BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Heteroscedastic Concomitant Lasso for sparse multimodal electromag
netic brain imaging
DTSTART;VALUE=DATE-TIME:20180111T150000Z
DTEND;VALUE=DATE-TIME:20180111T160000Z
DTSTAMP;VALUE=DATE-TIME:20200530T060220Z
UID:indico-event-2899@indico.math.cnrs.fr
DESCRIPTION:In high dimension\, it is customary to consider Lasso-type est
imators to enforce sparsity.\nFor standard Lasso theory to hold though\, t
he regularization parameter should be proportional to the noise level\, wh
ich is generally unknown in practice.\nA remedy is to consider estimators\
, such as the Concomitant Lasso\, which jointly optimize over the regressi
on coefficients and the noise level.\nHowever\, when data from different s
ources are pooled to increase sample size\, or when dealing with multimoda
l data\, noise levels differ and new dedicated estimators are needed.\nWe
provide new statistical and computational solutions to perform heterosceda
stic regression\, with an emphasis on functional brain imaging with magnet
o- and electroencephalographic (M/EEG) signals.\nWhen instantiated to de-c
orrelated noise\, our framework leads to an efficient algorithm.Experiment
s demonstrate improved prediction and support identification with correct
estimation of noise levels. Results on multimodal neuroimaging problems w
ith M/EEG data are also reported.\n\nThis is joint work with M. Massias an
d A. Gramfort.\n\nhttps://indico.math.cnrs.fr/event/2899/
LOCATION:UMPA salle 435
URL:https://indico.math.cnrs.fr/event/2899/
END:VEVENT
END:VCALENDAR