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:Signal Processing and Communications Lab Seminars
SUMMARY:Classification on the Grassmann Manifold: Performa
nce Limits of Compressive Classifiers - Dr. Miguel
Rodrigues\, University College London
DTSTART;TZID=Europe/London:20140227T140000
DTEND;TZID=Europe/London:20140227T150000
UID:TALK50511AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/50511
DESCRIPTION:The reliable classification of high-dimensional si
gnals from low-dimensional measurements is an incr
easingly crucial task in the age of the data delug
e. This talk demonstrates that tools and intuition
s from Shannon theory enable the derivation of fun
damental limits on the performance of such classif
ication systems\, by focusing on the classificatio
n of high-dimensional (rank-deficient) Gaussian si
gnals from noisy\, low-dimensional signal projecti
ons.\n\nLeveraging the syntactic equivalence of di
scrimination between Gaussian classes and communic
ation over vector wireless channels\, bounds on cl
assifier performance will be presented that are as
ymptotic in two regimes. First\, the notion of cla
ssification capacity is introduced\, which charact
erizes the number of classes that can be discrimin
ated reliably as the signal dimensionality approac
hes infinity\; tight bounds on the classification
capacity associated with Gaussian classes are pres
ented. Second\, the notion of diversity-discrimina
tion tradeoff is also introduced\, which\, by anal
ogy with the diversity-multiplexing tradeoff of ve
ctor channels\, characterizes the tradeoff between
the misclassification probability and the number
of discernible classes as the signal-to-noise rati
o goes to infinity\; again tight bounds on this tr
adeoff are also proven.\n\nThese results reveal th
at the “easiest” classes to discriminate correspon
d to (affine) subspaces drawn from an appropriate
Grassmann manifold\; they further reveal a precise
relationship between signal and measurement geome
try and classifier performance. Numerical results\
, including a face recognition application\, valid
ate this relationship in practice.\n\nThis represe
nts joint work with Matthew Nokleby (Duke Universi
ty\, USA) and Robert Calderbank (Duke University\,
USA)\n\n*BIO*: Miguel Rodrigues is a Senior Lectu
rer with the Department of Electronic and Electric
al Engineering\, University College London\, U.K.
He was previously with the Department of Computer
Science\, University of Porto\, Portugal\, rising
through the ranks from Assistant to Associate Prof
essor\, where he also led the Information Theory a
nd Communications Research Group at Instituto de T
elecomunicações – Porto. He received the Licenciat
ura degree in Electrical Engineering from the Facu
lty of Engineering of the University of Porto\, Po
rtugal in 1998 and the Ph.D. degree in Electronic
and Electrical Engineering from University College
London\, UK in 2002. He has carried out postdocto
ral research work both at Cambridge University\, U
K\, as well as Princeton University\, USA\, in the
period 2003 to 2007. He has also held visiting ap
pointments at Princeton University\, Duke Universi
ty\, Cambridge University\, and University College
London in the period 2007 to 2013.\n\nHis researc
h interests are in the general areas of informatio
n theory\, communications theory and statistical s
ignal processing. He was the recipient of the IEEE
Communications and Information Theory Societies J
oint Paper Award in 2011 for the work on Wireless
Information-Theoretic Security (with M. Bloch\, J.
Barros and S. W. McLaughlin).
LOCATION:BE4-38 (CBL Meeting Room)
CONTACT:Dr Ramji Venkataramanan
END:VEVENT
END:VCALENDAR