BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Bayesian approaches to autonomous Bayesian real-time learning - Jo
 -Anne Ting (University of Southern California)
DTSTART:20080915T130000Z
DTEND:20080915T140000Z
UID:TALK13141@talks.cam.ac.uk
CONTACT:Zoubin Ghahramani
DESCRIPTION:I propose a set of Bayesian methods to help us work towards th
 e goal of autonomous real-time learning. Specifically\, I am interested in
  scenarios where the input data has thousands of dimensions and where real
 -time\, incremental learning may be needed\, as in robotics\, real-time vi
 sion\, brain-computer interfaces\, autonomous vehicles etc. Real-time auto
 nomous learning in such data-rich environments is challenging\, due to iss
 ues such as outliers\, noisy sensory data\, redundant and irrelevant dimen
 sions\, and the need for computational efficiency in real-time conditions.
  I introduce a set of automatic methods to address these challenges\, usin
 g Bayesian inference -- combined with variational approximations -- in ord
 er to eliminate open parameters in a principled way. All these methods can
  be leveraged together to develop a Bayesian local kernel shaping for nonl
 inear regression. Bayesian local kernel shaping is computationally efficie
 nt\, requires no sampling and automatically rejects outliers. It can be us
 ed for nonparametric regression with local polynomials (e.g.\, for real-ti
 me learning) or as a novel method to achieve nonstationary regression with
  Gaussian processes. The usefulness and improved performance of our algori
 thms are illustrated in various robotic applications such as parameter ide
 ntification in robot dynamics\, real-time outlier detection in tracking an
 d learning a task-level control law.
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR