[Eeglablist] ICA - message
Blefari Maria Laura
m.blefari at sssup.it
Thu Jan 28 23:59:04 PST 2010
Thanks Manousos for your clear and quick answer. I will
On Fri, 29 Jan 2010 09:51:02 +0200
Klados Manousos <mklados at gmail.com> wrote:
>Dear Maria Laura
>The problem relies on the real dimensionality of the
>data. The real
>independent sources are much more than the recorded
>electrodes, so fast-ICA
>tries to exctract many sources in a single independent
>component. If you
>consider the central limit theorem this component tends
>to have gaussian
>distribution, which is a great problem...especially when
>you use algorithms
>based on high order statistics
>According to my opinion is better to use extended -
>infomax ICA (the default
>ICA of EEGLAB) or if you want something faster you can
>is based on second order statistics and can retrieve
>distributions close to gaussian.
>I hope i helped...
>2010/1/28 Blefari Maria Laura <m.blefari at sssup.it>
>> I'm using fastICA algorithm in order to remove
>> artifact. I have 16 electrodes.
>> Sometimes I got this messagge
>> Component 13 did not converge in 1000 iterations.
>> Too many failures to converge(6). Giving up. Adding the
>> mean to the data.
>> eeg_checkset:recompiuting the ICA activation matrix.
>> and then stopped.
>> What happen, any idea? the signal is too noisy and it is
>> not possible to perform the decomposition?
>> Thanks in advance,
>> Maria Laura
>> Eeglablist page:
>> To unsubscribe, send an empty email to
>> eeglablist-unsubscribe at sccn.ucsd.edu
>> For digest mode, send an email with the subject "set
>>digest mime" to
>> eeglablist-request at sccn.ucsd.edu
>Manousos A. Klados
>PhD Candidate -- Research Assistant
>Group of Applied Neurosciences
>Lab of Medical Informatics
>School of Medicine
>Aristotle University of Thessaloniki
>P.O. Box 323 54124 Thessaloniki Greece
More information about the eeglablist