BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:Sampling and Generative Modeling Using Dynamical Representatio
 ns of Transport
X-WR-TIMEZONE:Eastern Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20260520T163221Z
UID:tag:localist.com\,2008:EventInstance_47914906662790
DTSTART:20241031T180000Z
DTEND:20241031T190000Z
DESCRIPTION:Widely Applied Mathematics presents...\n\n"Sampling and Generat
 ive Modeling Using Dynamical Representations of Transport"\n\nProfessor Yo
 ussef Marzouk\, Massachusetts Institute of Technology\n\n \n\nThursday\, O
 ctober 31\n2:00 - 3:00pm\nPierce 301\n\n \n\nDrawing samples from a probab
 ility distribution is a central task in applied mathematics\, statistics\,
  and machine learning—with applications ranging from Bayesian computatio
 n to computational chemistry and generative modeling. Many powerful tools 
 for sampling employ transportation of measure\, where the essential idea i
 s to couple the target probability distribution with a simple\, tractable 
 “reference" distribution\, and to use this coupling (which may be determ
 inistic or stochastic) to generate new samples.\n\nWithin this broad area\
 , an emerging class of methods use dynamics to define a transport incremen
 tally\, e.g.\, via the flow map induced by trajectories of an ODE. These m
 ethods have shown great empirical success\, but their consistency and conv
 ergence properties\, and the ways in which they can exploit structure in t
 he underlying distributions\, are less well understood.  We will discuss p
 roperties and theoretical underpinnings of these new dynamical approaches 
 to transport. In particular\, we will discuss the statistical convergence 
 of generative models based on neural ODEs. We will also present a new dyna
 mical construction of transport: a gradient-free method which avoids compl
 ex training procedures by instead evolving an interacting particle system 
 that approximates a Fisher–Rao gradient flow. We will attempt to illumin
 ate the relative advantages and pitfalls of these dynamical methods\, and 
 the many design choices that they entail.\n\n\nSpeaker: Youssef Marzouk is
  a professor in the Department of Aeronautics and Astronautics at MIT and 
 co-director of the MIT Center for Computational Science and Engineering. H
 e is also a core member of MIT’s Statistics and Data Science Center and 
 director of MIT’s Aerospace Computational Design Laboratory. His researc
 h interests lie at the intersection of computation and statistical inferen
 ce with physical modeling. He develops new methodologies for uncertainty q
 uantification\, Bayesian modeling and computation\, data assimilation\, ex
 perimental design\, and machine learning in complex physical systems. His 
 methodological work is motivated by a wide variety of engineering and envi
 ronmental applications. He is an avid coffee drinker and occasional classi
 cal pianist.
GEO:42.3783;-71.117081
LOCATION:Pierce Hall\, 301
SUMMARY:Sampling and Generative Modeling Using Dynamical Representations of
  Transport
URL;VALUE=URI:https://events.seas.harvard.edu/event/sampling-and-generative
 -modeling-using-dynamical-representations-of-transport
CATEGORIES:Colloquia / Seminar / Lecture
END:VEVENT
END:VCALENDAR