model i s challenging to work with directly\, I will introduce two variants of th e popular $L_1$ norm for approximating the $L_0$ regularization: one know n as the transformed $L_1$ (TL1)\, and the other as the ratio of the $L_1 $ and $L_2$ norms\, denoted by $L_1/L_2$. Our theoretical analysis establ ishes the local optimality of the $L_1/L_2$ models based on the condition s that are analogous to those for the convex $L_1$ and nuclear norm metho ds. Additionally\, we provide a statistical analysis of the estimator der ived from a TL1-regularized matrix completion under a general sampling di stribution\, rather than the classic uniform sampling. Our results show t hat the model achieves a convergence rate\, comparable to that of the nuc lear norm-based model\, despite the challenges posed by the non-convexity of TL1. In this talk\, I will also present a range of real-world applic ations\, including limited-angle CT reconstruction and video background m odeling\, demonstrating the superior performance of these two $L_1$ varia nts compared to state-of-the-art methods. DURATION:PT1H DTSTAMP:20241107T180834Z DTSTART;TZID=America/New_York:20241108T153000 LAST-MODIFIED:20241107T180834Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Two Tales of L1 Variants for Sparse Signal and Low-rank Tensor Rec overy UID:CAL-8a000483-92c3adf6-0192-d9d1daae-00004a9bdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=stat-staff214@duke.edu:S tatSci Staff X-BEDEWORK-SUBMITTEDBY:kherndon for Statistical Science (agrp_StatisticalS cience) X-BEDEWORK-SPEAKER:Yifei Lou\, Associate Professor\, School of Data Scien ce and Society\, The University of North Carolina X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20241001T153525Z DESCRIPTION:In many fields\, including medicine\, education\, and public p olicy\, researchers and practitioners are interested in determining what works for whom - in other words\, identifying subgroups for whom specific interventions work particularly well. By finding these subgroups\, we ca n more effectively focus resources and give interventions to those who mi ght benefit the most. These individualized treatment decisions can improv e outcomes\, but answering these types of questions is challenging with a single dataset. Specifically\, randomized controlled trials have unconfo unded treatment assignment but are often too small to reliably estimate h eterogeneous treatment effects\, while larger observational datasets migh t suffer from confounding. Data integration methods can utilize the benef its of different sources of data while accounting for bias. In this talk\ , I first discuss non-parametric data integration approaches for combinin g multiple randomized controlled trials to estimate the effect of treatme nts conditional on observed characteristics. I explore the performance of these methods through a simulation study\, and I apply the approaches to four randomized controlled trials to examine effect heterogeneity of tre atments for major depressive disorder. I then discuss methods for applyin g these multi-study treatment effect models to an external\, observationa l target sample represented by electronic health records of a set of pati ents. With these methods\, we can utilize individual-level data across so urces to improve our ability to make intervention decisions that are tail ored to individuals or communities\, and we can ultimately apply our conc lusions to a given target population. DURATION:PT1H DTSTAMP:20241001T153525Z DTSTART;TZID=America/New_York:20241115T153000 LAST-MODIFIED:20241001T153525Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Data integration approaches to estimate heterogeneous treatment ef fects UID:CAL-8a00048d-91324965-0192-48b8d418-00000525demobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-SPEAKER:Carly Brantner\, Assistant Professor of Biostatistics & Bioinformatics\, Duke University X-BEDEWORK-DUKE-SERIES:Statistical Science X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=karen.whitesell@duke.edu :Karen Whitesell X-BEDEWORK-SUBMITTEDBY:kherndon for Statistical Science (agrp_StatisticalS cience) END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20241029T202322Z DESCRIPTION:For several decades\, political scientists have collected data sets of "dyadic events"-i.e.\, micro-records of the form "country i took action a to county j at time t". Such data sets provide an expansive and systematized view of the world that prompts data-driven approaches to th e study of international relations. \n \nHowever\, despite all the work t o collect massive event data sets\, there has been comparatively little w ork in political science to actually use them. This is partly due to the presence of "complex dependence structures" in the data\, which violate t he independence assumptions of many methods in the standard statistical t oolkit.\nIn this talk\, I will discuss a family of Bayesian models for me asuring complex dependence structure in dyadic events. These models blend aspects of tensor decomposition\, dynamical systems\, and discrete admix tures to capture rich multilayer network structure and excitatory tempora l dynamics in country-to-country interactions. While inspired by internat ional relations\, these models are tailored to the general statistical pr operties of sparse and high-dimensional discrete data and are widely appl icable to problems where such data sets arise. DURATION:PT1H DTSTAMP:20241029T202902Z DTSTART;TZID=America/New_York:20241122T153000 LAST-MODIFIED:20241029T202902Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Probabilistic Tensor Decomposition Model for Measuring Complex Dep endence Structure in Sparse Dyadic Event Data UID:CAL-8a000483-92c3adf6-0192-d9f2849d-00004d6ademobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=karen.whitesell@duke.edu :Karen Whitesell X-BEDEWORK-SPEAKER:Aaron Schein\, Assistant Professor\, Department of Stat istics and Data Science Institute\, University of Chicago X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar X-BEDEWORK-SUBMITTEDBY:kherndon for Statistical Science (agrp_StatisticalS cience) END:VEVENT BEGIN:VEVENT CATEGORIES:Natural Sciences CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20250108T200720Z DESCRIPTION:This talk introduces a novel framework for multitask learning in high-dimensional time series through hypothesis testing and data integ ration. The proposed procedure tests for shared structures across multipl e high-dimensional factor models\, determining whether they are driven by the same loading vectors up to a linear transformation. Leveraging repea ted applications of singular value decomposition\, the framework achieves consistent estimation of shared and non-shared loading vectors and intro duces a sequential testing procedure to estimate the number of shared com ponents. Theoretical results establish the asymptotic behavior of the tes t statistics and consistency. The method applies to multitask frameworks to uncover inter-individual relationships between datasets and to detecti ng structural changes over time in a single factor model. Applications to macroeconomic data demonstrate the framework's practical utility in reve aling shared and distinct structures across datasets. DURATION:PT1H DTSTAMP:20250108T200720Z DTSTART;TZID=America/New_York:20250113T153000 LAST-MODIFIED:20250108T200720Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Multitask Learning for High-dimensional Time Series UID:CAL-8a000483-92c3adf6-0194-47875a05-000029b2demobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Natural Sciences:/user/publi c-user/Topics/Natural Sciences X-BEDEWORK-SPEAKER:Marie-Christine Dueker\, Assistant Professor\, Mathemat ical Statistics & Data Science\, University of Erlangen-Nuremburg X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=statsci-seminar-coordina tor@duke.edu:Seminar Coordinator X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y2:353 X-BEDEWORK-IMAGE-ALT-TEXT:Marie-Christine Dueker X-BEDEWORK-SUBMITTEDBY:tnscott for Statistical Science (agrp_StatisticalSc ience) X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:529.5 X-BEDEWORK-IMAGE-CROP-WIDTH:529.5 X-BEDEWORK-IMAGE-CROP-HEIGHT:353 X-BEDEWORK-IMAGE:/public/Images/Dueker-jan13_20250108080625PM.jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/Dueker-jan13_20250108080625PM-thumb. png END:VEVENT BEGIN:VEVENT CATEGORIES:Natural Sciences CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20250108T202036Z DESCRIPTION:We investigate inference for a single coordinate in high-dimen sional linear regression\, focusing on settings where there is a single p arameter of interest (such as a treatment effect) in the presence of nume rous potential confounding