BEGIN:VCALENDAR
VERSION:2.0
PRODID:researchseminars.org
CALSCALE:GREGORIAN
X-WR-CALNAME:researchseminars.org
BEGIN:VEVENT
SUMMARY:Omar Montasser (TTIC)
DTSTART:20201028T170000Z
DTEND:20201028T180000Z
DTSTAMP:20260423T020956Z
UID:TCSPlus/15
DESCRIPTION:Title: <a href="https://researchseminars.org/talk/TCSPlus/15/"
 >Adversarially Robust Learnability: Characterization and Reductions</a>\nb
 y Omar Montasser (TTIC) as part of TCS+\n\n\nAbstract\nWe study the questi
 on of learning an adversarially robust predictor from uncorrupted samples.
  We show that any VC class H is robustly PAC learnable\, but we also show 
 that such learning must sometimes be improper (i.e. use predictors from ou
 tside the class)\, as some VC classes are not robustly properly learnable.
   In particular\, the popular robust empirical risk minimization approach 
 (also known as adversarial training)\, which is proper\, cannot robustly l
 earn all VC classes.  After establishing learnability\, we turn to ask whe
 ther having a tractable non-robust learning algorithm is sufficient for tr
 actable robust learnability and give a reduction algorithm for robustly le
 arning any hypothesis class H using a non-robust PAC learner for H\, with 
 nearly-optimal oracle complexity.\n\nThis is based on joint work with Stev
 e Hanneke and Nati Srebro\, available at https://arxiv.org/abs/1902.04217 
 and https://arxiv.org/abs/2010.12039..\n
LOCATION:https://researchseminars.org/talk/TCSPlus/15/
END:VEVENT
END:VCALENDAR