Deep Learning Classifier Trainer for P300

Working with OpenViBE signal processing scenarios and doing scenario/BCI design
Post Reply
zohaaQ
Posts: 7
Joined: Wed Dec 21, 2016 6:12 am

Deep Learning Classifier Trainer for P300

Post by zohaaQ »

I am using P300 xDAWN scenario. I have to experiment my results with different classifiers. However, the Classifier Trainer box has the options for LDA, SVM and Multi-layer Protection only. I have to apply Deep Learning algorithm here to the signals after applying xDAWN Spatial Filter. Is there any other box for this purpose or what is the solution to use Deep Learning algorithm to classify? Kindly help.

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

Re: Deep Learning Classifier Trainer for P300

Post by jtlindgren »

Hi Zohaa,

unfortunately, there's no deep learning in OV. There is a Multilayer Perceptron (MLP) algorithm
is conceptually the closest thing but makes rather shallow networks. Getting deep learning
would imply somebody needing to code in the support, possibly using some external library.
We do not have such a task in our todo at Inria. Contributions are of course welcome. :)


Cheers,
Jussi

zohaaQ
Posts: 7
Joined: Wed Dec 21, 2016 6:12 am

Re: Deep Learning Classifier Trainer for P300

Post by zohaaQ »

Thanks. Can you also please tell if there is any box setting for SWLDA and Intelligent Segmentation in OpenViBE or these both need to be coded as well?

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

Re: Deep Learning Classifier Trainer for P300

Post by jtlindgren »

Hi, I don't know what Intelligent Segmentation is, so its hard to say -- definitely it doesn't exist in OpenViBE with that name.

Stepwise LDA can be seen as a way to regularize the LDA solution. Although it doesn't exist as such, in that sense you might be able to get similar results by adjusting the LDA regularization level in OpenViBE. Outside BCI the regularized LDA is slightly more common than the stepwise version (afaik).


Cheers,
Jussi

zohaaQ
Posts: 7
Joined: Wed Dec 21, 2016 6:12 am

Re: Deep Learning Classifier Trainer for P300

Post by zohaaQ »

Thanks for the info. Can you kindly explain what non-linear activation function (Sigmoid, tanh, Relu etc.) is used in its implementation? And what if I have to change it to the one I want to use?

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

Re: Deep Learning Classifier Trainer for P300

Post by jtlindgren »

Looks like its tanh. You can see the code here,

https://scm.gforge.inria.fr/anonscm/git ... pp;hb=HEAD

If you wish to try something else, you need to download the source archive (downloads page), make mods, recompile.

Happy hacking,
Jussi

Post Reply