BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Optimizing  Similarity  Functions  in  Various  Pattern  Recogniti
 on  Problems - Sang Wan Lee\,  KAIST Korea
DTSTART:20081209T100000Z
DTEND:20081209T110000Z
UID:TALK15648@talks.cam.ac.uk
CONTACT:David MacKay
DESCRIPTION:Machines classify instances by using various similarity functi
 ons\,  such  as\nimage-based feature  extractors\,  clustering  functions\
 ,  kernel  functions\,\netc. In real world applications\, however\, they o
 ften work poor.  Since  they\nare characterized by a set of parameters\, t
 he choice of them hugely  affects\nperformance. To solve this parameter op
 timization problem\, in this  talk  is\npresented optimization strategies 
 in terms of class separability.  First\,  a\nGabor Wavelet Neural Network 
 for facial expression recognition is  presented\nand  its  optimization  m
 ethod  is  proposed.   Second\,   an   agglomerative\nclustering technique
  is  presented  and  a  strategy  for  choosing  optimal\nnumber of  clust
 ers  is  proposed.  A  concatenation  of  the  agglomerative\nclustering a
 nd Fuzzy-state Q-learning allows us to learn and  predict  human\nbehavior
   patterns.  The  last  part  of  the  talk  is  about  learning   a\ngene
 ralized  kernel  function  for  kernel-based   pattern   classification.\n
 Theoretical results on a generalized kernel function and  its  learning  v
 ia\nregularization are  presented.  Its  performance  is  demonstrated  wi
 th  an\napplication of EMG signal-based walking phase recognition.
LOCATION:TCM Seminar Room\, Cavendish Laboratory\, Department of Physics
END:VEVENT
END:VCALENDAR