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:20231229T143339Z
DESCRIPTION:The total electron content (TEC) maps can be used to estimate
the signal delay of GPS due to the ionospheric electron content between a
receiver and a satellite. This delay can result in a GPS positioning err
or. Thus\, it is important to monitor and forecast the TEC maps. However\
, the observed TEC maps have big patches of missingness in the ocean and
scattered small areas on the land. Thus\, precise imputation and predicti
on of the TEC maps are crucial in space weather forecasting. \n\nIn this
talk\, I first present several extensions of existing matrix completion a
lgorithms to achieve TEC map reconstruction\, accounting for spatial smoo
thness and temporal consistency while preserving essential structures of
the TEC maps. We show that our proposed method achieves better reconstruc
ted TEC maps as compared to existing methods in the literature. I will al
so briefly describe the use of our large-scale complete TEC database. The
n\, I present a new model for forecasting time series data distributed on
a matrix-shaped spatial grid\, using the historical spatiotemporal data
and auxiliary vector-valued time series data. Large sample asymptotics of
the estimators for both finite and high dimensional settings are establi
shed. Performances of the model are validated with extensive simulation s
tudies and an application to forecast the global TEC distributions.
DURATION:PT1H
DTSTAMP:20240102T213722Z
DTSTART;TZID=America/New_York:20240112T153000
LAST-MODIFIED:20240102T213722Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Video Imputation and Prediction Methods with Applications in Space
Weather
UID:CAL-8a018ccf-8b87f80e-018c-b5febd0f-00003128demobedework@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:Yang Chen\, Assistant Professor\, University of Michiga
n
X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar Series
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:20240113T005531Z
DESCRIPTION:The biochemical functions of proteins\, such as catalyzing a c
hemical reaction or binding to a virus\, are typically conferred by the g
eometry of only a handful of atoms. This arrangement of atoms\, known as
a motif\, is structurally supported by the rest of the protein\, referre
d to as a scaffold. A central task in protein design is to identify a di
verse set of stabilizing scaffolds to support a motif known or theorized
to confer function. This long-standing challenge is known as the motif-sc
affolding problem.\n \nIn this talk\, I describe a statistical approach I
have developed to address the motif-scaffolding problem. My approach in
volves (1) estimating a distribution supported on realizable protein stru
ctures and (2) sampling scaffolds from this distribution conditioned on a
motif. For step (1) I adapt diffusion generative models to fit example
protein structures from nature. For step (2) I develop sequential monte
carlo algorithms to sample from the conditional distributions of these mo
dels. I finally describe how\, with experimental and computational colla
borators\, I have generalized and scaled this approach to generate and ex
perimentally validate hundreds of proteins with various functional specif
ications.\n \nBio:\nBrian Trippe is a postdoctoral fellow at Columbia Uni
versity in the Department of Statistics\, and a visiting researcher at th
e Institute for Protein Design at the University of Washington. He comple
ted his Ph.D. in Computational and Systems Biology at the Massachusetts I
nstitute of Technology where worked on Bayesian methods for inference in
hierarchical linear models. In his research\, Brian develops statistical
machine learning methods to address challenges in biotechnology and medic
ine\, with a focus on generative modeling and inference algorithms for pr
otein engineering.
DURATION:PT1H
DTSTAMP:20240116T171502Z
DTSTART;TZID=America/New_York:20240119T153000
LAST-MODIFIED:20240116T171502Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Probabilistic methods for designing functional protein structures
UID:CAL-8a008bcc-8cfbd545-018d-00511b7e-00005306demobedework@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:Brian Trippe\, Columbia University
X-BEDEWORK-DUKE-SERIES:Statistical Science
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:20240109T033728Z
DESCRIPTION:Optimization techniques\, such as dual ascent\, alternating di
rection method of multipliers\, and majorization-minimization\, are widel
y used in high-dimensional applications. The strengths of optimization ar
e the high computing efficiency and the ease of inducing point estimates
on useful constrained spaces\, such as those satisfying low rank\, low ca
rdinality or combinatorial structure. For uncertainty quantification arou
nd point estimate\, a popular generalized Bayes solution known as Gibbs p
osterior exponentiates the negative loss function\, and forms a posterior
density. Despite successful theoretic justifications\, Gibbs posterior d
istribution is supported in a high-dimensional space and\, hence often do
es not inherit nice properties in computing efficiency and constraints fr
om optimization. In this work\, we are motivated by a discovery that a la
rge class of penalized profile likelihoods\, which partially maximize ove
r a subset of parameters\, in fact enjoy equivalence to another generativ
e model for the data. This leads us to explore a new generalized Bayes ap
proach that views the likelihood as an equality-constrained function\, ba
sed on data\, parameters\, and a conditionally deterministic latent varia
ble equal to an optimization solution. This new likelihood can be justifi
ed as a special case of augmented likelihood where the latent variable is
typically exploited to model dependency among the data. Therefore\, this
framework coined "bridged posterior'' conforms to the Bayesian methodolo
gy. A surprising theoretical finding is that under mild conditions\, the
square root n-adjusted bridged posterior distribution of the parameters c
onverges to the same asymptotical normal that the canonical integrated po
sterior converges to. Therefore\, our results formally dispel a long-time
belief that partial optimization over latent variables might lead to an
underestimation of parameter uncertainty. We demonstrate the practical ad
vantages of our approach in applications\, such as classification with pa
rtially labeled data and harmonization of multiple brain scan networks.
DURATION:PT1H
DTSTAMP:20240112T200140Z
DTSTART;TZID=America/New_York:20240122T114500
LAST-MODIFIED:20240112T200140Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Bridged Posterior: Optimization\, Profile Likelihood and a New App
roach of Generalized Bayes
UID:CAL-8a018ccf-8b87f80e-018c-ec4beffe-000055e3demobedework@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:Leo Duan\, Assistant Professor\, University of Florida
X-BEDEWORK-DUKE-SERIES:Statistical Science
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:20240110T024331Z
DESCRIPTION:The exploratory and interactive nature of modern data analysis
often introduces selection bias\, posing challenges for traditional stat
istical inference methods. A common strategy to address this bias is by c
onditioning on the selection event. However\, this often results in a con
ditional distribution that is intractable and requires Markov chain Monte
Carlo (MCMC) sampling for inference. Notably\, some of the most widely u
sed selection algorithms yield selection events that can be characterized
as polyhedra\, such as the lasso for variable selection and the epsilon-
greedy algorithm for multi-armed bandit problems. This talk will present
a method that is tailored for conducting inference following polyhedral s
election. The method transforms the variables constrained within a polyhe
dron into variables within a unit cube\, allowing for exact sampling. Com
pared to MCMC\, the proposed method offers superior speed and accuracy\,
providing a practical and efficient approach for conditional selective in
ference. Additionally\, it facilitates the computation of the selection-a
djusted maximum likelihood estimator\, enabling MLE-based inference. Nume
rical results demonstrate the enhanced performance of the proposed method
compared to alternative approaches for selective inference.
DURATION:PT1H
DTSTAMP:20240110T024331Z
DTSTART;TZID=America/New_York:20240126T153000
LAST-MODIFIED:20240110T024331Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:An Exact Sampler for Inference after Polyhedral Selection
UID:CAL-8a018ccf-8b87f80e-018c-f140e857-0000167edemobedework@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:Sifan Liu\, Stanford 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:20240110T140507Z
DESCRIPTION:Mixed effect modeling for longitudinal data is challenging whe
n the observed data are random objects\, which are complex data taking va
lues in a general metric space without either global linear or local line
ar (Riemannian) structure. In such settings the classical additive error
model and distributional assumptions are unattainable. Due to the rapid a
dvancement of technology\, longitudinal data containing complex random ob
jects\, such as covariance matrices\, data on Riemannian manifolds\, and
probability distributions are becoming more common. Addressing this chall
enge\, we develop a mixed-effects regression for data in geodesic spaces\
, where the underlying mean response trajectories are geodesics in the me
tric space and the deviations of the observations from the model are quan
tified by perturbation maps or transports. A key finding is that the geod
esic trajectories assumption for the case of random objects is a natural
extension of the linearity assumption in the standard Euclidean scenario
to the case of general geodesic metric spaces. Geodesics can be recovered
from noisy observations by exploiting a connection between the geodesic
path and the path obtained by global Fréchet regression for random object
s. The effect of baseline Euclidean covariates on the geodesic paths is m
odeled by another Fréchet regression step. We study the asymptotic conver
gence of the proposed estimates and provide illustrations through simulat
ions and real-data applications.
DURATION:PT1H
DTSTAMP:20240127T192305Z
DTSTART;TZID=America/New_York:20240129T114500
LAST-MODIFIED:20240127T192305Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:CANCELLED:Geodesic Mixed Effects Models for Repeatedly Observed/Lo
ngitudinal Random Objects
UID:CAL-8a018ccf-8b87f80e-018c-f3b0eca1-00002992demobedework@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:Satarupa Bhattacharjee\, Pennsylvania State
X-BEDEWORK-DUKE-SERIES:Statistical Science
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:20240110T022314Z
DESCRIPTION:Monte Carlo methods span a range of disciplines\, drawing inte
rest from statisticians\, computer scientists\, physicists\, among others
. In the Monte Carlo workflow\, the upstream task involves designing and
analyzing sampling algorithms\, while the downstream task focuses on effi
ciently using these samples to form estimators. In this talk\, I will dis
cuss recent developments of both aspects. The first part introduces the '
Spectral Telescope' framework for analyzing Gibbs samplers\, and discusse
s its relationship with the spectral independence technique recently deve
loped in theoretical computer science. The second part focuses on the dev
elopment of unbiased estimators through the combination of Markov Chain M
onte Carlo and Multilevel Monte Carlo methods\, highlighting their potent
ial in parallel computing.\n\nBiosketch: Guanyang Wang is an Assistant Pr
ofessor in the Department of Statistics at Rutgers University. He complet
ed his Ph.D. in Mathematics with a Ph.D. minor in Statistics at Stanford
University\, advised by Professor Persi Diaconis. Guanyang Wang's researc
h primarily centers on Monte Carlo methods\, applied probability\, and st
atistical computing. Recently\, he is also working on quantum computing.
His research receives support from both the NSF and an Adobe Data Science
Research Award.
DURATION:PT1H
DTSTAMP:20240110T165332Z
DTSTART;TZID=America/New_York:20240202T153000
LAST-MODIFIED:20240110T165332Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:The Many Facets of Monte Carlo Methods: From Sampling Algorithms t
o Unbiased Estimators
UID:CAL-8a018ccf-8b87f80e-018c-f12e53cb-000015bddemobedework@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:Guanyang Wang\, Assistant Professor\, Rutgers Universit
y
X-BEDEWORK-DUKE-SERIES:Statistical Science
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:20240110T184203Z
DESCRIPTION:Recent interest has centered on uncertainty quantification for
machine learning models. For the most part\, this work has assumed indep
endence of the observations. However\, many of the most important problem
s arising across scientific fields\, from genomics to climate science\, i
nvolve systems where dependence cannot be ignored. In this talk\, I will
investigate inference on machine learning models in the presence of depen
dence. \n\nIn the first part of my talk\, I will consider a common practi
ce in the field of genomics in which researchers compute a correlation ma
trix between genes and threshold its elements in order to extract groups
of independent genes. I will describe how to construct valid p-values ass
ociated with these discovered groups that properly account for the group
selection process. While this is related to the literature on selective
inference developed in the past decade\, this work involves inference abo
ut the covariance matrix rather than the mean\, and therefore requires an
entirely new technical toolset. This same toolset can be applied to quan
tify the uncertainty associated with canonical correlation analysis after
feature screening. \n\nIn the second part of my talk\, I will turn to an
important problem in the field of oceanography as it relates to climate
science. Oceanographers have recently applied random forests to estimate
carbon export production\, a key quantity of interest\, at a given locati
on in the ocean\; they then wish to sum the estimates across the world's
oceans to obtain an estimate of global export production. While quantifyi
ng uncertainty associated with a single estimate is relatively straightfo
rward\, quantifying uncertainty of the summed estimates is not\, due to t
heir complex dependence structure. I will adapt the theory of V-statistic
s to this dependent data setting in order to establish a central limit th
eorem for the summed estimates\, which can be used to quantify the uncert
ainty associated with global export production across the world's oceans.
\n\nThis is joint work with my postdoctoral supervisors\, Daniela Witten
(University of Washington) and Jacob Bien (University of Southern Califor
nia).
DURATION:PT1H
DTSTAMP:20240126T191301Z
DTSTART;TZID=America/New_York:20240205T114500
LAST-MODIFIED:20240126T191301Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Inference for machine learning under dependence
UID:CAL-8a018ccf-8b87f80e-018c-f4ae7643-00007e00demobedework@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:Arkajyoti Saha\, University of Washington
X-BEDEWORK-DUKE-SERIES:Statistical Science
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:20240122T180117Z
DESCRIPTION:This presentation offers insights into the work environment at
JMP Statistical Discovery LLC\, a prominent statistical software company
\, with a specific focus on the Research and Development (R&D) division.
Within this dynamic industry\, I'll showcase unique aspects of working at
JMP\, emphasizing the collaborative and innovative culture that defines
the company. In this talk\, I will share insights on software testing\, a
critical aspect of the company's operations. While the testing group at
JMP have graduate degrees in statistics and related fields\, software tes
ting is not a topic covered in most statistics programs. This talk will d
iscuss the challenges and intricacies of software testing and cutting-edg
e testing techniques used within JMP.\n \nDr. Ryan Lekivetz is a Senior M
anager of Advanced Analytics R&D at JMP\, heading the Design of Experimen
ts (DOE) and Reliability Development team. Ryan earned his doctorate in s
tatistics from Simon Fraser University in Burnaby\, British Columbia. He
has published papers on DOE topics in peer-reviewed journals and holds ma
ny patents that he shares with his team members. His research interests i
nclude design of experiments\, combinatorial testing\, and assessing the
usability of statistical software through designed experiments.
DURATION:PT1H15M
DTSTAMP:20240126T210345Z
DTSTART;TZID=America/New_York:20240207T150500
LAST-MODIFIED:20240126T210345Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:A Day in the Life: JMP R&D
UID:CAL-8a0292fd-8d13410f-018d-32557553-00001ea8demobedework@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=megan.deyncourt@duke.edu
:Megan Deyncourt
X-BEDEWORK-SPEAKER:Dr. Ryan Lekivetz\, Senior Manager\, Advanced Analytics
R&D\, JMP
X-BEDEWORK-DUKE-SERIES:Statistical Science Proseminar
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:20240112T194732Z
DESCRIPTION:In this talk\, I will discuss semi-parametric estimation when
nuisance parameters cannot be estimated consistently\, focusing in partic
ular on the estimation of average treatment effects\, conditional correla
tions\, and linear effects under high-dimensional GLM specifications. In
this challenging regime\, even standard doubly-robust estimators can be i
nconsistent. I describe novel approaches which enjoy consistency guarante
es for low-dimensional target parameters even though standard approaches
fail. For some target parameters\, these guarantees can also be used for
inference. Finally\, I will provide my perspective on the broader implica
tions of this work for designing methods which are less sensitive to bias
es from high-dimensional prediction models.
DURATION:PT1H
DTSTAMP:20240117T144245Z
DTSTART;TZID=America/New_York:20240209T153000
LAST-MODIFIED:20240117T144245Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Debiasing in the inconsistency regime
UID:CAL-8a008bcc-8cfbd545-018c-ff372232-00003c95demobedework@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:Michael Celentano\, University of California\, Berkeley
X-BEDEWORK-DUKE-SERIES:Statistical Science
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:20240216T212156Z
DESCRIPTION:Ensemble decision tree methods such as XGBoost\, RF\, and BART
have gained enormous popularity in data science for their superior perfo
rmance in machine learning regression and classification tasks. In this p
aper\, we develop a new Bayesian graph-split-based additive decision tree
s method\, called GS-BART\, to improve the performance of Bayesian additi
ve decision trees for complex dependent data with graph relations. The ne
w method adopts a highly flexible split rule complying with graph structu
re to relax the axis-parallel split rule assumption in most existing ense
mble decision tree models. We design a scalable informed MCMC algorithm l
everaging a gradient-based recursive algorithm on spanning trees or chain
s to sample the graph-split-based decision tree. The superior performance
of the method over conventional ensemble tree models and gaussian proces
s models is illustrated in various regression and classification tasks fo
r spatial and network data analysis.
DURATION:PT1H
DTSTAMP:20240216T220058Z
DTSTART;TZID=America/New_York:20240223T153000
LAST-MODIFIED:20240216T220058Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:GS-BART: Graph split additive decision trees for classification an
d nonparametric regression of spatial and network data
UID:CAL-8a0292fd-8d13410f-018d-b3cc235e-000015c7demobedework@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:Huiyan Sang Professor of Statistics Texas A&M Universit
y
X-BEDEWORK-DUKE-SERIES:Statistical Science
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:20240125T163537Z
DESCRIPTION:following the launch of GPT4-Agent\, GPT4 has demonstrated its
flexibility in utilizing tools like Advanced Data Analytics (ADA\, previ
ously known as code interpreter) and DALL- E3\, although the details of G
PT4-Agent have not been fully disclosed. Over the past years\, we have in
tensively studied the core functionalities of GPT4\, progressively develo
ping a system comparable to GPT4-Agent named InfiAgent. Initially\, we re
plicated Codex and discovered that while existing models such as CodeLlam
a\, StarCoder\, and WizardCoder excel in programming capabilities\, they
fall short in handling FreeformQA problems for coding. To address this\,
we created InfiCoder-the first open-source model capable of handling text
-to-code\, code-to-code\, and freeform code-related QA tasks simultaneous
ly. Building on this\, we developed InfiCoder-Eval (FreeformQA benchmark)
\, which includes 270 high-quality automated test questions. Our findings
indicate that even GPT4 has room for improvement in this area (achieving
a score rate of only 59.13%). Based on InfiCoder\, we launched the InfiA
gent framework. This framework first defines the problem framework and ev
aluation objectives for data analysis. Then\, in line with the data analy
sis scenarios\, we developed a specialized Agent system based on the Reac
t format and LLM. This system integrates an LLM with programming capabili
ties and a sandbox environment for executing Python code\, generating sol
utions\, and corresponding code through multiple rounds of dialogues. It
is the industry's first Agent framework closest to the capabilities of AD
A. Additionally\, we expanded the application scenarios of InfiAgent\, es
pecially in multimodal reasoning. We observed that there is significant r
oom for improvement in the current GPT4V (achieving a score rate of only
74.44%). These achievements not only reveal the tremendous potential of I
nfiAgent but also showcase our possible directions in surpassing the capa
bilities of GPT4.\n \nBio: Dr. Hongxia Yang\, Ph.D.\, from Duke Universit
y\, has published over 100 papers in top-tier conferences and journals an
d holds more than 50 patents in the USA and China. Currently\, she serves
as the Head of large models at ByteDance US. Previously\, she worked as
a research staff member at IBM T.J. Watson Research Center\, as a princip
al scientist for Computational Advertising at Yahoo!\, as an AI scientist
and director at Alibaba DAMO Academy\, and as an adjunct professor at Zh
ejiang University's Shanghai Advanced Research.
DURATION:PT1H15M
DTSTAMP:20240220T213214Z
DTSTART;TZID=America/New_York:20240228T150500
LAST-MODIFIED:20240220T213214Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:InfiAgent: A Multi-Tool Agent for AI Operating Systems
UID:CAL-8a0292fd-8d13410f-018d-417a1af6-00000d6ademobedework@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=megan.deyncourt@duke.edu
:Megan Deyncourt
X-BEDEWORK-SPEAKER:Dr. Hongxia Yang\, Head of Large Models\, ByteDance US
X-BEDEWORK-DUKE-SERIES:Statistical Science Proseminar
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:20240226T154511Z
DESCRIPTION:Visualizations allow analysts to rapidly explore and make sens
e of their data. The ways we visualize data directly influence the conclu
sions we draw and decisions we make\; however\, our knowledge of how visu
alization design influences data analysis is largely grounded in heuristi
cs and intuition. My research instead empirically models how people inter
pret visualized data to understand limitations in current visualization s
ystems. We use these results to develop novel visualization systems that
support accurate analysis of complex data and better scale to the needs o
f modern analytics challenges by incorporating interactive statistical an
alytics and immersive display technologies to increase the accessibility\
, scalability\, and pervasiveness of data-driven reasoning. In this talk\
, I will discuss our efforts towards improving exploratory data analysis
guidelines and tools across a variety of domains.
DURATION:PT1H
DTSTAMP:20240226T154511Z
DTSTART;TZID=America/New_York:20240301T153000
LAST-MODIFIED:20240226T154511Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Leveraging Visual Cognition in Data Visualization
UID:CAL-8a0292fd-8d13410f-018d-e6176e9e-00002743demobedework@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:Danielle Szafir\, Assistant Professor of Computer Scien
ce at the University of North Carolina-Chapel Hill
X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar Series
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:20240226T184141Z
DESCRIPTION:Tree-based methods are popular nonparametric tools for capturi
ng spatial heterogeneity and making predictions in multivariate problems.
In unsupervised learning\, trees and their ensembles have also been appl
ied to a wide range of statistical inference tasks\, such as multi-resolu
tion sketching of distributional variations\, localization of high-densit
y regions\, and design of efficient data compression schemes. In this tal
k\, we will focus on the density estimation problem---a fundamental one i
n unsupervised learning. We consider the optional P{\\'o}lya tree (Wong a
nd Ma\, 2010) prior and the Dirichlet prior or their variations on indivi
dual trees. First we show that Bayesian density trees can achieve minimax
(up to a logarithmic term) convergence over the anisotropic Besov class\
, which implies that tree based methods can adapt to spatially inhomogene
ous features of the underlying density function\, and can achieve fast co
nvergence as the dimension increases. We will also introduce a novel Baye
sian model for forests\, and show that for a class of anisotropic H{\\"o}
lder continuous functions\, such type of density forests can achieve fast
er convergence than trees. The convergence rate is adaptive in the sense
that to achieve such a rate we do not need any prior knowledge of the smo
othness level of the density. The Bayesian framework naturally provides a
stochastic search algorithm over either the tree space or the forest one
. For both Bayesian density trees and forests\, we will provide several n
umerical results to illustrate their performance in the moderately high-d
imensional case.
DURATION:PT1H
DTSTAMP:20240302T141658Z
DTSTART;TZID=America/New_York:20240308T153000
LAST-MODIFIED:20240302T141658Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Bayesian trees and forests for unsupervised learning and their spa
tial adaptation properties
UID:CAL-8a0292fd-8d13410f-018d-e6b90689-0000359edemobedework@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:Linxi Liu\, Assistant Professor\, University of Pittsbu
rgh
X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar
X-BEDEWORK-SUBMITTEDBY:kherndon for Statistical Science (agrp_StatisticalS
cience)
END:VEVENT
BEGIN:VEVENT
CATEGORIES:Utilities
CATEGORIES:Main
CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None
CREATED:20240311T182540Z
DESCRIPTION:Inverse scattering aims to infer information about a hidden ob
ject by using the received scattered waves and training data collected fr
om forward mathematical models. Recent advances in computing have led to
increasing attention towards functional inverse inference\, which can rev
eal more detailed properties of a hidden object. However\, rigorous studi
es on functional inverse\, including the reconstruction of the functional
input and quantification of uncertainty\, remain scarce. Motivated by an
inverse scattering problem where the objective is to infer the functiona
l input representing the refractive index of a bounded scatterer\, a new
Bayesian framework will be discussed in the first part of this talk. It c
ontains a surrogate model that takes into account the functional inputs d
irectly through kernel functions\, and a Bayesian procedure that infers f
unctional inputs through the posterior distribution. In the second part o
f this talk\, we will introduce a novel procedure that\, given sparse dat
a generated from a stationary deterministic nonlinear dynamical system\,
can characterize specific local and/or global dynamic behavior with rigor
ous probability guarantees. More precisely\, the sparse data is used to c
onstruct a statistical surrogate model based on a Gaussian process (GP).
The dynamics of the surrogate model is interrogated using combinatorial
methods and characterized using algebraic topological invariants (Conley
index). The GP predictive distribution provides a lower bound on the conf
idence that these topological invariants\, and hence the characterized dy
namics\, apply to the unknown dynamical system.
DURATION:PT1H
DTSTAMP:20240311T182540Z
DTSTART;TZID=America/New_York:20240329T153000
LAST-MODIFIED:20240311T182540Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Surrogate modeling and uncertainty quantification for inverse prob
lems and dynamical systems
UID:CAL-8a0292fd-8d13410f-018e-2ec36393-000029c8demobedework@mysite.edu
X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili
ties/Main
X-BEDEWORK-SPEAKER:Ying Hung\, Professor\, Rutgers University
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:20240226T171835Z
DESCRIPTION:In the talk I will present some of my recent works in the fiel
d of Adversarial Risk Analysis. In the first part I will talk about Adve
rsarial Classification. In multiple domains such as malware detection\, a
utomated driving systems\, or fraud detection\, classification algorithms
are susceptible to being attacked by malicious agents willing to perturb
the value of instance covariates in search of certain goals. Such proble
ms pertain to the field of adversarial machine learning and have been mai
nly dealt with\, perhaps implicitly\, through game-theoretic ideas with s
trong underlying common knowledge assumptions. These are not realistic in
numerous application domains in relation to security. We present an alte
rnative statistical framework that accounts for the lack of \nknowledge a
bout the attacker's behavior using adversarial risk analysis concepts.\n\
nIn the second part I will discuss about an adversarial risk analysis fra
mework for the software release problem. A major issue in software engine
ering is the decision of when to release a software product to the market
. This problem is complex due to\, among other things\, the uncertainty s
urrounding the software quality and its faults\, the various costs involv
ed\, and the presence of competitors. \n\nA general adversarial risk anal
ysis framework is proposed to support a software developer in deciding wh
en to release a product and showcased with an example.
DURATION:PT1H
DTSTAMP:20240311T193100Z
DTSTART;TZID=America/New_York:20240405T153000
LAST-MODIFIED:20240311T193100Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Advances in Adversarial Risk Analysis
UID:CAL-8a0292fd-8d13410f-018d-e66cefd4-00002d14demobedework@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 Whiteselll
X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us
er/Lectures_Conferences/Lecture_Talk
X-BEDEWORK-SPEAKER:Fabrizio Ruggeri\, CNR IMATI\, Milano\, Italy
X-BEDEWORK-SUBMITTEDBY:kherndon for Statistical Science (agrp_StatisticalS
cience)
X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar Series
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:20240311T195848Z
DESCRIPTION:I discuss how one can use quantum circuits to accelerate multi
proposal MCMC and point to promising avenues of future research\, includi
ng quantum HMC\, quantum-accelerated nonreversible MCMC and quantum-acce
lerated locally-balanced MCMC.
DURATION:PT1H
DTSTAMP:20240311T195848Z
DTSTART;TZID=America/New_York:20240412T153000
LAST-MODIFIED:20240311T195848Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Quantum Markov chain Monte Carlo(s)
UID:CAL-8a0292fd-8d13410f-018e-2f18a91d-00003198demobedework@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:Andrew Holbrook\, Assistant Professor\, UCLA
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:Panel/Seminar/Colloquium
CATEGORIES:Main
CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None
CREATED:20240417T131621Z
DESCRIPTION:By modeling documents as mixtures of topics\, Topic Modeling a
llows the discovery of latent thematic structures within large text corpo
ra\, and has played an important role in natural language processing over
the past decades. Beyond text data\, topic modeling has proven itself ce
ntral to the analysis of microbiome data\, population genetics\, or\, mor
e recently\, single-cell spatial transcriptomics. Given the model's exten
sive use\, the development of estimators - particularly those capable of
leveraging known structure in the data - presents a compelling challenge.
\nIn this talk\, we focus more specifically on the probabilistic Latent S
emantic Indexing model\, which assumes that the expectation of the corpus
matrix is low-rank and can be written as the product of a topic-word mat
rix and a word-document matrix. Although various estimators of the topic
matrix have recently been proposed\, their error bounds highlight a numbe
r of data regimes in which the error can grow substantially - particularl
y in the case where the size of the dictionary p is large.\nIn this talk\
, we propose studying the estimation of the topic-word matrix under the a
ssumption that the ordered entries of its columns rapidly decay to zero.
This sparsity assumption is motivated by the empirical observation that t
he word frequencies in a text often adhere to Zipf's law. We introduce a
new spectral procedure for estimating the topic-word matrix that threshol
ds words based on their corpus frequencies\, and show that its ℓ1-error r
ate under our sparsity assumption depends on the vocabulary size p only v
ia a logarithmic term. Our error bound is valid for all parameter regimes
and in particular for the setting where p is extremely large\; Our proce
dure also empirically performs well relative to well-established methods
when applied to a large corpus of research paper abstracts\, as well as t
he analysis of single-cell and microbiome data where the same statistical
model is relevant but the parameter regimes are vastly different.
DURATION:PT1H
DTSTAMP:20240417T131621Z
DTSTART;TZID=America/New_York:20240419T153000
LAST-MODIFIED:20240417T131621Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Sparse topic modeling via spectral decomposition and thresholding
UID:CAL-8a0292fd-8d13410f-018e-ec33818b-00006353demobedework@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:Claire Donnat\, Department of Statistics\, University o
f Chicago
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:Other
CATEGORIES:Utilities
CATEGORIES:Meeting
CATEGORIES:Main
CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo
ri
CREATED:20240531T141620Z
DESCRIPTION:The Statistical Science Department encourages all to attend th
e defense of this dissertation.
DURATION:PT2H
DTSTAMP:20240531T141620Z
DTSTART;TZID=America/New_York:20240607T100000
LAST-MODIFIED:20240531T141620Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Utilizing Network Structure to Flexibly Model Areal Data
UID:CAL-8a008ae5-8f05e4c1-018f-cf02392b-000050e6demobedework@mysite.edu
X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili
ties/Main
X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Meeting:/user/public-user/Ot
her/Meeting
X-BEDEWORK-SPEAKER:Michael Christensen
X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc
e)
END:VEVENT
BEGIN:VEVENT
CATEGORIES:Other
CATEGORIES:Utilities
CATEGORIES:Meeting
CATEGORIES:Main
CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo
ri
CREATED:20240611T171143Z
DESCRIPTION:The Statistical Science Department invites all to attend the d
efense of this dissertation.
DURATION:PT2H
DTSTAMP:20240611T171143Z
DTSTART;TZID=America/New_York:20240711T110000
LAST-MODIFIED:20240611T171143Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Graphical and Isoperimetric Perspectives in Sampling and Regulariz
ation
UID:CAL-8a008ae5-8f05e4c1-0190-0848bfea-00000169demobedework@mysite.edu
X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili
ties/Main
X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Meeting:/user/public-user/Ot
her/Meeting
X-BEDEWORK-SPEAKER:Ed Tam
X-BEDEWORK-SUBMITTEDBY:lr3 for Statistical Science (agrp_StatisticalScienc
e)
END:VEVENT
BEGIN:VEVENT
CATEGORIES:Other
CATEGORIES:Utilities
CATEGORIES:Meeting
CATEGORIES:Main
CONTACT;X-BEDEWORK-UID=18832e99-2a84ac8f-012a-85e0def0-00000038:Rauch\, Lo
ri
CREATED:20240702T203018Z
DESCRIPTION:The Statistical Science Department encourages all to attend th
e defense of this dissertation.
