[Eeglablist] Effect of pass band on ICA learning rate

[Makoto Miyakoshi]

Dear David,

I think I wrote a reply to this post. Please refer to it.

> 2: explain the nature of said relationship?

The higher the high-pass cutoff frequency (up to 2 Hz), the better the
decomposition.
This is probably because 1) low-freq data has the highest power in 1/f PSD
data, which attracts most attention of ICA, yet 2) such low-freq data does
not have much response of interest in the conventional analysis (i.e., does
not impact ICA results even if cut out), and 3) such low-freq data do
contain known source of artifacts (sweating etc)

Makoto

On Mon, Sep 2, 2019 at 1:24 AM Jenson, David Evans <david.jenson at wsu.edu>
wrote:

> EEGLABers
>
> I filtered a dataset today from 3-30 Hz and ran ICA, only for the learning
> rate to lower several times (restarting the step count each time) before
> finally continuing with decomposition.  However, when processing the same
> dataset with a wider pass-band (1-40 Hz), ICA proceeds as normal without
> lowering the learning rate and starting over.
>
> Is anyone able to:
>
> 1: confirm a relationship between with of the pass-band of a filter and
> this learning rate issue with ICA?
>
> 2: explain the nature of said relationship?
>
> Thanks
>
>
> David Jenson
> Assistant Professor
> Department of Speech and Hearing Sciences
> o: 509-368-6913 | <mailto:david.jenson at wsu.edu> david.jenson at wsu.edu
> <mailto:david.jenson at wsu.edu><mailto:linda.gallup at wsu.edu>
>
> _______________________________________________
> 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
Assistant Project Scientist, Swartz Center for Computational Neuroscience
Institute for Neural Computation, University of California San Diego