BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20250811T012315EDT-77875zBSUE@132.216.98.100
DTSTAMP:20250811T052315Z
DESCRIPTION:Title:\n\nOverview of adaptive stochastic optimization methods
 \n\nAbstract: \n\nRecently a variety of stochastic variants of adaptive me
 thods have been developed and analyzed. These include stochastic step sear
 ch\, trust region and cubicly regularized Newton methods. Such methods ada
 pt the step size parameter and use it to dictate the accuracy required or 
 stochastic approximations. The requirements on stochastic approximations a
 re\, thus\, also adaptive and in principle can be biased and even inconsis
 tent. The step size parameters in these methods can increase and decrease 
 based on the perceived progress\, but unlike the deterministic case they a
 re not bounded away from zero. This creates obstacles in complexity analys
 is of such methods. We will show how by viewing such algorithms as stochas
 tic processes with martingale behavior we can derive bounds on expected co
 mplexity that also apply in high probability. We also show that it is poss
 ible to derive a lower bound on step size parameters in high probability f
 or the methods in this general framework. We will discuss various stochast
 ic settings\, where the framework easily applies\, such as expectation min
 imization\, black box and simulation optimization\, expectation minimizati
 on with corrupt samples\, etc.\n\n \n
DTSTART:20230313T203000Z
DTEND:20230313T213000Z
LOCATION:Room 1104\, Burnside Hall\, CA\, QC\, Montreal\, H3A 0B9\, 805 rue
  Sherbrooke Ouest
SUMMARY:Katya Scheinberg (Cornell University)
URL:/mathstat/channels/event/katya-scheinberg-cornell-
 university-346629
END:VEVENT
END:VCALENDAR