DURATION:PT2H
DTSTAMP:20240702T203018Z
DTSTART;TZID=America/New_York:20240716T100000
LAST-MODIFIED:20240702T203018Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Ensembling and Coordination Methods for Sequential Decision Making
Under Uncertainty
UID:CAL-8a001f84-905b348a-0190-752418e5-0000617edemobedework@mysite.edu
X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili
ties/Main
X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Meeting:/user/public-user/Ot
her/Meeting
X-BEDEWORK-SPEAKER:Joseph Lawson
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=00f1fcdb-0f068baf-010f-068baf83-00000004:None
CREATED:20240820T024539Z
DESCRIPTION:Some context and perspective: On statistical analysis with mul
tiple- or many- candidate models defining model-specific predictions and
decision recommendations. How to calibrate\, collate and combine for form
al subjective Bayesian inference and resulting decisions? \n\nSome highli
ghts: Historical contexts\, influences\, and the relevance of revisiting
"old" and broad thinking on Bayesian analysis\; Applied contexts includ
ing financial portfolios and macroeconomic policy decisions\; And\, of co
urse\, foundations and theory\, methodological advances\, and current res
earch frontiers.
DURATION:PT1H
DTSTAMP:20240821T191132Z
DTSTART;TZID=America/New_York:20240906T153000
LAST-MODIFIED:20240821T191132Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Bayesian forecasting and decisions under model uncertainty: Here
and back again (1988–2024)
UID:CAL-8a00048d-91324965-0191-6dacfe10-00004e8ddemobedework@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-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=karen.whitesell@duke.edu
:Karen Whitesell
X-BEDEWORK-SPEAKER:Mike West\, The Arts & Sciences Distinguished Professor
(Emeritus Fall 2024) of Statistics & Decision Sciences\, Duke University
X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar Series
X-BEDEWORK-SUBMITTEDBY:kherndon for Statistical Science (agrp_StatisticalS
cience)
END:VEVENT
BEGIN:VEVENT
CATEGORIES:Europe focus
CATEGORIES:Humanities
CATEGORIES:Other
CATEGORIES:Utilities
CATEGORIES:Reception
CATEGORIES:Social
CATEGORIES:Information Session
CATEGORIES:Main
CATEGORIES:Free Food and Beverages
CONTACT;X-BEDEWORK-UID=8a0870ef-538374e0-0153-a51edf9a-00007189:Thorpe-Tur
ner\, Dorothy
CREATED:20240904T152455Z
DESCRIPTION:Come mingle with teachers and fellow students\, majors & minor
s\, and learn more about the department in an informal setting.\nAsk ques
tions\, find out more about German club\, study abroad programs & more!\n
Pizza & drinks will be served!
DTEND;TZID=America/New_York:20240912T203000
DTSTAMP:20240904T154023Z
DTSTART;TZID=America/New_York:20240912T183000
LAST-MODIFIED:20240904T154023Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:German Studies Fall 2024 Teacher & Student Mixer!
UID:CAL-8a00048d-91324965-0191-bda3846f-0000775cdemobedework@mysite.edu
URL:https://german.duke.edu/events
X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili
ties/Main
X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Free Food and Beverages:/use
r/public-user/Other/Free Food and Beverages
X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Information Session:/user/pu
blic-user/Other/Information Session
X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Reception:/user/public-user/
Other/Reception
X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Social:/user/public-user/Oth
er/Social
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-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Humanities:/user/public-user
/Topics/Humanities
X-BEDEWORK-SUBMITTEDBY:dt96 for German Studies (agrp__ArtsandSciences_Germ
an)
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:20240907T134810Z
DESCRIPTION:We study how designing ballots with and without party designat
ions impacts electoral outcomes when partisan voters rely on party-order
cues to infer candidate affiliation in races without designations. If the
party orders of candidates in races with and without party designations
differ\, these voters might cast their votes incorrectly. We identify a q
uasi-randomized natural experiment with contest-level treatment assignmen
t pertaining to North Carolina judicial elections and use double machine
learning to accurately capture the magnitude of such incorrectly cast vot
es. Using precinct-level election and demographic data\, we estimate that
12.08% (95% confidence interval: 4.95%\, 19.20%) of democratic partisan
voters and 13.63% (95% confidence interval: 5.14%\, 22.10%) of republican
partisan voters cast their votes incorrectly due to the difference in pa
rty orders. Our results indicate that ballots mixing contests with and wi
thout party designations mislead many voters\, leading to outcomes that d
o not reflect true voter preferences. To accurately capture voter intent\
, such ballot designs should be avoided. (Joint work with Alexandre Bello
ni\, Fei Fang\, and Sasa Pekec.)
