BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Non-parametric Bayesian Learning of User Preferences: Elicitation\
 , Sparsification and Beyond - Edwin Bonilla (NICTA/ANU)
DTSTART:20120704T100000Z
DTEND:20120704T110000Z
UID:TALK38822@talks.cam.ac.uk
CONTACT:Novi Quadrianto
DESCRIPTION:Bayesian approaches to preference elicitation (PE) are particu
 larly attractive due to their ability to explicitly model uncertainty in u
 sers' latent utility functions. However\, previous approaches to Bayesian 
 PE have ignored the important problem of generalizing from previous users 
 to an unseen user in order to reduce the elicitation burden on new users. 
 In this talk\, I will present an approach that addresses this deficiency b
 y introducing a Gaussian Process (GP) prior over users' latent utility fun
 ctions on the joint space of user and item features. The hyper-parameters 
 of this GP are learned on a set of preferences of previous users and this 
 information is leveraged to aid in the elicitation process for a new user.
  This approach provides a flexible model of a multi-user utility function\
 , facilitates an efficient value of information (VOI) heuristic query sele
 ction strategy\, and provides a principled way to incorporate the elicitat
 ions of multiple users back into the model. I will show the effectiveness 
 of this method in comparison to previous work on a real dataset of user pr
 eferences over sushi types. Additionally\, I will present recent work on l
 oss-sensitive sparsification approaches to scale up GP-based preference mo
 dels.
LOCATION:Engineering Department\, CBL Room BE-438
END:VEVENT
END:VCALENDAR