BEGIN:VCALENDAR
VERSION:2.0
PRODID:researchseminars.org
CALSCALE:GREGORIAN
X-WR-CALNAME:researchseminars.org
BEGIN:VEVENT
SUMMARY:Ery Arias-Castro (UC San Diego)
DTSTART;VALUE=DATE-TIME:20200417T150000Z
DTEND;VALUE=DATE-TIME:20200417T161200Z
DTSTAMP;VALUE=DATE-TIME:20201029T102900Z
UID:sss/1
DESCRIPTION:Title: On using graph distances to estimate Euclidean and rela
ted distances\nby Ery Arias-Castro (UC San Diego) as part of Stochastics a
nd Statistics Seminar Series\n\n\nAbstract\nGraph distances have proven qu
ite useful in machine learning/statistics\, particularly in the estimation
of Euclidean or geodesic distances. The talk will include a partial revie
w of the literature\, and then present more recent developments on the est
imation of curvature-constrained distances on a surface\, as well as on th
e estimation of Euclidean distances based on an unweighted and noisy neigh
borhood graph.\n
END:VEVENT
BEGIN:VEVENT
SUMMARY:Sébastien Bubeck (Microsoft Research)
DTSTART;VALUE=DATE-TIME:20200424T150000Z
DTEND;VALUE=DATE-TIME:20200424T160000Z
DTSTAMP;VALUE=DATE-TIME:20201029T102900Z
UID:sss/2
DESCRIPTION:Title: How to Trap a Gradient Flow\nby Sébastien Bubeck (Micr
osoft Research) as part of Stochastics and Statistics Seminar Series\n\n\n
Abstract\nIn 1993\, Stephen A. Vavasis proved that in any finite dimension
\, there exists a faster method than gradient descent to find stationary p
oints of smooth non-convex functions. In dimension 2 he proved that 1/eps
gradient queries are enough\, and that 1/sqrt(eps) queries are necessary.
We close this gap by providing an algorithm based on a new local-to-global
phenomenon for smooth non-convex functions. Some higher dimensional resul
ts will also be discussed. I will also present an extension of the 1/sqrt(
eps) lower bound to randomized algorithms\, mainly as an excuse to discuss
some beautiful topics such as Aldous’ 1983 paper on local minimization
on the cube\, and Benjamini-Pemantle-Peres’ 1998 construction of unpredi
ctable walks.\n\nJoint work with Dan Mikulincer\n
END:VEVENT
BEGIN:VEVENT
SUMMARY:Alexandre d'Aspremont (ENS\, CNRS)
DTSTART;VALUE=DATE-TIME:20200501T150000Z
DTEND;VALUE=DATE-TIME:20200501T160000Z
DTSTAMP;VALUE=DATE-TIME:20201029T102900Z
UID:sss/3
DESCRIPTION:Title: Naive feature selection: Sparsity in naive Bayes\nby Al
exandre d'Aspremont (ENS\, CNRS) as part of Stochastics and Statistics Sem
inar Series\n\n\nAbstract\nDue to its linear complexity\, naive Bayes clas
sification remains an attractive supervised learning method\, especially i
n very large-scale settings. We propose a sparse version of naive Bayes\,
which can be used for feature selection. This leads to a combinatorial max
imum-likelihood problem\, for which we provide an exact solution in the ca
se of binary data\, or a bound in the multinomial case. We prove that our
bound becomes tight as the marginal contribution of additional features de
creases. Both binary and multinomial sparse models are solvable in time al
most linear in problem size\, representing a very small extra relative cos
t compared to the classical naive Bayes. Numerical experiments on text dat
a show that the naive Bayes feature selection method is as statistically e
ffective as state-of-the-art feature selection methods such as recursive f
eature elimination\, l1-penalized logistic regression and LASSO\, while be
ing orders of magnitude faster. For a large data set\, having more than wi
th 1.6 million training points and about 12 million features\, and with a
non-optimized CPU implementation\, our sparse naive Bayes model can be tra
ined in less than 15 seconds. Authors: A. Askari\, A. d’Aspremont\, L.
El Ghaoui.\n
END:VEVENT
BEGIN:VEVENT
SUMMARY:Gesine Reinert (University of Oxford)
DTSTART;VALUE=DATE-TIME:20200911T150000Z
DTEND;VALUE=DATE-TIME:20200911T160000Z
DTSTAMP;VALUE=DATE-TIME:20201029T102900Z
UID:sss/4
DESCRIPTION:Title: Stein’s method for multivariate continuous distributi
ons and applications\nby Gesine Reinert (University of Oxford) as part of
Stochastics and Statistics Seminar Series\n\n\nAbstract\nStein’s method
is a key method for assessing distributional distance\, mainly for one-dim
ensional distributions. In this talk we provide a general approach to Stei
n’s method for multivariate continuous distributions. Among the applicat
ions we consider is the Wasserstein distance between two continuous probab
ility distributions under the assumption of existence of a Poincare consta
nt.\n\nThis is joint work with Guillaume Mijoule (INRIA Paris) and Yvik Sw
an (Liege).\n
END:VEVENT
BEGIN:VEVENT
SUMMARY:Caroline Uhler (MIT)
DTSTART;VALUE=DATE-TIME:20200918T150500Z
DTEND;VALUE=DATE-TIME:20200918T160500Z
DTSTAMP;VALUE=DATE-TIME:20201029T102900Z
UID:sss/5
DESCRIPTION:Title: Causal Inference and Overparameterized Autoencoders in
the Light of Drug Repurposing for SARS-CoV-2\nby Caroline Uhler (MIT) as p
art of Stochastics and Statistics Seminar Series\n\n\nAbstract\nMassive da
ta collection holds the promise of a better understanding of complex pheno
mena and ultimately\, of better decisions. An exciting opportunity in this
regard stems from the growing availability of perturbation / intervention
data (drugs\, knockouts\, overexpression\, etc.) in biology. In order to
obtain mechanistic insights from such data\, a major challenge is the deve
lopment of a framework that integrates observational and interventional da
ta and allows predicting the effect of yet unseen interventions or transpo
rting the effect of interventions observed in one context to another. I wi
ll present a framework for causal inference based on such data and particu
larly highlight the role of overparameterized autoencoders. We end by demo
nstrating how these ideas can be applied for drug repurposing in the curre
nt SARS-CoV-2 crisis.\n
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dylan Foster (MIT)
DTSTART;VALUE=DATE-TIME:20200925T150500Z
DTEND;VALUE=DATE-TIME:20200925T160500Z
DTSTAMP;VALUE=DATE-TIME:20201029T102900Z
UID:sss/6
DESCRIPTION:Title: Separating Estimation from Decision Making in Contextua
l Bandits\nby Dylan Foster (MIT) as part of Stochastics and Statistics Sem
inar Series\n\n\nAbstract\nThe contextual bandit is a sequential decision
making problem in which a learner repeatedly selects an action (e.g.\, a n
ews article to display) in response to a context (e.g.\, a user’s profil
e) and receives a reward\, but only for the action they selected. Beyond t
he classic explore-exploit tradeoff\, a fundamental challenge in contextua
l bandits is to develop algorithms that can leverage flexible function app
roximation to model similarity between contexts\, yet have computational r
equirements comparable to classical supervised learning tasks such as clas
sification and regression. To this end\, we provide the first universal an
d optimal reduction from contextual bandits to online regression. We show
how to transform any oracle for online regression with a given value funct
ion class into an algorithm for contextual bandits with the induced policy
class\, with no overhead in runtime or memory requirements. Conceptually\
, our results show that it is possible to provably separate estimation and
decision making into separate algorithmic building blocks\, and that this
can be effective both in theory and in practice. Time permitting\, I will
discuss extensions of these techniques to more challenging reinforcement
learning problems.\n
END:VEVENT
BEGIN:VEVENT
SUMMARY:Richard Nickl (University of Cambridge)
DTSTART;VALUE=DATE-TIME:20201002T150500Z
DTEND;VALUE=DATE-TIME:20201002T160500Z
DTSTAMP;VALUE=DATE-TIME:20201029T102900Z
UID:sss/7
DESCRIPTION:Title: Bayesian inverse problems\, Gaussian processes\, and pa
rtial differential equations\nby Richard Nickl (University of Cambridge) a
s part of Stochastics and Statistics Seminar Series\n\nAbstract: TBA\n
END:VEVENT
BEGIN:VEVENT
SUMMARY:Gábor Lugosi (Pompeu Fabra University)
DTSTART;VALUE=DATE-TIME:20201009T150500Z
DTEND;VALUE=DATE-TIME:20201009T160500Z
DTSTAMP;VALUE=DATE-TIME:20201029T102900Z
UID:sss/8
