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:Cambridge Image Analysis Seminars
SUMMARY:Randomized second-order algorithm for MAP estimati
on and fast MSE estimator applied to Computed Tomo
graphy - Alessandro Perelli (DTU)
DTSTART;TZID=Europe/London:20191001T130000
DTEND;TZID=Europe/London:20191001T140000
UID:TALK131263AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/131263
DESCRIPTION:Two families of algorithms for MAP and MSE estimat
ion respectively will be presented and applied to
a monoenergetic X-ray Computed Tomography (CT) acq
uisition model.\nSecond order methods for solving
regularized optimization problems with generalized
linear models have been widely studied but despit
e the superior convergence rate compared to first
order methods one weakness relies on the computati
onal cumbersome for calculating the Hessian matrix
. Additionally\, in imaging applications where the
input prior is difficult to model\, powerful regu
larization techniques are based on data-driven mod
els or denoisers.\nFor MAP estimation\, an effici
ent and accurate randomized second order method fo
r model based CT reconstruction is proposed. The a
lgorithm combines the idea of dimensionality reduc
tion of the Hessian of the likelihood cost functio
n by sketching\, using ridge leverage scores\, and
an explicit regularizer term which can be impleme
nted by a generic denoiser through the score match
ing formulation. We show how to compute the gradie
nt and the Hessian of the likelihood and regulariz
er together with simulated results.\nFinally\, a f
ist order iterative method\, called approximate me
ssage passing\, will be presented for performing M
SE estimation efficiently.
LOCATION:MR 14\, Centre for Mathematical Sciences
CONTACT:Carola-Bibiane Schoenlieb
END:VEVENT
END:VCALENDAR