BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Unsupervised Risk Estimation with only Structural Assumptions - Ja
 cob Steinhardt (Stanford University)
DTSTART:20160316T140000Z
DTEND:20160316T150000Z
UID:TALK64869@talks.cam.ac.uk
CONTACT:Zoubin Ghahramani
DESCRIPTION:Given a model θ and unlabeled samples from a distribution p \
 , we show how to estimate the labeled risk of θ while only making structu
 ral (i.e.\, conditional independence) assumptions about p. This lets us es
 timate a model’s test error on distributions very different than its tra
 ining distribution\, thus performing unsupervised domain adaptation even w
 ithout assuming the true predictor remains constant (covariate shift). Fur
 thermore\, we can perform discriminative semi-supervised learning\, even u
 nder model mis-specification. Our technical tool is the method of moments\
 , which allows us to exploit conditional independencies without relying on
  a specific parametric model. Finally\, we introduce a new theoretical fra
 mework for grappling with the non-identifiability of the class identities 
 fundamental to unsupervised learning.
LOCATION:Engineering Department\, CBL Room BE-438
END:VEVENT
END:VCALENDAR