BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:New Methods in Bayesian Optimization for Machine Learning - Jasper
  Snoek (Harvard University)
DTSTART:20140925T100000Z
DTEND:20140925T110000Z
UID:TALK54245@talks.cam.ac.uk
CONTACT:Dr Jes Frellsen
DESCRIPTION:Bayesian optimization is a methodology for the global optimiza
 tion of expensive\, noisy and multimodal black-box functions.  When applie
 d to the optimization of hyperparameters and model parameters of machine l
 earning algorithms\, this provides a reproducible and efficient methodolog
 y for running experiments that often outperforms domain experts.  This tal
 k will highlight our recent work on tailoring Bayesian optimization to pro
 blems in machine learning.  Developing a principled statistical framework 
 in the context of Bayesian optimization allows us to reason about useful a
 dditional information and model the specific structure of the problem.  Pe
 rtaining to the former\, I will discuss Bayesian optimization with multipl
 e related tasks and reasoning about various forms of constraints.  Then I 
 will talk about our recent work on using a model tailored to optimization 
 curves to forecast the result of a training run long before it is finished
 .  This allows us to reason about when to stop training a model\, when to 
 start training a new one\, and when to revisit an already partially traine
 d one.  If time permits\, I will also discuss our recent efforts in reachi
 ng out to the wider scientific community to optimize expensive experiments
  in various other domains such as robotics and aeronautics.
LOCATION:Engineering Department\, CBL Room BE-438
END:VEVENT
END:VCALENDAR