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:Machine Learning @ CUED
SUMMARY:Population Inference for Functional Brain Connecti
vity - Genevera I. Allen (Rice University)
DTSTART;TZID=Europe/London:20150304T110000
DTEND;TZID=Europe/London:20150304T120000
UID:TALK58206AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/58206
DESCRIPTION:Functional brain connections are the set of statis
tical relationships between neural activity in dif
ferent parts of the brain\; these are typically es
timated from neuroimaging data such as functional
MRI. In multi-subject studies\, many are intereste
d in identifying the individual functional brain c
onnections or patterns of connections that are dif
ferent between two groups of subjects or across a
clinical population. Popular approaches to this p
roblem include estimating a network for each subje
ct\, and then assuming the subject networks are fi
xed\, conducting inference over network metrics.
These approaches\, however\, fail to account for t
he variability and network estimation error associ
ated with estimating each subjectâ€™s brain network\
, thus resulting in incorrect inferences. \n\nIn
this talk\, we study this problem using Markov Net
works as the model for brain connectivity. Statis
tically\, our problem can be described as conducti
ng large scale inference over network edges or gro
ups of edges post graphical model selection\, part
of a novel statistical paradigm we call Populatio
n Post Selection Inference (popPSI). We show that
for this popPSI problem\, current approaches in n
euroimaging have both low statistical power and hi
ghly inflated false positive rates. We then devel
op a new procedure which we term R^3\, standing fo
r resampling\, random effects\, and random penaliz
ation. Our approach uses the correct two-level ra
ndom effects model to account for network variabil
ity within a subject (due to network estimation) a
s well as between subjects. Through simulation st
udies we show that our method solves many of the p
roblems associated with existing techniques\, yiel
ding substantial improvements in terms of both err
or control and statistical power. We conclude our
talk by applying our methods in two case studies -
a color-sequence synesthesia study and a neurofib
romatosis one study.\n\nJoint work with Manjari Na
rayan and Steffie Tomson.
LOCATION:Engineering Department\, CBL Room BE-438
CONTACT:Dr R.E. Turner
END:VEVENT
END:VCALENDAR