Doubts About P300 Magic and Speller Training and Online

Working with OpenViBE signal processing scenarios and doing scenario/BCI design
Post Reply
apdullahyayik
Posts: 9
Joined: Sat Jul 11, 2015 10:02 am

Doubts About P300 Magic and Speller Training and Online

Post by apdullahyayik »

SERIOUS PROBLEM FOR ME...

I have serious problem with OpenVIBE training and online session.
As I have learned P300 train.xml file detects P300 and non-P300 classes. These classes are chosen in Simulation Based Epoches boxes that are connected to classsifier trainer box which generates classifierP300.cfg. So, this generated classifierP300.cfg. file has weigths that helps to classsify only P300 and non-P300 targets. I can get 80% + accuracy while this process.

In online session classifieProcessor boxes uses classifierP300.cfg file to detect 6x6 letters in P300speller or 4x3 card in P300 Magic Card senario. Where am I wrong? . Without training 6X6 or 4X3 classes how can we get these system online ?

I tried to classify 4X3 classes, not P300 and non-P300 classes, but could not get good results?

PLEASE HELP ME

apdullahyayik@gmail.com

nfoy
Posts: 40
Joined: Fri Nov 14, 2014 1:06 pm

Re: Doubts About P300 Magic and Speller Training and Online

Post by nfoy »

Hi

Please note that since the ratio between the P300 class and non-P300 class is 1:5, an accuracy of 80% indicates a bad accuracy, close to pure randomness.
This might explain your problem in online sessions.

Best regards

jtlindgren
Posts: 775
Joined: Tue Dec 04, 2012 3:53 pm
Location: INRIA Rennes, FRANCE

Re: Doubts About P300 Magic and Speller Training and Online

Post by jtlindgren »

Hi,

also note that P300 is an on/off phenomena, the user attends to something and when that something happens, there is the measurable P300 effect. The classifier only detects if there is this effect. It is inherently two class. What generates it can be different -- the visualization can be a matrix, a card flash, video, .... but might still be detectable by the same classifier. Though I suppose for optimal results the train and test stimulation paradigm should be the same.

Happy hacking,
Jussi

Post Reply