DURATION:PT1H
DTSTAMP:20240907T135915Z
DTSTART;TZID=America/New_York:20240913T153000
LAST-MODIFIED:20240907T135915Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Ballot design and electoral outcomes: The role of candidate order
and party affiliation
UID:CAL-8a00048d-91324965-0191-ccbe0310-000047f8demobedework@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@gmail.co
m:Karen Whitesell
X-BEDEWORK-SPEAKER:Alessandro Arlotto\, Associate Professor of Business Ad
ministration and Mathematics. Decision Sciences\, The Fuqua School of Bus
iness\, Duke University
X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar Series
X-BEDEWORK-SUBMITTEDBY:kherndon for Statistical Science (agrp_StatisticalS
cience)
END:VEVENT
BEGIN:VEVENT
CATEGORIES:Lectures/Conferences
CATEGORIES:Utilities
CATEGORIES:Panel/Seminar/Colloquium
CATEGORIES:Main
CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None
CREATED:20240821T214513Z
DESCRIPTION:Rubrics are thought to improve quality and decrease social bia
s in scientific peer review. However\, rubrics cannot serve these functio
ns if reviewers sequence their judgments in a normatively backwards order
. If reviewers determine the overall merit of a submission before scoring
for specific criteria\, criteria scores serve as post hoc rationalizatio
ns that can\, intentionally or unintentionally\, mask intellectual and so
cial biases. Despite the importance of proper sequencing in rubric review
and the wide adoption of rubrics in high-stakes peer review contexts\, t
here is little to no research on the order with which reviewers score rub
ric elements in practice. Using a large dataset of preliminary scores for
R01 proposals submitted to the National Institutes of Health (NIH) in fi
scal years 2014-2016\, we employ causal discovery methodology to investig
ate the causal direction with which assigned reviewers tended to score cr
iteria (Significance\, Investigator(s)\, Innovation\, Approach\, and Envi
ronment) and Overall Impact before panel discussion. Strikingly\, we find
that Overall Impact tends to be evaluated before Approach - which focuse
s on scientific strategy\, methodology\, analyses\, and feasibility. We a
lso find that Investigator and Environment tend to be evaluated first\, b
efore evaluations of scientific criteria relevant to the content of the p
roposed research. This evidence stresses the importance of structuring an
d sequencing rubric review processes to minimize the potential for normat
ively backwards assessment. \nThis is joint work with Carole J. Lee\, Fan
Xia\, Kwun C. G. Chan\, Sheridan Grant\, and Thomas S. Richardson.\n\n*T
his seminar will not be recorded
DURATION:PT1H
DTSTAMP:20240919T150247Z
DTSTART;TZID=America/New_York:20240920T153000
LAST-MODIFIED:20240919T150247Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Normatively Backwards Rubric Scoring: Evidence from NIH Peer Revie
w
UID:CAL-8a00048d-91324965-0191-76e6a86e-0000385cdemobedework@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-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=Karen.whitesell@duke.edu
:Karen Whitesell
X-BEDEWORK-SPEAKER: Elena Erosheva\, Professor\, University of Washington
X-BEDEWORK-DUKE-SERIES:Statistical Science Seminar Series
X-BEDEWORK-SUBMITTEDBY:kherndon for Statistical Science (agrp_StatisticalS
cience)
END:VEVENT
BEGIN:VEVENT
CATEGORIES:Lectures/Conferences
CATEGORIES:Utilities
CATEGORIES:Panel/Seminar/Colloquium
CATEGORIES:Main
CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None
CREATED:20240815T140730Z
DESCRIPTION:We introduce the BREASE framework for the Bayesian analysis of
randomized controlled trials with a binary treatment and a binary outcom
e. Approaching the problem from a causal inference perspective\, we propo
se parameterizing the likelihood in terms of the baseline risk\, efficacy
\, and adverse side effects of the treatment\, along with a flexible\, ye
t intuitive and tractable jointly independent beta prior distribution on
these parameters\, which we show to be a generalization of the Dirichlet
prior for the joint distribution of potential outcomes. Our approach has
a number of desirable characteristics when compared to current mainstream
alternatives: (i) it naturally induces prior dependence between expected
outcomes in the treatment and control groups\; (ii) as the baseline risk
\, efficacy and risk of adverse side effects are quantities commonly pres
ent in the clinicians' vocabulary\, the hyperparameters of the prior are
directly interpretable\, thus facilitating the elicitation of prior knowl
edge and sensitivity analysis\; and (iii) we provide analytical formulae
for the marginal likelihood\, Bayes factor\, and other posterior quantiti
es\, as well as exact posterior sampling via simulation\, in cases where
traditional MCMC fails. Empirical examples demonstrate the utility of our
methods for estimation\, hypothesis testing\, and sensitivity analysis o
f treatment effects.
DURATION:PT1H
DTSTAMP:20240815T140730Z
DTSTART;TZID=America/New_York:20240927T153000
LAST-MODIFIED:20240815T140730Z
LOCATION;X-BEDEWORK-UID=18832edc-1b27e154-011b-28365daf-0000006c:Old Chemi
stry 116
STATUS:CONFIRMED
SUMMARY:Causally Sound Priors for Binary Experiments
UID:CAL-8a00048d-91324965-0191-565d754e-00005e3edemobedework@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:Carlos Cinelli\, Assistant Professor\, University of Wa
shington
X-BEDEWORK-DUKE-SERIES:Statistlcal Science Seminar Series
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: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: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
END:VCALENDAR