BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.bham.ac.uk//v3//EN
BEGIN:VEVENT
CATEGORIES:Artificial Intelligence and Natural Computation se
minars
SUMMARY:Online Kernel Density Estimation for learning and
classification - Dr. Matej Kristan\, Visual Cognit
ive Systems Laboratory\, Faculty of Computer and I
nformation Science\, University of Ljubljana
DTSTART:20130517T150000Z
DTEND:20130517T160000Z
UID:TALK1065AT
URL:/talk/index/1065
DESCRIPTION:\nPlease note that coffee will be served at 3:15 a
t the coffee room (School of Computer Science)\, f
ollowed by the one-hour seminar at 4pm.\n\nMany pr
actical applications require building probabilisti
c models of a perceived environment\, or of an obs
erved process\, in form of probability density fun
ctions (pdf) over some moderate-dimensional featur
e space. Building the pdf models from large batche
s of data may be computationally infeasible. Moreo
ver\, all the data may not be available in advance
. We might want to observe some process for an ind
efinite duration\, while continually providing the
best estimate of the pdf from the data observed s
o far. To guarantee the model's low computational
complexity at the its application\, the model shou
ld remain simple enough even after observing many
new data-points.\n\nIn this talk I will present an
approach to online estimation of probability dens
ity functions\, which is based on kernel density e
stimation (KDE). The approach is called the online
Kernel Density Estimation (oKDE) and generates a
Gaussian mixture model from a stream of data-point
s. It retains the nonparametric quality of a KDE\,
allows adaptation from as little as a single data
-point at a time\, while keeping the model's compl
exity low. The oKDE allows estimation of stationar
y as well as nonstationary distributions and is ex
tendable to online construction of classifiers. I
will show the main results of oKDE comparison to c
ompeting batch KDEs and support vector machines on
problems of online estimation of generative model
s as well as classifiers.
LOCATION:UG09\, Learning Centre
CONTACT:
END:VEVENT
END:VCALENDAR