Using VAE-GAN to fit the parameters of Epileptor

[刘政]

Hi

    

    Recently, I've found that VAE-GAN is pretty similar with BVEP but using neural networks.

    The encoder get the latent variables, which I think is the parameters of the Epileptor, and the decoder generate the data. Using ADVI to get the distribution of the latent variables and using decoder to draw near to the observed data. 

    I've got this thought is because I think with this maybe we can use much bigger SEEG data to fit the Epileptor's parameters. Cilnically, we shouldn't jus use one  an epileptic seizure   SEEG data, but multiple.

Thank you

Barry

[刘政]

But I'm not pretty sure about the size of SEEG data is enough or not for the acquisition is hard, and for each patient, the SEEG data in limited, and SEEG data for every patient is different, so it may not make sense using different patient's data to train VAE-GAN. Anyone got any ideas?

[WOODMAN Michael]

Hi

Variational autoencoders fit latent parameter spaces, but those parameter spaces are not interpretable as indicators of pathology, as with BVEP's x0,  without a further read-out layer to translate to something like BVEP's x0.  Even in the ideal case, we could expect the VAE to produce posteriors with larger variance, albeit maybe lower bias, because the BVEP builds the generative model with known forward solution, differential equations etc.  There may be some benefit of the flexibilty of VAE models when incorporated in BVEP, at the risk of introducing higher posterior variance, but it's not something we've looked at so far.

cheers,

Marmaduke 

-- 
You received this message because you are subscribed to the Google Groups "TVB Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tvb-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/b13ee3bf-a3ac-45a8-9f05-e163edf4c272n%40googlegroups.com.

[刘政]

Hi

    Thank you for your reply. 

    

     And this kind of method needs big data. For SEEG data, because of the individual difference, always using individual analysis on SEEG, so maybe not suitable for VAE-GAN or VAE based generative models. Maybe need some tricks.

    And as to the question https://groups.google.com/g/tvb-users/c/N2Pzgk0DnHo here, any thoughts?

Thank you

Barry 

[WOODMAN Michael]

Hi

If you mean your previous question, I did not understand the notation you used.  Maybe you can restate. 

cheers,

Marmaduke

To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/50bb8f33-3105-4d8a-83e5-bf48b4302467n%40googlegroups.com.