covariates. A common Bayesian procedure in suc h scenarios is to model the nuisance covariates using a model selection p rior\, which encourages sparsity in the regression coefficients. \n\nThis work characterizes the behavior of the model selection posterior distrib ution for the coordinate of interest as sample size grows. Under certain conditions\, we establish that the posterior marginal distribution achiev es optimal inferential properties through a Bernstein-von Mises theorem: it converges to a normal distribution centered at an efficient oracle est imator with optimal variance. Strikingly\, such performance is attainable under conditions where other priors or frequentist procedures such as th e LASSO require debiasing. \n\nHowever\, we also identify settings where the posterior marginal exhibits problematic limiting behavior\; convergin g to a multimodal mixture with components that contain substantial bias a nd/or suboptimal variance\, resulting in poor coverage. We propose adjust ments to the prior specification that provably restore desirable asymptot ic properties whenever such issues arise. DURATION:PT1H DTSTAMP:20250108T211042Z DTSTART;TZID=America/New_York:20250117T153000 LAST-MODIFIED:20250108T211042Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Bayesian Linear Regression with a Sparse Prior: Inference for a Si ngle Coordinate UID:CAL-8a000483-92c3adf6-0194-47937fe4-00002ad6demobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Natural Sciences:/user/publi c-user/Topics/Natural Sciences X-BEDEWORK-SPEAKER:Lasse Vuursteen\, Postdoctoral Scholar\, University of Pennsylvania\, Wharton School of Business X-BEDEWORK-SUBMITTEDBY:tnscott for Statistical Science (agrp_StatisticalSc ience) X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:-0.3333333333333144 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Lasse Vuursteen X-BEDEWORK-IMAGE:/public/Images/Lasse-1.17_20250108091042PM.jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/Lasse-1.17_20250108091042PM-thumb.pn g END:VEVENT BEGIN:VEVENT CATEGORIES:Natural Sciences CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo ri CREATED:20250120T200357Z DESCRIPTION:We derive an asymptotic expansion of posterior integrals in th e regime in which dimension grows together with sample size. We also pres ent related work on the accuracy of the Laplace approximation (LA) to hig h-dimensional posterior densities\, and derive a higher-order correction to the LA. These results are both theoretically significant and useful fo r the computations involved e.g. in Bayesian model selection and construc tion of credible sets. Finally\, we prove the tightest known high-dimensi onal Bernstein-von Mises theorem\, closing the long-standing gap between conditions for asymptotic normality in Bayesian and frequentist inference . \n\nOur expansion of posterior integrals\, which are naturally of Lapla ce type for large sample size\, is also of theoretical significance in as ymptotic analysis. It fills the gap in the theory between the classical f ixed-dimensional regime dating back to Laplace\, and more recent work on the asymptotic expansion of infinite-dimensional Laplace-type integrals d ue to Ben Arous. DURATION:PT1H DTSTAMP:20250120T200357Z DTSTART;TZID=America/New_York:20250122T153000 LAST-MODIFIED:20250120T200357Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Asymptotics of High-Dimensional Bayesian Inference UID:CAL-8a000483-92c3adf6-0194-855092a3-000031cbdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Natural Sciences:/user/publi c-user/Topics/Natural Sciences X-BEDEWORK-SPEAKER:Anya Katsevich X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc e) END:VEVENT BEGIN:VEVENT CATEGORIES:Natural Sciences CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo ri CREATED:20250120T202050Z DESCRIPTION:Dietary patterns are essential for understanding dietary behav iors and their health implications in nutritional epidemiology\, yet comp lexities in dietary assessments pose analytical challenges. Heterogeneity in dietary behaviors across populations and the multivariate nature of d ietary assessment data underscore the need for latent variable models to uncover meaningful patterns. However\, estimation of dietary patterns fac es challenges of strong similarities and highly correlated acculturation exposure measures. Overlooking these challenges often results in numerica l instability and inaccuracy that hinder scientific interpretation. We pr opose novel latent variable models to address these challenges by incorpo rating tree regularization and a multilayered modeling framework. Through studies of migrant populations in the US\, we discuss improved identific ation of nuanced differences in dietary patterns in small subpopulations\ , and the interplay between acculturation and dietary behaviors. We also highlight insights into public health and nutrition interventions. DURATION:PT1H DTSTAMP:20250120T202050Z DTSTART;TZID=America/New_York:20250124T153000 LAST-MODIFIED:20250120T202050Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Latent Variable Models for Advancing Nutrition Epidemiology UID:CAL-8a000483-92c3adf6-0194-856006fc-000031ccdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Natural Sciences:/user/publi c-user/Topics/Natural Sciences X-BEDEWORK-SPEAKER:Mengbing Li X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc e) END:VEVENT BEGIN:VEVENT CATEGORIES:Natural Sciences CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo ri CREATED:20250120T202428Z DESCRIPTION:The success of Bayesian inference with MCMC depends critically on Markov chains rapidly reaching the posterior distribution. Despite th e plentitude of inferential theory for posteriors in Bayesian non-paramet rics\, convergence properties of MCMC algorithms that simulate from such ideal inferential targets\, are not thoroughly understood. This work focu ses on the Bayesian CART algorithm\, which forms a building block of Baye sian Additive Regression Trees (BART). We derive upper bounds on mixing t imes for typical posteriors under various proposal distributions. Exploit ing the wavelet representation of trees\, we provide sufficient condition s for Bayesian CART to mix well (polynomially) under certain hierarchical connectivity restrictions on the signal. We also derive a negative resul t showing that Bayesian CART (based on simple grow and prune steps) canno t reach deep isolated signals in faster than superpolynomial mixing time. To remediate myopic tree exploration\, we propose Twiggy Bayesian CART\, which attaches/detaches entire twigs (not just single nodes) in the prop osal distribution. We show polynomial mixing of Twiggy Bayesian CART with out assuming that the signal is connected on a tree. Going further\, we s how that informed variants achieve even faster mixing. A thorough simulat ion study highlights discrepancies between spike-and-slab priors and Baye sian CART under a variety of proposals. DURATION:PT1H DTSTAMP:20250120T202428Z DTSTART;TZID=America/New_York:20250127T153000 LAST-MODIFIED:20250120T202428Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:On Mixing Rates for Bayesian CART UID:CAL-8a000483-92c3adf6-0194-85635cc6-000031cddemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Natural Sciences:/user/publi c-user/Topics/Natural Sciences X-BEDEWORK-SPEAKER:Jungeum Kim X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc e) END:VEVENT BEGIN:VEVENT CATEGORIES:Europe focus CATEGORIES:Global CATEGORIES:Other CATEGORIES:Utilities CATEGORIES:Information Session CATEGORIES:Main CATEGORIES:Student CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20250114T174847Z DESCRIPTION:Join the Duke in Berlin Resident Director to learn more about studying in Berlin this summer! No German language experience required! DURATION:PT30M DTSTAMP:20250114T174847Z DTSTART;TZID=America/New_York:20250129T170000 LAST-MODIFIED:20250114T174847Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Duke in Berlin Summer Info Session UID:CAL-8a000483-92c3adf6-0194-65eeaa9d-00001853demobedework@mysite.edu URL:https://my.globaled.duke.edu/_portal/tds-program-brochure?programid=10 198 X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Student:/user/public-user/Ut ilities/Student X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Information Session:/user/pu blic-user/Other/Information Session X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Global:/user/public-user/Top ic of Event Focused on a Country or Continent (if applicable)/Global X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Europe focus:/user/public-us er/Topic of Event Focused on a Country or Continent (if applicable)/Europ e focus X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=globaled@duke.edu:Global