[Eeglablist] ICA strategy on multi epoch data in EEGLab

[arno]

Dear Peng,

peng wrote:
> I have a question on the ICA strategy of multi epoch data in EEGLab:
>    When we import NeuroScan eeg file into EEGLab, the data would be 3
> dimension: time points * channels * epochs. I think ICA would deal
> with 2-dimention data, so there may be several strategies.
>   

The 3-D matrix is transformed into 2-D before running Infomax  ICA 
(runica function). This is equivalent to concatenating all data epochs. 
Note that since ICA is an instantaneous algorithm, and does not take 
into account time information, epochs boundaries are irrelevant. Epoch 
boundaries are not irrelevant to the SOBI ICA algorithm for instance 
that does take into account time information. For SOBI the 3-D array is 
given as input and ICA and the covariance matrix is computed for each 
epoch, then averaged across epochs.

Best regards,

Arno