BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1550
DTSTAMP:20250425T065111Z
SUMMARY:Rare events in stochastic multi-armed bandits
DESCRIPTION:Speaker: Anirban Bhattacharjee (TIFR)\n\nAbstract: \nThis synop
 sis summarizes the problems addressed in the dissertation and the main res
 ults obtained in the course of studying these problems. We examine stochas
 tic multi-armed bandit (MAB) problems in rare event regimes with emphasis 
 on Best Arm Identification (BAI) and also touch upon regret minimization.
  When arm rewards occur infrequently but are high in magnitude\, we develo
 p algorithms for BAI which are based on Poisson approximation and drastic
 ally reduce computational effort at the cost of negligible increase in sa
 mple complexity. For identifying the safest system among a given set of s
 afety-critical systems with rare failures\, we consider simulation models 
 of the systems\, and simulate from them following BAI methods. Reasonably
  accurate approximations of the lower bound on sample complexity reveal t
 hat sample complexity depends on the failure rate of the secondbest system
 \, as opposed to the best. Further\, standard regret minimization algorith
 ms are shown to perform poorly in rare-event regimes where rewards are hi
 gh-value but rarely seen\, necessitating scaled modifications that ensure
  optimality\n
URL:https://www.tcs.tifr.res.in/web/events/1550
DTSTART;TZID=Asia/Kolkata:20250425T160000
DTEND;TZID=Asia/Kolkata:20250425T170000
LOCATION:via Zoom in A201
END:VEVENT
END:VCALENDAR