[Eeglablist] Separation of an ICA component and filtering ICA components

[Tamar Cohen]

Let's assume that you have a great ICA component, but it seems to include
two sources which you wish to separate (and keep one of them as the other
component is contaminated).

Of course- you may in some cases run repeated ICA or run it with a larger
number of components, but what one would expect to get if one would filter
the ICA components and then transform them back to the original space?

The inverse of the weights (“patterns”) would not be valid after filtering
the ICA components.
What do you think about this idea?

Any attempt to apply a similar method in the past?

Obviously- it would make sense to filter ICA components if one sticks with
the ICA space, but filtering and mapping back sounds slightly undefined to
me.

What do you think?