MI-CSP Alternatives

Working with OpenViBE signal processing scenarios and doing scenario/BCI design
Post Reply
mhadji05
Posts: 53
Joined: Tue Mar 16, 2021 9:37 am

MI-CSP Alternatives

Post by mhadji05 »

Hello all.
Happy new year.

I m trying to improve my MI (Left-right hand imagination) system with better feature extraction algorithms.

I saw the Regularized CSP Trainer Box and I m wondering what are the best values in the settings to work in MI-CSP scenario.
a. Filters per condition.
b. Save filters as box config
c. Covariance update
d. Trace normalization
e. shrinkage coefficient
f. Tikhonov coefficient

Please explain me what are this values and what i have to set.


There is also the CSP Spatial Filter Trainer Metabox. Is it just a fusion of temporal filter and CSP or maybe can achieve better features?

Finally is there a way to use the following CSP feature extraction variations?
a. Adaptive Composite Common Spatial Pattern (ACCSP)
b. Self Adaptive Common Spatial Pattern (SACSP) algorithms

Thanks in advance!

Thomas
Posts: 210
Joined: Wed Mar 04, 2020 3:38 pm

Re: MI-CSP Alternatives

Post by Thomas »

Hi mhadji05,

This is a good question and there is no absolute best values for these settings. It can depend on the data and require some experimentation to find out what works best.

I invite you to have a look at the documentation of the box that provides details on how the CSP algorithm works and publications providing better understanding of some parameters.

The variations of the CSP that you mention are unfortunately not implemented at the moment.
However, OpenViBE is an open source software and any contributions of new algorithms are more than welcome.

Cheers,
Thomas

Post Reply