Education Office X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y2:353 X-BEDEWORK-IMAGE-ALT-TEXT:Duke in Berlin Summer 2025 Info Session X-BEDEWORK-SUBMITTEDBY:des52 for Global Education Office for Undergraduate s (agrp_ArtsandSciences_GlobalEd) X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:529.5 X-BEDEWORK-IMAGE-CROP-WIDTH:529.5 X-BEDEWORK-IMAGE-CROP-HEIGHT:353 X-BEDEWORK-IMAGE:/public/Images/DukeinBerlinSummerEventCalendar_2025011405 4828PM.png X-BEDEWORK-THUMB-IMAGE:/public/Images/DukeinBerlinSummerEventCalendar_2025 0114054828PM-thumb.png X-BEDEWORK-CS;X-BEDEWORK-PARAM-DESCRIPTION="/principals/users/agrp__Artsan dSciences_German,":German Studies END:VEVENT BEGIN:VEVENT CATEGORIES:Europe focus CATEGORIES:Global CATEGORIES:Other CATEGORIES:Utilities CATEGORIES:Information Session CATEGORIES:Main CATEGORIES:Student CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20250114T174454Z DESCRIPTION:Join the Duke in Berlin Resident Director to learn more about studying in Berlin in Fall 2025! No German language experience required! DURATION:PT1H DTSTAMP:20250114T174454Z DTSTART;TZID=America/New_York:20250129T173000 LAST-MODIFIED:20250114T174454Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Duke in Berlin Fall 2025 Info Session UID:CAL-8a000483-92c3adf6-0194-65eb1ca3-00001852demobedework@mysite.edu URL:https://my.globaled.duke.edu/_portal/tds-program-brochure?programid=10 196 X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Student:/user/public-user/Ut ilities/Student X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Information Session:/user/pu blic-user/Other/Information Session X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Global:/user/public-user/Top ic of Event Focused on a Country or Continent (if applicable)/Global X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Europe focus:/user/public-us er/Topic of Event Focused on a Country or Continent (if applicable)/Europ e focus X-BEDEWORK-CS;X-BEDEWORK-PARAM-DESCRIPTION="/principals/users/agrp__Artsan dSciences_German,":German Studies X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=globaled@duke.edu:Global Education Office X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:-0.3333333333333144 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Duke in Berlin Fall 2025 Info Session X-BEDEWORK-SUBMITTEDBY:des52 for Global Education Office for Undergraduate s (agrp_ArtsandSciences_GlobalEd) X-BEDEWORK-IMAGE:/public/Images/DukeinBerlinFallEventCalendar_202501140544 54PM.png X-BEDEWORK-THUMB-IMAGE:/public/Images/DukeinBerlinFallEventCalendar_202501 14054454PM-thumb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Natural Sciences CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo ri CREATED:20250120T202727Z DESCRIPTION:Feature/variable selection is a fundamental technique in data science that aims to identify the relevant features in a dataset. A criti cal component of feature selection is false positive control. Several pop ular methods---including the classical Benjamini-Hochberg procedure\, sta bility selection\, and\, most recently\, knockoffs---address this challen ge\, but often at the expense of identifying few true positives. In this talk\, I will introduce a new version of stability selection\, integrated path stability selection (IPSS)\,that yields significantly more true pos itives in practice than existing methods while still controlling false po sitives at desired rates. Furthermore\, IPSS is computationally efficient \, easy to implement\, and effective in high dimensions. It also offers p arametric and nonparametric versions\, with the latter capable of capturi ng nonlinear relationships in data. After introducing the method\, I will demonstrate its performance with applications to cancer data DURATION:PT1H DTSTAMP:20250120T202727Z DTSTART;TZID=America/New_York:20250131T153000 LAST-MODIFIED:20250120T202727Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Integrated Path Stability Selection UID:CAL-8a000483-92c3adf6-0194-856615a4-000031cedemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Natural Sciences:/user/publi c-user/Topics/Natural Sciences X-BEDEWORK-SPEAKER:Omar Melikechi X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc e) END:VEVENT BEGIN:VEVENT CATEGORIES:Natural Sciences CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo ri CREATED:20250120T203123Z DESCRIPTION:Sampling from a target distribution is a recurring theme in st atistics and generative artificial intelligence (AI). In Bayesian statist ics\, posterior sampling offers a flexible inferential framework\, enabli ng uncertainty quantification\, probabilistic prediction\, as well as the estimation of intractable quantities. In generative AI\, sampling aims t o generate unseen instances that emulate a target population\, such as th e natural distributions of texts\, images\, and molecules. \nIn this talk \, I will present my works on designing provably efficient sampling algor ithms\, addressing challenges in both statistics and generative AI. (1) I n the first part\, I will focus on posterior sampling for Bayes sparse re gression. In general\, such posteriors are high-dimensional and contain m any modes\, making them challenging to sample from. To address this\, we develop a novel sampling algorithm based on decomposing the target poster ior into a log-concave mixture of simple distributions\, reducing samplin g from a complex distribution to sampling from a tractable log-concave on e. We establish provable guarantees for our method in a challenging regim e that was previously intractable. (2) In the second part\, I will descri be a training-free acceleration method for diffusion models\, which are d eep generative models that underpin cutting-edge applications such as Alp haFold\, DALL-E and Sora. Our approach is simple to implement\, wraps aro und any pre-trained diffusion model\, and comes with a provable convergen ce rate that strengthens prior theoretical results. We demonstrate the ef fectiveness of our method on several real-world image generation tasks. \nLastly\, I will outline my vision for bridging the fields of statistics and generative AI\, exploring how insights from one domain can drive pro gress in the other. DURATION:PT1H DTSTAMP:20250120T203123Z DTSTART;TZID=America/New_York:20250203T153000 LAST-MODIFIED:20250120T203123Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Modern Sampling Paradigms: from Posterior Sampling to Generative A I UID:CAL-8a000483-92c3adf6-0194-8569b032-000031cfdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Natural Sciences:/user/publi c-user/Topics/Natural Sciences X-BEDEWORK-SPEAKER:Yuchen Wu X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc e) END:VEVENT BEGIN:VEVENT CATEGORIES:Natural Sciences CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo ri CREATED:20250120T203537Z DESCRIPTION:Verbal autopsy (VA) algorithms are routinely employed in low-a nd middle-income countries to determine individual causes of death (COD)\ , which are then aggregated to estimate population-level cause-specific m ortality fractions (CSMFs) essential for public health policymaking. Howe ver\, VA algorithms often misclassify COD\, introducing bias in CSMF esti mates. A recent method\, VA-calibration\, addresses this bias by utilizin g a VA misclassification rate matrix derived from limited labeled COD dat a collected in the CHAMPS project. Due to limited labeled samples\, the d ata are pooled across countries to improve estimation precision\, thereby implicitly assuming uniform misclassification rates. In this presentatio n\, I will highlight substantial cross-country heterogeneity in VA miscla ssification\, challenging this homogeneity assumption and revealing its i mpact on VA-calibration bias. To address this\, I will propose a comprehe nsive country-specific VA misclassification matrix modeling framework in data-scarce settings. The framework introduces a novel base model that pa rsimoniously characterizes the misclassification matrix through two laten t mechanisms: intrinsic accuracy and systematic preference. We theoretica lly prove that these mechanisms are identifiable from the data and manife st as a form of invariance in misclassification odds\, a pattern evident in the CHAMPS data. Building on this\, the framework then incorporates cr oss-country heterogeneity through interpretable effect sizes and uses shr inkage priors to balance the bias-variance tradeoff in misclassification matrix estimation. This effort broadens VA-calibration's future applicabi lity and strengthens ongoing efforts of using VA for mortality surveillan ce. I will illustrate this through applications to mortality surveillance projects\, such as COMSA in Mozambique and CA CODE. DURATION:PT1H DTSTAMP:20250120T203537Z DTSTART;TZID=America/New_York:20250207T153000 LAST-MODIFIED:20250120T203537Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Bayesian Modeling of Misclassification Matrices for Improved Verba l Autopsy-Based Mortality Estimates in LMICs UID:CAL-8a000483-92c3adf6-0194-856d8f38-000031d0demobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Natural Sciences:/user/publi c-user/Topics/Natural Sciences X-BEDEWORK-SPEAKER:Sandipan Pramanik X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc e) END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Main CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo ri CREATED:20250212T180544Z DESCRIPTION:A major challenge in the age of Big Data is the integration of disparate data types into a single data analysis. That is tackled here in the context of data blocks measured on a common set of experimental ca ses. Joint variation is defined in terms of modes of variation having id entical scores across data blocks. That allows mathematically rigorous f ormulation of individual variation within each data block in terms of ind ividual modes. These are mathematically defined through modes of variati on with common scores. DIVAS improves earlier methods using a novel rand om direction approach to statistical inference\, and by treating partiall y shared blocks. Usefulness is illustrated using mortality\, cancer and neuroimaging data sets. DURATION:PT1H DTSTAMP:20250213T140652Z DTSTART;TZID=America/New_York:20250221T153000 LAST-MODIFIED:20250213T140652Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Data Integration Via Analysis of Subspaces (DIVAS) UID:CAL-8a000483-92c3adf6-0194-fb569b8a-00002a3ademobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-SPEAKER:J.S. Marron\, Prof of Biostatistics\, School of Data Sc ience & Society\, UNC-Chapel Hill X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc e) END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Main CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo ri CREATED:20250212T181625Z DESCRIPTION:The denoising diffusion probabilistic model (DDPM) has become a cornerstone of generative AI. While sharp convergence guarantees have b een established for DDPM\, the iteration complexity typically scales with the ambient data dimension of target distributions\, leading to overly c onservative theory that fails to explain its practical efficiency. This h as sparked recent efforts to understand how DDPM can achieve sampling spe ed-ups through automatic exploitation of intrinsic low dimensionality of data.\n\nThis talk explores two key scenarios: (1) For a broad class of d ata distributions with intrinsic dimension k\, we prove that the iteratio n complexity of the DDPM scales nearly linearly with k\, which is optimal under the KL divergence metric\; (2) For mixtures of Gaussian distributi ons with k components\, we show that DDPM learns the distribution with it eration complexity that grows only logarithmically in k. These results pr ovide theoretical justification for the practical efficiency of diffusion models.\n\nDr. Yuting Wei is currently an Assistant Professor in the Sta tistics and Data Science Department at the Wharton School\, University of Pennsylvania. Prior to that\, Dr. Wei spent two years at Carnegie Mellon University as an assistant professor and one year at Stanford University as a Stein's Fellow. She received her Ph.D. in statistics at the Univers ity of California\, Berkeley. She received the 2023 Google Research Schol ar Award\, 2022 NSF Career award\, and the Erich L. Lehmann Citation from the Berkeley statistics department. Her research interests include high- dimensional and non-parametric statistics\, reinforcement learning\, and diffusion models. DURATION:PT1H DTSTAMP:20250213T191835Z DTSTART;TZID=America/New_York:20250228T153000 LAST-MODIFIED:20250213T191835Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:To Intrinsic Dimension and Beyond: Efficient Sampling in Diffusion Models UID:CAL-8a000483-92c3adf6-0194-fb6061b5-00002afedemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-SPEAKER:Yuting Wei\, Asst Prof Statistics & Data Science\, Univ ersity of Pennsylvania\, Wharton School of Business X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc e) END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo ri CREATED:20250410T125416Z DESCRIPTION:Random partitions are fundamental probabilistic objects in Bay esian statistics\, particularly in nonparametric models\, as they enable flexible clustering and relax strong distributional assumptions about the data. Exchangeable partitions arise naturally in Dirichlet process mixtu res\, allowing the number of clusters to grow with the data\, which are a lso assumed to be exchangeable. More general forms\, such as partially ex changeable partitions and product partition models with covariates\, exte nd their applicability to settings where\, alongside the data\, a set of covariates is also available. These models further highlight the connecti on between the symmetry assumptions imposed on the data-corresponding to the adopted experimental design-and those reflected in the law governing the partition.\n\nA less explored experimental design for random partitio n models is that of repeated measurements\, which has gained particular a ttention only in recent years. This setting arises when the inferential p roblem requires estimating a collection of partitions of the same items\, commonly referred to as multi-view clustering models. These models shoul d incorporate temporal dynamics or separate exchangeability assumptions w ithin random partition frameworks.\n\nThis talk will provide an overview of recent advancements in this area\, with a particular focus on conditio nal partial exchangeability (CPE)\, a unifying dependence condition for c onstructing dependent partitions of the same objects. CPE differs from tr aditional partial exchangeability due to its conditional nature and its r equirement for marginal invariance. Together\, these conditions ensure lo cal dependence at the level of the items across partitions\, aligning the symmetry assumptions in the partition law with the experimental design o f observing multiple instances corresponding to the same items (i.e.\, re peated measures design). The theoretical and modeling advancements of CPE will be illustrated through two real data applications: clustering metab olomics data and analyzing diffusion tensor imaging data. DURATION:PT1H DTSTAMP:20250410T162329Z DTSTART;TZID=America/New_York:20250411T153000 LAST-MODIFIED:20250410T162329Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Random Partitions for Multi-view Data: How to Encode Repeated Meas ures Design into Nonparametric Bayesian Models. UID:CAL-8a000483-92c3adf6-0196-1fc3ee1c-00005d8ademobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-SPEAKER: Beatrice Franzolini\, Jr. Asst. Professor\, Institute for Data Science & Analytics\, Bocconi University X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc e) X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Beatrice Franzolini X-BEDEWORK-IMAGE:/public/Images/beatrice-sized_20250410030332PM.jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/beatrice-sized_20250410030332PM-thum b.png END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo ri CREATED:20250415T133939Z DESCRIPTION:Min-norm interpolators naturally emerge as implicit regularize d limits of modern machine learning algorithms. Recently\, their out-of-d istribution risk was studied when test samples are unavailable during tra ining. However\, in many applications\, a limited amount of test data is typically available during training. The properties of min-norm interpola tion in this setting are not well understood. In this talk\, I will prese nt a characterization of the generalization error of pooled min-L2-norm i nterpolation under covariate and model shifts. I will demonstrate that th e pooled interpolator captures both early fusion and a form of intermedia te fusion. Our results have several implications. Under model shift\, add ing data always hurts prediction when the signal-to-noise ratio is low. H owever\, for higher signal-to-noise ratios\, transfer learning helps as l ong as the shift-to-signal ratio lies below a threshold that I will defin e. I will derive precise thresholds capturing when the pooled interpolato r outperforms the target-based interpolator\, and further characterize th e optimal number of target samples that minimizes the generalization erro r. Our results also show that under covariate shift\, if the source sampl e size is small relative to the dimension\, heterogeneity between domains improves the risk. This is based on joint work with Yanke Song and Soho m Bhattacharya. DURATION:PT1H DTSTAMP:20250416T193849Z DTSTART;TZID=America/New_York:20250418T153000 LAST-MODIFIED:20250416T193849Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Quantifying the Effects of Transfer Learning in Min-norm Interpola tion UID:CAL-8a000483-92c3adf6-0196-39ad480c-00002ebfdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-SPEAKER:Pragya Sur\, Assistant Professor of Statistics\, Harvar d University X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc e) X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:12 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:365.