BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Scalable Gaussian Processes for Scientific Discovery - Dr Andrew W
 ilson\, Carnegie Mellon University
DTSTART:20150729T100000Z
DTEND:20150729T110000Z
UID:TALK60149@talks.cam.ac.uk
CONTACT:12852
DESCRIPTION:Large datasets provide unprecedented opportunities to automati
 cally discover rich statistical structure\, from which we can derive new s
 cientific discoveries.  Gaussian processes are flexible distributions over
  functions\, which can learn interpretable structure through covariance ke
 rnels.  In this talk\, I introduce an O(N) Gaussian process framework whic
 h is capable of learning expressive kernel functions on large datasets.  T
 his framework generalizes and provides alternative derivations for classic
 al inducing point methods\, and allows one to exploit kernel structure for
  significant further gains in scalability and accuracy\, without requiring
  severe assumptions.  I evaluate this approach for kernel matrix reconstru
 ction\, kernel learning\, time series modeling\, image inpainting\, and lo
 ng range forecasting in spatiotemporal statistics.\n\n\nReferences:\n\nhtt
 p://www.cs.cmu.edu/~andrewgw/pattern \n\nhttp://jmlr.org/proceedings/paper
 s/v37/wilson15.pdf\n\nhttp://jmlr.org/proceedings/papers/v37/wilson15-supp
 .pdf
LOCATION:Engineering Department\, CBL Room BE-438
END:VEVENT
END:VCALENDAR