[QUESTION] Representing windows as covariances?

[sebasarango1180]

Awesome work, thanks for making it public!

I'm just getting started with this all so I wanted to better understand - is there a technical motivation or preference for having time window samples represented as covariances for training? Is that the standard? Are there any other alternatives worth checking (including other methods for statistical summarization, time series...)?
If there are any educational resources related to processing EEG data for ML training that you can share, that would be much appreciated!

Thanks!