<html>
  <head>
    <meta content="text/html; charset=ISO-8859-1"
      http-equiv="Content-Type">
  </head>
  <body bgcolor="#FFFFFF" text="#000000">
    Hello all,<br>
    <br>
    We have perfomed some experiments on the necessary data lenght topic
    for ICA. They are described in a paper we have presented at the
    Biosignals 2012 conference a month ago (G. Korats et al, "Impact of
    window length and decorrelation step on ICA algorithms for EEG blind
    source separation"). <br>
    <br>
    Best,<br>
    Radu<br>
    <br>
    On 18/03/2012 14:55, Scott Makeig wrote:
    <blockquote
cite="mid:CAAHbppXjMSiQY5c7icAHrUtYpA+JwzoVXoYHcCviWuAaWivCSA@mail.gmail.com"
      type="cite">The data_length/channels^2 = k > 30 heuristic was
      based on observing our 256-channel data decompositions. For fewer
      channels, I don't think this (k > 30) may be necessary (though
      a small k will very likely prove a problem!). 
      <div>
        <br>
      </div>
      <div>But for small numbers of channels, data length is not so much
        an obstacle -- for example, decomposing 32 channels with k=30
        would require only 30*32^2= 32K data points: At a sampling rate
        of (say) 256 Hz, this would be ~2 minutes of data. With 128
        channels, the equivalent heuristic would be ~32 minutes.
        <div>
          <br>
        </div>
        <div>I hope one of us will be able to do a proper study,
          decomposing data subsets of various lengths and measuring
          mutual information reduction and dipolarity (see Delorme et
          al., PLoS One 2012). Also, Jason Palmer has been working on
          theoretic lower bounds on ICA accuracy at given data lengths
          and channel numbers. I'll summarize result with him when
          possible.</div>
        <div><br>
        </div>
        <div>Scott Makeig<br>
          <br>
          <div class="gmail_quote">On Sat, Mar 17, 2012 at 5:17 PM,
            Tarik S Bel-Bahar <span dir="ltr"><<a
                moz-do-not-send="true"
                href="mailto:tarikbelbahar@gmail.com">tarikbelbahar@gmail.com</a>></span>
            wrote:<br>
            <blockquote class="gmail_quote" style="margin:0 0 0
              .8ex;border-left:1px #ccc solid;padding-left:1ex">
              <div>1. not sure if you have to decimate, have seen some
                people to do this to better meet requirements for good
                ICA decompositions.</div>
              <div>2. don't think higher sampling rate will give you
                better ICA. I think it's more of a matter that ICA is
                fed data that gives an accurate and lengthy
                representation of the whole "Data space".</div>
              <div>3. quote from earlier eeglablist post: "<span>the key
                  factor is how much data you have  (</span><span>timepoints</span><span> / </span><span>channels</span><span>^2).
                  If this is > 30 (or near to it), then we find it
                  preferable to return all possible components (since
                  pca does a rather poor job of separating sources)."</span></div>
              <div><span>So if below this threshold there is some reason
                  for adding more time to the protocol, or reducing
                  channels, or  decimation</span></div>
              <br>
              <br>
              <br>
              <div class="gmail_quote">
                <div>
                  <div class="h5">On Tue, Mar 13, 2012 at 6:00 AM,
                    Modestino, Edward J *HS <span dir="ltr"><<a
                        moz-do-not-send="true"
                        href="mailto:EJM9F@hscmail.mcc.virginia.edu"
                        target="_blank">EJM9F@hscmail.mcc.virginia.edu</a>></span>
                    wrote:<br>
                  </div>
                </div>
                <blockquote class="gmail_quote" style="margin:0 0 0
                  .8ex;border-left:1px #ccc solid;padding-left:1ex">
                  <div>
                    <div class="h5">
                      <div link="blue" vlink="purple" lang="EN-US">
                        <div>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d">Dear
                              EEGLAB experts,</span></p>
                          <p class="MsoNormal">
                            <span style="font-family:"Times New
                              Roman","serif";color:#1f497d"> </span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d">(1)
                              Is it true that  ICA must be subject, like
                              all the model-based spectral analysis
                              methods, to a recommendation that one
                              decimate to the lowest frequency capable
                              of representing the actual signal content
                              of the data without alienation effects?  <b>Does
                                one <u>NEED</u> to decimate the data
                                before running ICA?</b>  For instance,
                              we have a data set recorded at 1,000 Hz. 
