[Eeglablist] Question about the use of ICA and GC

[Makoto Miyakoshi]

I can point you to the SIFT manual written by Tim.

Without checking the paper you sent me, as long as the tool is advertised
as multivariate, it should address the spurious connectivity problem. My
point was that if you repeat GC without addressing this issue you will get
wrong results.

Makoto


On Mon, Jul 28, 2014 at 5:00 PM, Iman M.Rezazadeh <irezazadeh at ucdavis.edu>
wrote:

> I think it depends on how we calculate GC and also eliminate the volume
> conductance problem. Have you looked at the following paper? Also, it would
> be appreciated if you please cite some papers for your logic/statementJ
>
> The MVGC multivariate Granger causality toolbox: A new approach to
> Granger-causal inference Lionel Barnett∗, Anil K. Seth
>
>
>
> *From:* Makoto Miyakoshi [mailto:mmiyakoshi at ucsd.edu]
> *Sent:* Monday, July 28, 2014 4:43 PM
> *To:* Iman M.Rezazadeh
> *Cc:* Salim Al-wasity; EEGLAB List
>
> *Subject:* Re: [Eeglablist] Question about the use of ICA and GC
>
>
>
> Dear Iman,
>
>
>
> Because it can detect spurious connections that should be suppressed.
>
>
>
> Makoto
>
>
>
> On Mon, Jul 28, 2014 at 3:49 PM, Iman M.Rezazadeh <irezazadeh at ucdavis.edu>
> wrote:
>
> Hi Makoto,
>
> Could you please let me know how do you get to this point “By the way if
> you have more than 3 ICs for GC don't use GC but use RPDC or normalized
> dDTF” ?
>
> Thanks !
>
> Iman
>
>
>
> *From:* eeglablist-bounces at sccn.ucsd.edu [mailto:
> eeglablist-bounces at sccn.ucsd.edu] *On Behalf Of *Makoto Miyakoshi
> *Sent:* Monday, July 28, 2014 3:30 PM
> *To:* Salim Al-wasity
> *Cc:* eeglablist at sccn.ucsd.edu
> *Subject:* Re: [Eeglablist] Question about the use of ICA and GC
>
>
>
> Dear Salim,
>
>
>
> > 1- Are the predicted S1(60) and S2(60) independent as the actual ones
> which obtained using ICA.
>
>
>
> If the S1(60) and S2(60) are the predictions made by i.e. AR model from
> S1(1-59) and S2(1-59) respectively, then they are most likely independent.
>
>
>
> > 2- The past values that it used to predict S1(60) and S2(60) are
> independent. Does this independence affect the predict variables?
>
>
>
> Yes, why not?
>
>
>
> Makoto
>
>
>
> On Sun, Jul 27, 2014 at 8:22 AM, Salim Al-wasity <salim_alwasity at yahoo.com>
> wrote:
>
> Dear Mr. Miyakoshi
>
> Thanks for your reply. Let assumes that an ICA is applied to an observe
> matrix X (of 2 channels and 100 samples), therefore an S decomposed
> independent source signal is obtain.
>
> After that GC is used to find the connectivity between S1 and S2 (channel
> 1 and channel 2 of S), for instance to predict S1(60) and S2(60) based on
> the past values of S1 and S2 (for a predefined order), then:
>
> 1- Are the predicted S1(60) and S2(60) independent as the actual ones
> which obtained using ICA.
>
> 2- The past values that it used to predict S1(60) and S2(60) are
> independent. Does this independence affect the predict variables?
>
>
>
> Sincerely
>
> Salim
>
>
>
> On Friday, 25 July 2014, 20:41, Makoto Miyakoshi <mmiyakoshi at ucsd.edu>
> wrote:
>
>
>
> Dear Salim,
>
>
>
> ICA assures *instantaneous* independence, while GC calculates *temporal*
> causality. This means that ICA does not know what happens in the next
> moment, but GC does.
>
>
>
> By the way if you have more than 3 ICs for GC don't use GC but use RPDC or
> normalized dDTF.
>
>
>
> Makoto
>
>
>
> On Fri, Jul 25, 2014 at 4:41 AM, Salim Al-wasity <salim_alwasity at yahoo.com>
> wrote:
>
> Dears
>
> Have a nice day. Kindly I have a question regarding ICA:
> I have EEG data of (44 channels X 294000 samples)
>
> 1- I applied the ICA to separate the noise and find the ICs which are
> belong to brain activities.
> 2- I used the Granger Causality (GC) in SIFT to find the connectivity
> between these ICs and discover which component influence which. However I
> am not sure about the results that I have got.
>
>
> My question is that If I used ICA to decompose the EEG signal into their
> sources (hint: the decomposed ICs is much less than the actual brain
> sources), theoretically these ICs would be independent, and the use of GC
> would be useless since the latter algorithm search for the dependence
> across ICs?
> Or the ICA will minimize the mutual information, and the separated
> components will not be ~100% independent, therefore each component has more
> that one source, and GC  can find some influence across these components
> for the remaining not separated sources.
>
>
>
> Your cooperation is highly appreciated
>
>
>
> Yours
>
> Salim Al-Wasity
>
> PhD student
>
> Rehabilitation Centre
>
> Biomedical Engineering Department-School of Engineering
>
> University of Glasgow
>
> Glasgow-United Kingdom
>
> +44 742 371 4444
>
>
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> 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
>
>
>
>
>
> --
>
> Makoto Miyakoshi
> Swartz Center for Computational Neuroscience
> Institute for Neural Computation, University of California San Diego
>
>
>
>
>
>
>
> --
>
> Makoto Miyakoshi
> Swartz Center for Computational Neuroscience
> Institute for Neural Computation, University of California San Diego
>
>
>
>
>
> --
>
> Makoto Miyakoshi
> Swartz Center for Computational Neuroscience
> Institute for Neural Computation, University of California San Diego
>



-- 
Makoto Miyakoshi
Swartz Center for Computational Neuroscience
Institute for Neural Computation, University of California San Diego
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20140728/2c8f73fd/attachment.html>