BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1107
DTSTAMP:20230914T125950Z
SUMMARY:Accelerating Black Box Estimation of Distribution Tails Using Self 
 structuring Importance Samplers
DESCRIPTION:Speaker: Anand Deo\n\nAbstract: \nMotivated by the increasing a
 doption of models which facilitate greater automation in risk management a
 nd decision-making\, this talk presents a novel Importance Sampling (IS) s
 cheme for estimating distribution tails for a rich class of objectives mod
 elled with tools such as mixed integer linear programs\, deep neural netwo
 rks\, etc. A key challenge with the conventional efficient sampling approa
 ches in these settings is the need to intricately tailor the samplers base
 d on the underlying probability distribution and the objective. This chall
 enge is overcome in the proposed black-box scheme by automating the select
 ion of an effective IS density with a transformation that implicitly learn
 s and replicates the concentration properties observed in less rare sample
 s.  Despite its simple and scalable implementation\, this self structuring
  IS scheme achieves asymptotically optimal variance reduction across a spe
 ctrum of multivariate distributions involving light as well as heavy tails
 .\n\nThis approach is guided by a large deviations principle that brings o
 ut the phenomenon of self-similarity of optimal IS distributions in consid
 erable generality.  In addition to helping certify variance reduction\, th
 e large deviations principle serves as a tool for readily yielding new tai
 l risk asymptotics and algorithms in settings such as distribution network
 s.\n\nThis is joint work with Karthyek Murthy (SUTD).\n\nZoom link: https:
 //zoom.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1107
DTSTART;TZID=Asia/Kolkata:20201218T171500
DTEND;TZID=Asia/Kolkata:20201218T181500
END:VEVENT
END:VCALENDAR