BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Does data interpolation contradict statistical optimality? - Alexa
 ndre Tsybakov (CREST: Centre de Recherche en Économie et Statistique)
DTSTART:20180629T080000Z
DTEND:20180629T084500Z
UID:TALK107518@talks.cam.ac.uk
CONTACT:INI IT
DESCRIPTION:We exhibit estimators that interpolate the data\, and yet achi
 eve optimal rates of convergence for the problems of nonparametric regress
 ion and prediction with square loss. This curious observation goes against
  the usual intuition that a good statistical procedure should forego the e
 xact fit to data in favor of a more smooth representation.  A motivation f
 or this work is the recent focus within the machine learning community on 
 the out-of-sample performance of neural networks. These flexible models ar
 e typically trained to fit the data exactly (either in their sign or in th
 e actual value)\, and yet they have good prediction behaviour. This talk i
 s based on joint work with Misha Belkin and Sasha Rakhlin.
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR