BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Universal Learning for Individual Data
DTSTART:20190128T110000
DTEND:20190128T120000
DTSTAMP:20260407T043630Z
UID:6ddab3f3bafcc3714a65084d5d9952cdc24243b0e897f01dab75ea8a
CATEGORIES:Conferences - Seminars
DESCRIPTION:Prof. Meir Feder Meir Feder is a Professor at the School of El
 ectrical Engineering\, Tel-Aviv University\, the incumbent of the Informat
 ion Theory Chair. An internationally recognized authority in signal proces
 sing\, communication and information theory\, Professor Feder holds Sc.D. 
 degree from the Massachusetts Institute of Technology (MIT) was a visiting
  Professor in MIT\, and had visiting positions at Bell laboratories and Sc
 ripps Institute of Oceanography. He is an IEEE Fellow\, and received sever
 al academic awards including the IEEE Information Theory society best pape
 r award.\nUniversal learning is considered from an information theoretic p
 oint of view following the universal prediction approach pursued in the 90
 's by F&Merhav. Interestingly\, the extension to learning is not straight-
 forward. In previous works we considered on-line learning and supervised l
 earning in a stochastic setting. Yet\, the most challenging case is batch 
 learning where prediction is done on a test sample once the entire trainin
 g data is observed\, in the individual setting where the features and labe
 ls\, both training and test\, are specific individual quantities. This wor
 k provides schemes that for any individual data compete with a "genie" (or
  reference) that knows the true test label. It suggests design criteria an
 d derive the corresponding universal learning schemes. The main proposed s
 cheme is termed Predictive Normalized Maximum Likelihood (pNML). As demons
 trated\, pNML learning and its variations provide robust\, "stable" learni
 ng solutions that outperforms the current leading approach based on Empiri
 cal Risk Minimization (ERM). Furthermore\, the pNML construction provides 
 a pointwise indication for the learnability that measures the uncertainty 
 in learning the specific test challenge with the given training examples\;
  thus letting the learner know when it does not know. Joint work with Yani
 v Fogel and Koby Bibas
LOCATION:INR 113 https://plan.epfl.ch/?room=INR113
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR