BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1628
DTSTAMP:20251014T090649Z
SUMMARY:Generalization Bounds for Dependent Data using Online-to-Batch Conv
 ersion
DESCRIPTION:Speaker: Sagnik Chatterjee (IIIT Delhi)\n\nAbstract: \n\nIn thi
 s work\, we upper bound the generalization error of batch learning algorit
 hms trained on samples drawn from a mixing stochastic process (i.e.\, a de
 pendent data source) both in expectation and with high probability. Unlike
  previous results by Mohri et al. (JMLR 2010) and Fu et al. (ICLR 2023)\, 
 our work does not require any additional stability assumptions on the batc
 h learner itself. This is made possible due to our use of the Online-to-Ba
 tch ( OTB ) conversion framework\, which allows us to shift the burden of 
 stability from the batch learner to an artificially constructed online lea
 rner. We prove that our bounds are equal to the bounds in the i.i.d. setti
 ng (i.e.\, optimal) up to a term that depends on the decay rate of the und
 erlying mixing stochastic process. Central to our analysis is a new notion
  of algorithmic stability for online learning algorithms based on Wasserst
 ein distances of order one. Furthermore\, we prove that the EWA algorithm\
 , a textbook family of online learning algorithms\, satisfies our new noti
 on of stability.\nShort Bio:Sagnik Chatterjee recently obtained his Ph.D. 
 from IIIT Delhi\, where he was advised by Dr. Debajyoti Bera. Prior to his
  Ph.D.\, Sagnik was a software engineer at Oracle in Bangalore. Sagnik's m
 ain research interest is in Quantum Computing with a particular focus on L
 earning Theory and Shallow Circuit Complexity. \n
URL:https://www.tcs.tifr.res.in/web/events/1628
DTSTART;TZID=Asia/Kolkata:20251021T160000
DTEND;TZID=Asia/Kolkata:20251021T170000
LOCATION:via Zoom in A201
END:VEVENT
END:VCALENDAR