                              Do we need to decimate this to
                              approximately 128 or 256?</span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d"> </span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d">(2
                              ) According to the formula that Dr. Makeig
                              gave to determine the optimal amount of
                              data,  <b>#timepoints/(#channels)^2</b>,
                              it would appear that a <b>higher sampling
                                rate will give better ICA results</b>. 
                              Is this the case?</span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d"> </span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d">(3)
                              Finally, using this formula, <b>#timepoints/(#channels)^2</b>,
                              is there a <b>threshold or cutoff</b> one
                              needs to be exceeded to have the optimal
                              amount of data to run ICA. Simply doing
                              the equation without any way to interpret
                              the output is not helpful.</span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d"> </span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d">Thanks
                              for your help,</span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d">Dr.
                              Modestino</span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d"> </span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d">Edward
                              Justin Modestino, Ph.D.</span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d">Postdoctoral
                              Research Associate</span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d">Ray
                              Westphal Neuroimaging Laboratory</span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d">Division
                              of Perceptual Studies</span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d">Department
                              of Psychiatry and Neurobehavioral Sciences
                            </span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d">University
                              of Virginia</span></p>
                          <p class="MsoNormal"><span
                              style="font-family:"Times New
                              Roman","serif";color:#1f497d">Email:
                              <a moz-do-not-send="true"
                                href="mailto:ejm9f@virginia.edu"
                                target="_blank">ejm9f@virginia.edu</a></span></p>
                          <p class="MsoNormal">
                            <span
style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d"> </span></p>
                          <p class="MsoNormal"><span
style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d"> </span></p>
                        </div>
                      </div>
                      <br>
                    </div>
                  </div>
                  _______________________________________________<br>
                  Eeglablist page: <a moz-do-not-send="true"
                    href="http://sccn.ucsd.edu/eeglab/eeglabmail.html"
                    target="_blank">http://sccn.ucsd.edu/eeglab/eeglabmail.html</a><br>
                  To unsubscribe, send an empty email to <a
                    moz-do-not-send="true"
                    href="mailto:eeglablist-unsubscribe@sccn.ucsd.edu"
                    target="_blank">eeglablist-unsubscribe@sccn.ucsd.edu</a><br>
                  For digest mode, send an email with the subject "set
                  digest mime" to <a moz-do-not-send="true"
                    href="mailto:eeglablist-request@sccn.ucsd.edu"
                    target="_blank">eeglablist-request@sccn.ucsd.edu</a><br>
                </blockquote>
              </div>
              <br>
              <br>
              _______________________________________________<br>
              Eeglablist page: <a moz-do-not-send="true"
                href="http://sccn.ucsd.edu/eeglab/eeglabmail.html"
                target="_blank">http://sccn.ucsd.edu/eeglab/eeglabmail.html</a><br>
              To unsubscribe, send an empty email to <a
                moz-do-not-send="true"
                href="mailto:eeglablist-unsubscribe@sccn.ucsd.edu">eeglablist-unsubscribe@sccn.ucsd.edu</a><br>
              For digest mode, send an email with the subject "set
              digest mime" to <a moz-do-not-send="true"
                href="mailto:eeglablist-request@sccn.ucsd.edu">eeglablist-request@sccn.ucsd.edu</a><br>
            </blockquote>
          </div>
          <br>
          <br clear="all">
          <div><br>
          </div>
          -- <br>
          Scott Makeig, Research Scientist and Director, Swartz Center
          for Computational Neuroscience, Institute for Neural
          Computation; Prof. of Neurosciences (Adj.), University of
          California San Diego, La Jolla CA 92093-0559, <a
            moz-do-not-send="true" href="http://sccn.ucsd.edu/%7Escott"
            target="_blank">http://sccn.ucsd.edu/~scott</a><br>
        </div>
      </div>
      <br>
      <fieldset class="mimeAttachmentHeader"></fieldset>
      <br>
      <pre wrap="">_______________________________________________
Eeglablist page: <a class="moz-txt-link-freetext" href="http://sccn.ucsd.edu/eeglab/eeglabmail.html">http://sccn.ucsd.edu/eeglab/eeglabmail.html</a>
To unsubscribe, send an empty email to <a class="moz-txt-link-abbreviated" href="mailto:eeglablist-unsubscribe@sccn.ucsd.edu">eeglablist-unsubscribe@sccn.ucsd.edu</a>
For digest mode, send an email with the subject "set digest mime" to <a class="moz-txt-link-abbreviated" href="mailto:eeglablist-request@sccn.ucsd.edu">eeglablist-request@sccn.ucsd.edu</a></pre>
    </blockquote>
  </body>
</html>