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Pragya Sur X-BEDEWORK-IMAGE:/public/Images/Calendar-photos_20250416073849PM.jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/Calendar-photos_20250416073849PM-thu mb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo ri CREATED:20250410T123555Z DESCRIPTION:Abstract: Instrumental variables are a popular tool to infer c ausal effects under unobserved confounding\, but choosing suitable instru ments is challenging in practice. We propose gIVBMA\, a Bayesian model av eraging procedure that addresses this challenge by averaging across diffe rent sets of instrumental variables and covariates in a structural equati on model. Our approach extends previous work through a scale-invariant pr ior structure and accommodates non-Gaussian outcomes and treatments\, off ering greater flexibility than existing methods. The computational strate gy uses conditional Bayes factors to update models separately for the out come and treatments. We prove that this model selection procedure is cons istent. By explicitly accounting for model uncertainty\, gIVBMA allows in struments and covariates to switch roles and provides robustness against invalid instruments. In simulated and real data experiments\, gIVBMA outp erforms current state-of-the-art methods. A software implementation of gI VBMA is available in Julia. An application to returns to education will b e discussed. DURATION:PT1H30M DTSTAMP:20250410T150745Z DTSTART;TZID=America/New_York:20250421T113000 LAST-MODIFIED:20250410T150745Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Bayesian Model Averaging in Causal Instrumental Variable Models UID:CAL-8a000483-92c3adf6-0196-1fb320fa-00005b3fdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-SPEAKER:Mark Steel\, Professor of Statistics\, University of Wa rwick X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc e) X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Mark Steel X-BEDEWORK-IMAGE:/public/Images/Mark.Steel-sized_20250410030745PM.jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/Mark.Steel-sized_20250410030745PM-th umb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo ri CREATED:20250415T134251Z DESCRIPTION:Accurate multimodal prediction-spanning tabular\, textual\, an d visual data-is crucial for advancing analytics across diverse domains. However\, traditional models often struggle to integrate heterogeneous da ta while preserving high predictive accuracy. In this talk\, we present G enerative Distribution Prediction\, a flexible framework that enhances pr edictive performance through multimodal synthetic data generation\, inclu ding conditional diffusion models. This framework facilitates transfer le arning\, adapts to various loss functions for risk minimization\, and pro vides statistical guarantees on predictive accuracy. We empirically valid ate its versatility and effectiveness across four supervised tasks: tabul ar data prediction\, question answering\, image captioning\, and adaptive quantile regression.\n\nJoint work with Dr. Xinyu Tian\, School of Stati stics\, University of Minnesota. DURATION:PT1H DTSTAMP:20250416T191022Z DTSTART;TZID=America/New_York:20250425T153000 LAST-MODIFIED:20250416T191022Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Generative Distribution Prediction for Multimodal Learning UID:CAL-8a000483-92c3adf6-0196-39b034ee-00002ec0demobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-SPEAKER:Xiaotong Shen\, Professor of Statistics\, University of Minnesota X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc e) X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:-0.3333333333333144 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Xiaotong Shen X-BEDEWORK-IMAGE:/public/Images/Calendar-photos_20250416071022PM.jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/Calendar-photos_20250416071022PM-thu mb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Main CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo ri CREATED:20250807T142830Z DESCRIPTION:Modeling autocovariance functions (ACFs) is fundamental in tim e-series\, spatial\, and spatio-temporal statistics. Standard parametric models can struggle with complex\, non-separable dependence\, and non-par ametric approaches must ensure the ACF remains positive semi-definite. In this talk\, I present a new family of non-parametric\, closed-form ACFs that are provably dense in a broad class of continuous processes\, provid e optimally efficient functional representations\, and extend naturally t o multivariate and multidimensional settings. By avoiding rigid assumptio ns such as separability\, the method captures realistic space-time intera ctions and can handle irregularly observed data. Illustrations are given from oceanographic spatio-temporal applications. DURATION:PT1H DTSTAMP:20250812T204115Z DTSTART;TZID=America/New_York:20250814T153000 LAST-MODIFIED:20250812T204115Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:The Fast and the Fourier-ous: Non-Parametric Autocovariance Modeli ng with Spline Kernels UID:CAL-8a00ec8b-979413b9-0198-84eefaf6-00000c29demobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-SPEAKER:Lacchlan Astfalck\, Research Fellow\, University of Wes tern Australia\, School of Physics\, Mathematics & Computing X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc e) X-BEDEWORK-IMAGE-X1:533 X-BEDEWORK-IMAGE-Y1:40 X-BEDEWORK-IMAGE-X2:533 X-BEDEWORK-IMAGE-Y2:40 X-BEDEWORK-IMAGE-CROP-WIDTH:0 X-BEDEWORK-IMAGE-CROP-HEIGHT:0 X-BEDEWORK-IMAGE-ALT-TEXT:Lachlan Astfalck headshot END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Main CONTACT;X-BEDEWORK-UID=8a00ec8b-979413b9-0198-a4d89b28-00001675:Thompson\, Ekaterina CREATED:20250903T130026Z DESCRIPTION:We propose an algorithm that iteratively transforms a target d istribution\, specified by an unnormalized density\, into a standard Gaus sian. At each iteration\, the target is rotated to make the coordinates a s independent as possible\, after which a mean-field variational inferenc e step is applied to bring each marginal closer to a standard Gaussian. T he effectiveness of each iteration depends critically on the choice of ro tation. We show that a principled choice arises from the principal compon ents of a covariance matrix formed from the relative score function of th e target\, leading to a natural PCA-type procedure. The resulting sequenc e of transformations progressively maps the target to a Gaussian\, while the inverse transformation can be used to generate samples from the targe t. We establish convergence guarantees for Gaussian targets and demonstra te the effectiveness of the algorithm through numerical experiments on po sterior sampling tasks. Compared to applying mean-field variational infer ence in the standard coordinate axes\, using the proposed principal compo nent axes yields substantial accuracy gains with negligible computational overhead. Compared to conventional normalizing flows\, our approach achi eves comparable flexibility with far fewer parameters and lower training cost. DURATION:PT1H DTSTAMP:20250904T131646Z DTSTART;TZID=America/New_York:20250905T153000 LAST-MODIFIED:20250904T131646Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Sampling via Iterative Gaussianization UID:CAL-8a00ec8b-979413b9-0199-0faa0e8d-00003fbfdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-SPEAKER:Sifan Liu X-BEDEWORK-DUKE-SERIES:Friday Seminar in the Department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) X-BEDEWORK-IMAGE-X1:117 X-BEDEWORK-IMAGE-Y1:17 X-BEDEWORK-IMAGE-X2:647 X-BEDEWORK-IMAGE-Y2:370.