Covariate shift adaptation strategies for Muse?


Does anyone with a little BCI background have any strategies for dealing with the covariate shift adaptation problem on Muse, specifically dealing with problems like:

  1. Picking a target power and variability (for example, in the alpha band), when all the usual EEG confounds can vary so much between sessions, in order to calibrate what a target range should be.
  2. How to detect and adjust for nonstationarities. For example, looking for changes in band power and variability, determining that a statistically significant change has occurred, and recalibrating/renormalizing.

As far as I can tell, most work done in this area is either ad hoc or proprietary, so I was curious if there was any more formal literature that might be available, or if anyone on the forum had any experience with this? Most of the examples I found were a lot more sophisticated than what I really wanted to get into, specifically on how to do it with a PCA/CSP model…