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:Imagers Interest Group
SUMMARY:Fast validation of LDA classifiers for cross-valid
ation and permutation testing + new MVPA toolbox -
Matthias Treder (University of Birmingham)
DTSTART;TZID=Europe/London:20180305T123000
DTEND;TZID=Europe/London:20180305T133000
UID:TALK100270AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/100270
DESCRIPTION:In multivariate pattern analysis (MVPA) of neuroim
aging data\, it is often important to establish wh
ether a given classification performance is statis
tically significant. Unfortunately\, one of the mo
st popular approaches\, permutation testing\, is c
omputationally expensive\, since a classifier has
to be trained and tested thousands of times. For L
inear Discriminant Analysis (LDA) classifiers\, th
is problem can be circumvented by exploiting its r
elationship with linear regression. An update rule
can be used to calculate the classifier outputs i
n each cross-validation fold without ever re-train
ing the classifier. Remarkably\, since the hat mat
rix is independent of the class labels\, this appr
oach immediately extends to permutations. The appr
oach can be generalised to multi-class LDA via the
notion of optimal scoring. The update rule allows
for drastic computational improvements especially
in large feature spaces. Analyses using simulatio
ns and a publicly available MEG dataset (Wakeman a
nd Henson\, 2015) demonstrate speed improvements o
f several orders of magnitude over the classical r
e-training approach while providing identical resu
lts.\n\nThe function will be implemented in MVPA-L
ight (github.com/treder/MVPA-Light)\, an easy-to-u
se toolbox on multivariate pattern classification
in MATLAB. The toolbox provides a fast interface f
or cross-validation\, searchlight analysis\, time
classification and time x time generalisation. So
far\, Linear Discriminant Analysis (LDA) with shri
nkage regularisation\, logistic regression with L2
regularisation\, linear and nonlinear Support Vec
tor Machines\, multi-class LDA\, and ensemble meth
ods have been implemented. To provide a fast\, out
-of-the-box solution\, MVPA-Light uses custom impl
ementations of popular optimisation algorithms suc
h as Dual Coordinate Descent.
LOCATION:Lecture Theatre\, MRC Cognition and Brain Sciences
Unit\, 15 Chaucer Road\, Cambridge
CONTACT:Johan Carlin
END:VEVENT
END:VCALENDAR