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Sifan Liu X-BEDEWORK-IMAGE:/public/Images/sep5_20250903010551PM.jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/sep5_20250903010551PM-thumb.png END:VEVENT BEGIN:VEVENT RECURRENCE-ID;TZID=America/New_York:20250912T153000 CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=8a00ec8b-979413b9-0198-a4d89b28-00001675:Thompson\, Ekaterina CREATED:20250904T132242Z DESCRIPTION:Conditional moment equality models are regularly encountered i n empirical economics\, yet they are difficult to estimate. These models map a conditional distribution of data to a structural parameter via the restriction that a conditional mean equals zero. Using this observation\, I introduce a Bayesian inference framework in which an unknown condition al distribution is replaced with a nonparametric posterior\, and structur al parameter inference is then performed using an implied posterior. The method has the same flexibility as frequentist semiparametric estimators and does not require converting conditional moments to unconditional mome nts. Importantly\, I prove a semiparametric Bernstein-von Mises theorem\, providing conditions under which\, in large samples\, the posterior for the structural parameter is approximately normal\, centered at an efficie nt estimator\, and has variance equal to the Chamberlain (1987) semiparam etric efficiency bound. As byproducts\, I show that Bayesian uncertainty quantification methods are asymptotically optimal frequentist confidence sets and derive low-level sufficient conditions for Gaussian process prio rs. The latter sheds light on a key prior stability condition and relates to the numerical aspects of the paper in which these priors are used to predict the welfare effects of price changes. DURATION:PT1H DTSTAMP:20250908T131546Z DTSTART;TZID=America/New_York:20250912T153000 LAST-MODIFIED:20250908T131546Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Semiparametric Bayesian Inference for a Conditional Moment Equalit y Model UID:CAL-8a00ec8b-979413b9-0199-14e4ca9c-0000776cdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER:Christopher D. Walker X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) X-BEDEWORK-IMAGE-X1:12 X-BEDEWORK-IMAGE-Y1:110 X-BEDEWORK-IMAGE-X2:1268 X-BEDEWORK-IMAGE-Y2:947.3333333333334 X-BEDEWORK-IMAGE-CROP-WIDTH:1256 X-BEDEWORK-IMAGE-CROP-HEIGHT:837.3333333333334 X-BEDEWORK-IMAGE-ALT-TEXT:Christopher D. Walker\, Assistant Professor in t he Department of Economics at Duke University. X-BEDEWORK-IMAGE:/public/Images/Chris Walker-240827-0646_20250908011258PM. jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/Chris Walker-240827-0646_20250908011 258PM-thumb.png END:VEVENT BEGIN:VEVENT RECURRENCE-ID;TZID=America/New_York:20250919T153000 CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=8a00ec8b-979413b9-0198-a4d89b28-00001675:Thompson\, Ekaterina CREATED:20250904T132242Z DESCRIPTION:Markov chain Monte Carlo (MCMC) algorithms are foundational me thods for Bayesian statistics. However\, most MCMC algorithms are inheren tly sequential\, and their time complexity scales linearly with the numbe r of samples generated. Previous work on adapting MCMC to modern hardware has therefore focused on running many independent chains in parallel. We take an alternative approach: we propose algorithms to evaluate MCMC sam plers in parallel across the chain length. To do this\, we build on recen t methods for parallel evaluation of nonlinear recursions that formulate the state sequence as a solution to a fixed-point problem\, which can be obtained via a parallel form of Newton's method. We show how this approac h can be used to parallelize Gibbs\, Metropolis-adjusted Langevin\, and H amiltonian Monte Carlo sampling across the sequence length. Moreover\, we prove theoretical results that link the convergence rate of the fixed-po int algorithm to the stability of the recursion\, yielding sublinear time complexity for certain problems. Finally\, to lower memory costs and red uce runtime in practice\, we develop quasi-Newton methods that are genera lly applicable for parallel evaluation of nonlinear recursions. Across s everal examples\, we demonstrate the simulation of up to hundreds of thou sands of MCMC samples with only tens of parallel Newton iterations. We fi nd that the proposed parallel algorithms accelerate MCMC\, in some cases by more than an order of magnitude compared to sequential evaluation. DURATION:PT1H DTSTAMP:20250915T131047Z DTSTART;TZID=America/New_York:20250919T153000 LAST-MODIFIED:20250915T131047Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Parallelizing MCMC Across the Sequence Length: T samples in O(log2 T) time UID:CAL-8a00ec8b-979413b9-0199-14e4ca9c-0000776cdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER:Scott W. Linderman\, Assistant Professor of Statistics and an Institute Scholar in the Wu Tsai Neurosciences Institute at Stanfo rd University X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) X-BEDEWORK-IMAGE-X1:126 X-BEDEWORK-IMAGE-Y1:129 X-BEDEWORK-IMAGE-X2:897 X-BEDEWORK-IMAGE-Y2:643 X-BEDEWORK-IMAGE-CROP-WIDTH:771 X-BEDEWORK-IMAGE-CROP-HEIGHT:514 X-BEDEWORK-IMAGE-ALT-TEXT:Scott W. Linderman X-BEDEWORK-IMAGE:/public/Images/Scott_20250915011047PM.png X-BEDEWORK-THUMB-IMAGE:/public/Images/Scott_20250915011047PM-thumb.png END:VEVENT BEGIN:VEVENT RECURRENCE-ID;TZID=America/New_York:20250926T153000 CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=8a00ec8b-979413b9-0198-a4d89b28-00001675:Thompson\, Ekaterina CREATED:20250904T132242Z DESCRIPTION:Every weighted\, undirected graph describes a random walk on i ts vertex set\, which is reversible Markov chain. I will describe recent work on graph joinings\, which leverages this connection and ideas from optimal transport to gain insights into both graph isomorphisms and coupl ings of Markov chains. A joining of two graphs is a product graph that d escribes a reversible coupling of their random walks. Given two graphs wi th labeled vertices\, the optimal graph joining (OGJ) problem identifies a joining that minimizes the total weight of vertex pairs with different labels. I will present new results showing that\, for suitable families of labeled graphs\, OGJ can detect and identify isomorphisms between grap hsin the family. In the opposite direction\, the study of joinings leads naturally to a notion of disjointness that quantifies structural discorda nce between graphs. I will present a simple characterization of weak dis jointness\, and show how this yields new insights into reversible couplin gs of reversible Markov chains.\n\nJoint work with Yang Xiang\, Phuong Ho ang\, Bongsoo Yi\, and Kevin McGoff. DURATION:PT1H DTSTAMP:20250922T122008Z DTSTART;TZID=America/New_York:20250926T153000 LAST-MODIFIED:20250922T122008Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Graph Joinings and Reversible Markov Chains UID:CAL-8a00ec8b-979413b9-0199-14e4ca9c-0000776cdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER:Andrew B. Nobel\, PhD X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:55 X-BEDEWORK-IMAGE-X2:1230 X-BEDEWORK-IMAGE-Y2:875 X-BEDEWORK-IMAGE-CROP-WIDTH:1230 X-BEDEWORK-IMAGE-CROP-HEIGHT:820 X-BEDEWORK-IMAGE-ALT-TEXT:Andrew B. Nobel\, PhD X-BEDEWORK-IMAGE:/public/Images/Screenshot 2025-09-22 at 8.18.44 AM_202509 22122009PM.png X-BEDEWORK-THUMB-IMAGE:/public/Images/Screenshot 2025-09-22 at 8.18.44 AM_ 20250922122009PM-thumb.png END:VEVENT BEGIN:VEVENT RECURRENCE-ID;TZID=America/New_York:20251003T153000 CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=8a00ec8b-979413b9-0198-a4d89b28-00001675:Thompson\, Ekaterina CREATED:20250904T132242Z DESCRIPTION:In recent years\, longitudinal studies increasingly collect da ta where the primary measurements are functions or surfaces observed repe atedly over time. This talk introduces parsimonious modeling frameworks f or such functional data\, designed to extract meaningful low-dimensional features while respecting the longitudinal design. The methodology is com putationally efficient and well-suited for characterizing the dynamic evo lution of the underlying process. We then extend the framework to accommo date pointwise skewness in the data\, broadening its applicability. Build ing on the key ideas\, we discuss inference in the form of significance t ests for hypotheses of scientific interest in this setting. We conclude b y highlighting several open challenges and emerging directions in the ana lysis of longitudinal functional data. DURATION:PT1H DTSTAMP:20250929T132604Z DTSTART;TZID=America/New_York:20251003T153000 LAST-MODIFIED:20250929T132604Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Longitudinal Functional Data Methods for Emerging Repeated Measure ments UID:CAL-8a00ec8b-979413b9-0199-14e4ca9c-0000776cdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER: Ana-Maria Staicu X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:141 X-BEDEWORK-IMAGE-X2:1149 X-BEDEWORK-IMAGE-Y2:907 X-BEDEWORK-IMAGE-CROP-WIDTH:1149 X-BEDEWORK-IMAGE-CROP-HEIGHT:766 X-BEDEWORK-IMAGE-ALT-TEXT:Ana-Maria Staicu X-BEDEWORK-IMAGE:/public/Images/ana-maria_20250929012604PM.png X-BEDEWORK-THUMB-IMAGE:/public/Images/ana-maria_20250929012604PM-thumb.png END:VEVENT BEGIN:VEVENT RECURRENCE-ID;TZID=America/New_York:20251017T153000 CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=8a00ec8b-979413b9-0198-a4d89b28-00001675:Thompson\, Ekaterina CREATED:20250904T132242Z DESCRIPTION:Beta regression is used routinely for continuous proportional data\, but it often encounters practical issues such as a lack of robustn ess to misspecification of the beta distribution and sensitivity to outli ers. We develop an improved class of generalized linear models starting w ith the continuous binomial (cobin) distribution and further extending to dispersion mixtures of cobin distributions (micobin). The proposed cobin regression and micobin regression models have attractive robustness\, co mputation\, and flexibility properties. A key innovation is the Kolmogoro v-Gamma data augmentation scheme\, which facilitates Gibbs sampling for B ayesian computation\, including in hierarchical cases involving nested\, longitudinal\, or spatial data. We demonstrate robustness\, ability to ha ndle responses exactly at the boundary (0 or 1)\, and computational effic iency relative to beta regression in simulation experiments and through a nalysis of the benthic macroinvertebrate multimetric index of US lakes us ing lake watershed covariates.\n\nJoint work with Benjamin Dahl (Duke U)\ , Otso Ovaskainen (U Jyväskylä)\, David Dunson (Duke U). DURATION:PT1H DTSTAMP:20251014T120547Z DTSTART;TZID=America/New_York:20251017T153000 LAST-MODIFIED:20251014T120547Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Scalable and robust regression models for continuous proportional data UID:CAL-8a00ec8b-979413b9-0199-14e4ca9c-0000776cdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER:Changwoo lee X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) X-BEDEWORK-IMAGE-X1:26 X-BEDEWORK-IMAGE-Y1:47 X-BEDEWORK-IMAGE-X2:556 X-BEDEWORK-IMAGE-Y2:400.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Changwoo Lee X-BEDEWORK-IMAGE:/public/Images/Screenshot 2025-10-14 at 8.05.17 AM_202510 14120547PM.png X-BEDEWORK-THUMB-IMAGE:/public/Images/Screenshot 2025-10-14 at 8.05.17 AM_ 20251014120547PM-thumb.png END:VEVENT BEGIN:VEVENT RECURRENCE-ID;TZID=America/New_York:20251024T153000 CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=8a00ec8b-979413b9-0198-a4d89b28-00001675:Thompson\, Ekaterina CREATED:20250904T132242Z DESCRIPTION:We present a unified offline decision making framework. In the first part\, we consider a class of assortment optimization problems in an offline data-driven setting. A firm does not know the underlying custo mer choice model but has access to an offline dataset consisting of the h istorically offered assortment set\, customer choice\, and revenue. The o bjective is to use the offline dataset to find an optimal assortment. Due to the combinatorial nature of assortment optimization\, the problem of insufficient data coverage is likely to occur in the offline dataset. The refore\, designing a provably efficient offline learning algorithm become s a significant challenge. To this end\, we propose an algorithm referred as Pessimistic ASsortment opTimizAtion (PASTA) following the spirit of p essimism. We show the algorithm identifies the optimal assortment by only requiring the offline data to cover the optimal assortment under general settings. In particular\, we establish a regret bound for the offline as sortment optimization problem under the celebrated multinomial logit mode l and its generalizations\, where the regret is shown to be minimax optim al. Joint work with Juncheng Dong\, Weibin Mo\, Zhengling Qi\, Cong Shi\, and Vahid Tarokh.\n \nIn the second part\, we consider the inferential p roblem in assortment optimization. Uncertainty quantification for the opt imal assortment is still largely unexplored despite its great practical significance. Instead of estimating and recovering the complete optimal o ffer set\, decision-makers may only be interested in testing whether a gi ven property holds true for the optimal assortment\, such as whether they should include several products of interest in the optimal set\, or how many categories of products the optimal set should include. We proposes a novel inferential framework for testing such properties. We reduce infer ring a general optimal assortment property to quantifying the uncertainty associated with the sign change point detection of the marginal revenue gaps. We show the asymptotic normality of the marginal revenue gap estima tor\, and construct a maximum statistic via the gap estimators to detect the sign change point. Joint work with Shuting Shen\, Alex Belloni\, Xi C hen\, and Junwei Lu. DURATION:PT1H30M DTSTAMP:20251023T162033Z DTSTART;TZID=America/New_York:20251024T150500 LAST-MODIFIED:20251023T162033Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Offline Data-Driven Decision Making with Applications to Assortmen t Optimization: Estimation and Inference UID:CAL-8a00ec8b-979413b9-0199-14e4ca9c-0000776cdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER:Ethan Fang X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:183 X-BEDEWORK-IMAGE-X2:2366 X-BEDEWORK-IMAGE-Y2:1760.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:2366 X-BEDEWORK-IMAGE-CROP-HEIGHT:1577.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Ethan Fang X-BEDEWORK-IMAGE:/public/Images/Ethan_Fang_Photo_20251020052429PM.jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/Ethan_Fang_Photo_20251020052429PM-th umb.png END:VEVENT BEGIN:VEVENT RECURRENCE-ID;TZID=America/New_York:20251031T153000 CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=8a00ec8b-979413b9-0198-a4d89b28-00001675:Thompson\, Ekaterina CREATED:20250904T132242Z DESCRIPTION:With increased data collection\, the need to fuse data sources has emerged as an important and rapidly growing research activity in the statistical community. In considering spatial and spatio-temporal datase ts to examine complex environmental and ecological processes of interest\ , we often have multiple sources that are jointly informative about featu res of interest of the processes. Model-based data fusion aims to leverag e information from these sources to improve inference and prediction. In the spatial statistics setting\, these data could be geostatistical\, are al\, or point patterns with varying spatial resolutions\, supports\, and domains. With focus on North Atlantic right whales\, we explore stochasti c modeling to implement a suitable fusion to inform about their abundance and distribution with full inference and uncertainty. The first source i s aerial distance sampling\, which provides the spatial locations of whal es detected in the region. The second source is passive acoustic monitori ng (PAM)\, returning calls received at hydrophones placed on the ocean fl oor. Due to limited time on the surface and detection limitations arising from sampling effort\, aerial distance sampling only provides a partial realization of locations. With PAM we never observe numbers or locations of individuals. To address these challenges\, we develop a novel thinned point pattern data fusion. We demonstrate performance gains of our approa ch compared to that from a single source through simulation and apply our model to North Atlantic right whale data collected throughout Cape Cod B ay\, Massachusetts in the U.S. DURATION:PT1H DTSTAMP:20251028T164557Z DTSTART;TZID=America/New_York:20251031T153000 LAST-MODIFIED:20251028T164557Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Spatial data fusion of point processes to infer marine mammal abun dance UID:CAL-8a00ec8b-979413b9-0199-14e4ca9c-0000776cdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER:Erin Schliep X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) X-BEDEWORK-IMAGE-X1:52 X-BEDEWORK-IMAGE-Y1:112 X-BEDEWORK-IMAGE-X2:1050 X-BEDEWORK-IMAGE-Y2:777.3333333333334 X-BEDEWORK-IMAGE-CROP-WIDTH:998 X-BEDEWORK-IMAGE-CROP-HEIGHT:665.3333333333334 X-BEDEWORK-IMAGE-ALT-TEXT:Erin Schliep X-BEDEWORK-IMAGE:/public/Images/Screenshot 2025-10-27 at 1.31.57 PM_202510 28044557PM.png X-BEDEWORK-THUMB-IMAGE:/public/Images/Screenshot 2025-10-27 at 1.31.57 PM_ 20251028044557PM-thumb.png END:VEVENT BEGIN:VEVENT RECURRENCE-ID;TZID=America/New_York:20251107T203000Z CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=8a00ec8b-979413b9-0198-a4d89b28-00001675:Thompson\, Ekaterina CREATED:20250904T132242Z DESCRIPTION:TBD DURATION:PT1H DTSTAMP:20251103T131718Z DTSTART;TZID=America/New_York:20251107T153000 LAST-MODIFIED:20251103T131718Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Friday Seminar in the department of Statistical Science UID:CAL-8a00ec8b-979413b9-0199-14e4ca9c-0000776cdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER:TBD X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) EXDATE:20251010T193000Z END:VEVENT