DESCRIPTION:Title: On Estimating the Mean of a Random Vector\nby Gábor Lu
gosi (Pompeu Fabra University) as part of Stochastics and Statistics Semin
ar Series\n\nAbstract: TBA\n
END:VEVENT
BEGIN:VEVENT
SUMMARY:Carola-Bibiane Schönlieb (University of Cambridge)
DTSTART;VALUE=DATE-TIME:20201016T150500Z
DTEND;VALUE=DATE-TIME:20201016T160500Z
DTSTAMP;VALUE=DATE-TIME:20201029T102900Z
UID:sss/9
DESCRIPTION:Title: Data driven variational models for solving inverse prob
lems\nby Carola-Bibiane Schönlieb (University of Cambridge) as part of St
ochastics and Statistics Seminar Series\n\nAbstract: TBA\n
END:VEVENT
BEGIN:VEVENT
SUMMARY:Jose Blanchet (Stanford University)
DTSTART;VALUE=DATE-TIME:20201023T150500Z
DTEND;VALUE=DATE-TIME:20201023T160500Z
DTSTAMP;VALUE=DATE-TIME:20201029T102900Z
UID:sss/10
DESCRIPTION:Title: Statistical Aspects of Wasserstein Distributionally Rob
ust Optimization Estimators\nby Jose Blanchet (Stanford University) as par
t of Stochastics and Statistics Seminar Series\n\n\nAbstract\nAbstract: Wa
sserstein-based distributional robust optimization problems are formulated
as min-max games in which a statistician chooses a parameter to minimize
an expected loss against an adversary (say nature) which wishes to maximiz
e the loss by choosing an appropriate probability model within a certain n
on-parametric class. Recently\, these formulations have been studied in th
e context in which the non-parametric class chosen by nature is defined as
a Wasserstein-distance neighborhood around the empirical measure. It turn
s out that by appropriately choosing the loss and the geometry of the Wass
erstein distance one can recover a wide range of classical statistical est
imators (including Lasso\, Graphical Lasso\, SVM\, group Lasso\, among man
y others). This talk studies a wide range of rich statistical quantities a
ssociated with these problems\; for example\, the optimal (in a certain se
nse) choice of the adversarial perturbation\, weak convergence of natural
confidence regions associated with these formulations\, and asymptotic nor
mality of the DRO estimators. (This talk is based on joint work with Y. Ka
ng\, K. Murthy\, and N. Si.)\n
END:VEVENT
BEGIN:VEVENT
SUMMARY:Daniela Witten (University of Washington)
DTSTART;VALUE=DATE-TIME:20201106T160500Z
DTEND;VALUE=DATE-TIME:20201106T170500Z
DTSTAMP;VALUE=DATE-TIME:20201029T102900Z
UID:sss/12
DESCRIPTION:by Daniela Witten (University of Washington) as part of Stocha
stics and Statistics Seminar Series\n\nAbstract: TBA\n
END:VEVENT
BEGIN:VEVENT
SUMMARY:Mary Wootters (Stanford University)
DTSTART;VALUE=DATE-TIME:20201113T160500Z
DTEND;VALUE=DATE-TIME:20201113T170500Z
DTSTAMP;VALUE=DATE-TIME:20201029T102900Z
UID:sss/13
DESCRIPTION:Title: Sharp Thresholds for Random Subspaces\, and Application
s\nby Mary Wootters (Stanford University) as part of Stochastics and Stati
stics Seminar Series\n\n\nAbstract\nAbstract: What combinatorial propertie
s are likely to be satisfied by a random subspace over a finite field? For
example\, is it likely that not too many points lie in any Hamming ball?
What about any cube? We show that there is a sharp threshold on the dimen
sion of the subspace at which the answers to these questions change from
“extremely likely” to “extremely unlikely\,” and moreover we give
a simple characterization of this threshold for different properties. Our
motivation comes from error correcting codes\, and we use this characteriz
ation to make progress on the questions of list-decoding and list-recovery
for random linear codes\, and also to establish the list-decodability of
random Low Density Parity-Check (LDPC) codes.\n\nThis talk is based on the
joint works with Venkatesan Guruswami\, Ray Li\, Jonathan Mosheiff\, Nico
las Resch\, Noga Ron-Zewi\, and Shashwat Silas.\nEvent Navigation\n
END:VEVENT
BEGIN:VEVENT
SUMMARY:Arnaud Doucet (University of Oxford)
DTSTART;VALUE=DATE-TIME:20201120T160500Z
DTEND;VALUE=DATE-TIME:20201120T170500Z
DTSTAMP;VALUE=DATE-TIME:20201029T102900Z
UID:sss/14
DESCRIPTION:by Arnaud Doucet (University of Oxford) as part of Stochastics
and Statistics Seminar Series\n\nAbstract: TBA\n
END:VEVENT
BEGIN:VEVENT
SUMMARY:Rong Ge (Duke University)
DTSTART;VALUE=DATE-TIME:20201204T160500Z
DTEND;VALUE=DATE-TIME:20201204T170500Z
DTSTAMP;VALUE=DATE-TIME:20201029T102900Z
UID:sss/15
DESCRIPTION:by Rong Ge (Duke University) as part of Stochastics and Statis
tics Seminar Series\n\nAbstract: TBA\n
END:VEVENT
END:VCALENDAR