BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1625
DTSTAMP:20251028T062631Z
SUMMARY:A Model predictive control approach for asymptotic optimality of Re
 stless Multi-armed Bandits
DESCRIPTION:Speaker: Dheeraj Narasimha (University of Grenoble\, France)\n\
 nAbstract: \nThe classical restless multi-armed bandit problem comprises o
 f $N$ bandit arms (slot machines) each with an associated state. A player 
 chooses a collection of at most $\\alpha N $ arms to play\, the remaining 
 arms are left untouched. At the next instance each arm evolves to a new st
 ate according to a stochastic transition process leaving the player with a
  reward depending on the player's action. Concretely\, the arms provide a 
 reward that is dependent on both the state and action of the arm at each t
 ime instance. This reward process is repeated over an infinite time horizo
 n. The player wishes to maximize the average reward obtained from this pro
 cess. Restless Multi-armed Bandits are known to be PSPACE hard combinatori
 al problems. Such problems find applications in a number of different fiel
 ds including wireless communication\, clinical trial design and web-crawl
 ing.This problem was first proposed by Whittle in 1988\, he conjectured t
 hat a priority order based on what is called the Whittle's index will prov
 ide an asymptotically optimal solution to this problem under a condition 
 called \\emph{indexability}. Unfortunately\, we now know of counter-examp
 les where this is not true. In this talk I will introduce a relaxation to
  the RMAB problem with a corresponding extension to the Whittle index pol
 icy that provably gives us asymptotically optimal solutions under easy to
  verify conditions.\nShort Bio:Dheeraj Narasimha obtained his PhD degree f
 rom Arizona State University under Prof. Lei Ying in 2021 while working on
  mean-field ames for density dependent continuous time markov processes. H
 e then went on to do a post-doctoral fellowship under Prof. Srinivas Shakk
 ottai at Texas A & M University and then completed a second post-doc at I
 NRIA\, CNRS under Dr. Nicolas Gast where he worked on restless multi-arme
 d bandits . His work on mean-field games has won runner-up best-paper priz
 es in ACM mobihoc as well as most recently in IEEE INFOCOM.The main focus
  of his work revolves around solving decision problems with many particip
 ants under weak interaction\, be it through weak graph limits\, potential
  games or mean field structures.\n
URL:https://www.tcs.tifr.res.in/web/events/1625
DTSTART;TZID=Asia/Kolkata:20251028T160000
DTEND;TZID=Asia/Kolkata:20251028T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
END:VCALENDAR