BEGIN:VEVENT RECURRENCE-ID;TZID=America/New_York:20251114T203000Z CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=8a00ec8b-979413b9-0198-a4d89b28-00001675:Thompson\, Ekaterina CREATED:20250904T132242Z DESCRIPTION:TBD DURATION:PT1H DTSTAMP:20251110T223611Z DTSTART;TZID=America/New_York:20251114T153000 LAST-MODIFIED:20251110T223611Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Friday Seminar in the department of Statistical Science UID:CAL-8a00ec8b-979413b9-0199-14e4ca9c-0000776cdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER:TBD X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) EXDATE:20251010T193000Z END:VEVENT BEGIN:VEVENT RECURRENCE-ID;TZID=America/New_York:20251121T203000Z CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=8a00ec8b-979413b9-0198-a4d89b28-00001675:Thompson\, Ekaterina CREATED:20250904T132242Z DESCRIPTION:TBD DURATION:PT1H DTSTAMP:20251119T160701Z DTSTART;TZID=America/New_York:20251121T153000 LAST-MODIFIED:20251119T160701Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Friday Seminar in the department of Statistical Science UID:CAL-8a00ec8b-979413b9-0199-14e4ca9c-0000776cdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER:TBD X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) EXDATE:20251010T193000Z END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20260109T162145Z DESCRIPTION:Many modern datasets exhibit low-dimensional organization desp ite living in high ambient dimensions\, yet this structure is often impli cit\, noisy\, and not explicitly modeled. In this talk\, I discuss statis tical inference in such settings through two complementary examples. I fi rst present recent theoretical results on Gaussian process regression for data supported on unknown low-dimensional structures. Using a real-domai n small-bandwidth analysis\, I show how intrinsic geometry governs approx imation and posterior contraction behavior without relying on spectral or Laplacian machinery\, leading to adaptive rates driven by intrinsic rath er than ambient dimension. I then turn to a biological application in sin gle-cell genomics\, where local neighborhood graphs are used to define op timal-transport-based gene affinities and reveal gene trajectories and un derlying biological processes. These examples illustrate how latent geome tric structure\, when present but only weakly revealed by data\, can none theless be harnessed to enable reliable statistical inference in high dim ensions. DURATION:PT1H DTSTAMP:20260113T145537Z DTSTART;TZID=America/New_York:20260116T153000 LAST-MODIFIED:20260113T145537Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Intrinsic geometry and latent low dimensionality in data: manifold s and beyond UID:CAL-8a00eca5-9af98aae-019b-a3905d65-00003415demobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=ekaterina.thompson@duke. edu:Ekaterina Thompson X-BEDEWORK-SPEAKER:Xiuyuan Cheng X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-IMAGE-X1:142 X-BEDEWORK-IMAGE-Y1:157 X-BEDEWORK-IMAGE-X2:1069 X-BEDEWORK-IMAGE-Y2:775 X-BEDEWORK-IMAGE-CROP-WIDTH:927 X-BEDEWORK-IMAGE-CROP-HEIGHT:618 X-BEDEWORK-IMAGE-ALT-TEXT:Xiuyuan Cheng X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) X-BEDEWORK-IMAGE:/public/Images/Screenshot 2026-01-09 at 11.20.07 AM_20260 113025537PM.png X-BEDEWORK-THUMB-IMAGE:/public/Images/Screenshot 2026-01-09 at 11.20.07 AM _20260113025537PM-thumb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20260120T121156Z DESCRIPTION:With breakthroughs in scientific computing\, virtual simulator s are increasingly used as "digital twins" for studying complex scientifi c phenomena\, e.g.\, particle collisions and nuclear reactions. Such simu lators\, however\, are highly time-intensive\, which hinders their effect ive use for scientific decision-making\, e.g.\, the optimization of a par ticle detector. There is a pressing need for novel Bayesian ML/AI methods that synergize with digital twins to enable timely\, uncertainty-aware\, and interpretable decision-making. One promising tool is Bayesian optimi zation (BO)\, which tackles the optimization of a costly black-box functi on (e.g.\, the response surface of the virtual simulator) using limited f unction evaluations. In this talk\, I will present a suite of novel BO me thods that address several important needs for scientific applications. T he first is a new BO method for identifying varied local optima of a blac k-box function\, which provides scientists with a basket of different opt imal solutions for flexible decision-making. The second method tackles th e challenging problem of high-dimensional black-box optimization\, offeri ng improved theoretical rates and empirical performance over existing tec hniques in the one-shot setting. The effectiveness of these methods will be investigated in ongoing collaborative projects in the physical and eng ineering sciences. DURATION:PT1H DTSTAMP:20260120T121156Z DTSTART;TZID=America/New_York:20260123T033000 LAST-MODIFIED:20260120T121156Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Advances in Bayesian optimization for accelerating scientific deci sion-making UID:CAL-8a00eca5-9af98aae-019b-db519adc-000026c6demobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER:Simon Mak X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=ekaterina.thompson@duke. edu:Ekaterina Thompson X-BEDEWORK-IMAGE-X1:69 X-BEDEWORK-IMAGE-Y1:133 X-BEDEWORK-IMAGE-X2:1179 X-BEDEWORK-IMAGE-Y2:873 X-BEDEWORK-IMAGE-CROP-WIDTH:1110 X-BEDEWORK-IMAGE-CROP-HEIGHT:740 X-BEDEWORK-IMAGE-ALT-TEXT:Simon Mak X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) X-BEDEWORK-IMAGE:/public/Images/Screenshot 2026-01-20 at 7.10.38 AM_202601 20121139PM.png X-BEDEWORK-THUMB-IMAGE:/public/Images/Screenshot 2026-01-20 at 7.10.38 AM_ 20260120121139PM-thumb.png END:VEVENT BEGIN:VEVENT RECURRENCE-ID;TZID=America/New_York:20260130T203000Z CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20260122T165024Z DESCRIPTION:TBD DURATION:PT1H DTSTAMP:20260126T141315Z DTSTART;TZID=America/New_York:20260130T153000 LAST-MODIFIED:20260126T141315Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Friday Seminar in the department of Statistical Science UID:CAL-8a00eca5-9af98aae-019b-e69d42e9-00000e2fdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER:TBD X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=statsci-friday-seminars@ duke.edu:StatSci Friday Seminars EXDATE:20260306T203000Z EXDATE:20260313T193000Z EXDATE:20260320T193000Z END:VEVENT BEGIN:VEVENT RECURRENCE-ID;TZID=America/New_York:20260206T203000Z CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20260122T165024Z DESCRIPTION:TBD DURATION:PT1H DTSTAMP:20260127T165335Z DTSTART;TZID=America/New_York:20260206T153000 LAST-MODIFIED:20260127T165335Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Friday Seminar in the department of Statistical Science UID:CAL-8a00eca5-9af98aae-019b-e69d42e9-00000e2fdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER:TBD X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=statsci-friday-seminars@ duke.edu:StatSci Friday Seminars EXDATE:20260306T203000Z EXDATE:20260313T193000Z EXDATE:20260320T193000Z END:VEVENT BEGIN:VEVENT RECURRENCE-ID;TZID=America/New_York:20260213T203000Z CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20260122T165024Z DESCRIPTION:TBD DURATION:PT1H DTSTAMP:20260209T145752Z DTSTART;TZID=America/New_York:20260213T153000 LAST-MODIFIED:20260209T145752Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Friday Seminar in the department of Statistical Science UID:CAL-8a00eca5-9af98aae-019b-e69d42e9-00000e2fdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER:TBD X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=statsci-friday-seminars@ duke.edu:StatSci Friday Seminars EXDATE:20260306T203000Z EXDATE:20260313T193000Z EXDATE:20260320T193000Z END:VEVENT BEGIN:VEVENT RECURRENCE-ID;TZID=America/New_York:20260220T203000Z CATEGORIES:Utilities CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20260122T165024Z DESCRIPTION:TBD DURATION:PT1H DTSTAMP:20260218T213537Z DTSTART;TZID=America/New_York:20260220T153000 LAST-MODIFIED:20260218T213537Z LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi stry 116 STATUS:CONFIRMED SUMMARY:Friday Seminar in the department of Statistical Science UID:CAL-8a00eca5-9af98aae-019b-e69d42e9-00000e2fdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-SPEAKER:TBD X-BEDEWORK-DUKE-SERIES:Friday Seminar in the department of Statistical Sci ence X-BEDEWORK-SUBMITTEDBY:ek217 for Statistical Science (agrp_StatisticalScie nce) X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=statsci-friday-seminars@ duke.edu:StatSci Friday Seminars EXDATE:20260306T203000Z EXDATE:20260313T193000Z EXDATE:20260320T193000Z END:VEVENT END:VCALENDAR