bci2000.org BBS

I am new to BCI and I'd like to know if I would like to implement a classifier filter for VEP other than the AR, what are the recommended neural network algorithms?

Hi,

I strongly suggest to begin using the simplest possible method, and one that is proven to work. Only when you get this to work would I implement a different classifier. My personal opinion is that neural networks are not very well suited to the BCI problem; even if you do everything right, it will be very difficult to improve performance over what is currently implemented.

Gerv

Thanks for your reply
I have another question related to VEP.I read that windows OS is not suitable for this kind of classifier because it is not real time. Is there a way to bypass this problem

Thanks,
BCI2000user

Hi,

Stimuli can be presented correctly, even under Windows. This can be done using DirectX programming. For example, the Presentation program can be configured to provide stimuli that would be appropriate to elicit VEPs, and BCI2000 could analyze the corresponding brain signals in real time. Alternatively, you may want to write your own BCI2000 module that does something similar to Presentation.

Gerv

Hi,
you have suggested using the simplest way to classify VEP. By this do you mean something Like k-nearest neighbor or somthing simpler?

Also we are trying to differentiate between green and red colors, what are the best positions for the electrodes? (Oz, Po7, Po8)!!

you have suggested using the simplest way to classify VEP. By this do you mean something Like k-nearest neighbor or somthing simpler?

You could use the linear classifier that is built into BCI2000. When we tried BCI2000 with SSVEPs some time ago, this worked very well.

Also we are trying to differentiate between green and red colors, what are the best positions for the electrodes? (Oz, Po7, Po8)!!

Those, and O1/2, seem to be good electrodes. However, similar to my comment with the P300, the key here is to first collect sufficient amounts of data, second, to analyze these data so that in fact you confirm that you are looking at a good signal, and only then proceed to BCI experiments.

Gerv