BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Active Learning - Ferenc Huszar (Budapest University of Technology
  and Economics)
DTSTART:20120412T130000Z
DTEND:20120412T143000Z
UID:TALK37504@talks.cam.ac.uk
CONTACT:Konstantina Palla
DESCRIPTION:This RCC is going to be about active learning. I will start by
  giving some motivating examples for where active learning can be useful i
 n experimental sciences and in large-scale machine learning applications. 
 I will draw a rough taxonomy of active learning methods\, mentioning the d
 ifference between transductive and inductive active learning\, loss-orient
 ed vs. information theoretic approaches. I will introduce a simple toy mod
 el for linearly separable binary classification\, and use it to illustrate
  the idea behind different approaches to information theoretic active lear
 ning. I will talk in detail about the methods proposed by Tong and Koller 
 (2001)\, resulting in one of the most highly cited papers in machine learn
 ing\, and contrast it with related methods based on as query by committee.
  I'm also going to touch upon relatively recent theoretical results by Ste
 ve Hanneke on the fast rates of convergence certain active learning method
 s can achieve.\n\nRecommended reading\nhttp://jmlr.csail.mit.edu/papers/vo
 lume2/tong01a/tong01a.pdf\nhttp://www.stat.cmu.edu/~shanneke/docs/2009/act
 ive-rates-annals.pdf
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR