BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/809
DTSTAMP:20230914T125939Z
SUMMARY:Accelerating Stochastic Gradient Descent
DESCRIPTION:Speaker: Praneeth Netrapalli (Microsoft Research\n"Vigyan"\, #9
 \, Lavelle Road\nShanthala Nagar\, Ashok Nagar\nBangalore 560001)\n\nAbstr
 act: \nThere is widespread sentiment that it is not possible to effectivel
 y utilize fast gradient methods (e.g. Nesterov's acceleration\, conjugate 
 gradient\, heavy ball) for the purposes of stochastic optimization due to 
 their instability and error accumulation\, a notion made precise in d'Aspr
 emont 2008 and Devolder\, Glineur\, and Nesterov 2014. This work considers
  these issues for the special case of stochastic approximation for the lea
 st squares regression problem\, and our main result refutes the convention
 al wisdom by showing that acceleration can be made robust to statistical e
 rrors. In particular\, this work introduces an accelerated stochastic grad
 ient method that provably achieves the minimax optimal statistical risk fa
 ster than stochastic gradient descent. Critical to the analysis is a sharp
  characterization of accelerated stochastic gradient descent as a stochast
 ic process. We hope this characterization gives insights towards the broad
 er question of designing simple and effective accelerated stochastic metho
 ds for more general convex and non-convex optimization problems (joint wor
 k with Prateek Jain\, Sham Kakade\, Rahul Kidambi and Aaron Sidford).\n \
 n
URL:https://www.tcs.tifr.res.in/web/events/809
DTSTART;TZID=Asia/Kolkata:20170919T160000
DTEND;TZID=Asia/Kolkata:20170919T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
END:VCALENDAR