BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Kernels for Sequentially Ordered Data - Dr Franz Kiraly
DTSTART:20161125T110000Z
DTEND:20161125T120000Z
UID:TALK69355@talks.cam.ac.uk
CONTACT:Martine Gregory-Jones
DESCRIPTION:Kernel learning is a general framework providing methodology f
 or descriptive/exploratory statistics\, non-linear regression and classifi
 cation learning\, for objects of any kind\, for example time series\, text
 \, matrices\, vectors up to mirror symmetries\, and so on - given the righ
 t feature representation encoded by the so-called kernel\, a non-linear sc
 alar product. After briefly reviewing the kernel learning framework and pr
 ior work on learning with structured objects or invariances\, we present m
 ethodological foundations for dealing with sequential data of any kind\, s
 uch as time series\, sequences of graphs\, or strings.  Our approach is ba
 sed on signature features which can be seen as an ordered variant of sampl
 e (cross-)moments\; it allows to obtain a "sequentialized" version of any 
 static kernel.  The sequential kernels are efficiently computable for disc
 rete sequences and are shown to approximate a continuous moment form in a 
 sampling sense. A number of known kernels for sequences arise as "sequenti
 alizations" of suitable static kernels:  string kernels may be obtained as
  a special case\, and alignment kernels are closely related up to a modifi
 cation that resolves  their open non-definiteness issue.  Our experiments 
 indicate that our signature-based sequential kernel framework may be a pro
 mising approach to learning with sequential data\, such as time series\, t
 hat allows to avoid extensive manual pre-processing.
LOCATION:Rayleigh seminar room\, 2nd floor\, Maxwell Building\, Cavendish 
 laboratory
END:VEVENT
END:VCALENDAR