BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Microsoft Research Cambridge\, public talks
SUMMARY:"\;Egalitarian Gradient Descent"\;: A Gene
ral Preconditioning Scheme for Deformable Image Re
gistration - Darko Zikic\, TU Munich
DTSTART;TZID=Europe/London:20110809T100000
DTEND;TZID=Europe/London:20110809T110000
UID:TALK32214AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/32214
DESCRIPTION:In this talk\, I will discuss a simple\, yet theor
etically justified approach for improving the effi
ciency of optimization in deformable registration
problems\, for arbitrary difference measures.\n\nG
iven the registration problem as optimization of a
n energy containing a difference measure\, the key
step of our approach is extremely simple: We modi
fy the vector field corresponding to the gradient
of the difference measure\, by normalizing the sin
gle point-wise vectors of the field to approximate
ly unit length. We show that this simple scheme im
proves the convergence speed of optimization\, whi
le being 1) theoretically justified\, 2) computati
onally efficient\, and 3) applicable to arbitrary
difference measures. Since the above scheme incorp
orates our idea to give roughly the same influence
to all relevant points in the image domain\, we d
ub the resulting approach â€œegalitarian".\n\n1) The
above scheme is theoretically justified\, as we d
emonstrate that it corresponds to a preconditionin
g of the difference measure. Actually\, the scheme
approximates the optimal preconditioning with res
pect to the analyzed model.\n\n2) Due to the simpl
icity of the gradient modification and the resulti
ng computational efficiency\, the improvement in c
onvergence speed is directly translated to shorter
effective runtimes.\n\n3) Finally\, a very import
ant aspect of our approach is that it is applicabl
e to arbitrary difference measures. This is of hig
h practical value for problems in which the range
of applicable optimization methods is limited\, su
ch as deformable registration tasks in multi-modal
settings\, which utilize statistical difference m
easures such as Mutual \nInformation.\n\nBesides d
iscussing the actual method\, the talk will focus
on providing an intuitive motivation for the appro
ach\, highlighting its relations to other optimiza
tion schemes\, and demonstrating its application i
n popular registration frameworks. Also\, the talk
will contain a brief overview of some of my other
image registration projects\, including deformabl
e 2D-3D registration\, and linear registration by
MRFs and discrete optimization.\n
LOCATION:Small lecture theatre\, Microsoft Research Ltd\, 7
J J Thomson Avenue (Off Madingley Road)\, Cambrid
ge
CONTACT:Microsoft Research Cambridge Talks Admins
END:VEVENT
END:VCALENDAR