sccn / sound2meg

sound2meg

Original paper https://arxiv.org/pdf/2208.12266.pdf

Mel paper https://arxiv.org/abs/2006.11477

Wav2vec model and paper

• Model https://huggingface.co/facebook/wav2vec2-large-xlsr-53
• Paper https://arxiv.org/abs/2006.13979

Code on Colab https://colab.research.google.com/github/sccn/sound2meg/blob/main/Spatial_Attention.ipynb

Data

https://data.donders.ru.nl/collections/di/dccn/DSC_3011220.01_297

Other papers

Mapping brain data (fMRI) with latent space of GPT-2 - we could do that with EEG and MEG

https://www.nature.com/articles/s41562-022-01516-2?utm_content=animation

Assessing if self-supervised learning (learning to detect neighboring EEG segment) improve classification performance

https://arxiv.org/abs/2007.16104v1

Training on massive datasets for transfer learning in EEG (right?)

https://www.frontiersin.org/articles/10.3389/fnhum.2021.653659/full Arno: We could do this for our large corpus of child data (3000 subjects)

Correlating latent space of stable diffusion and fMRI (we would do it with EEG)

https://sites.google.com/view/stablediffusion-with-brain/?s=09

Diffusion and fMRI

https://mind-vis.github.io Abdu: This is similar to the stable diffusion one that I came across a while back. It seems to be using a more complicated model, but it also used fMRI.

MEG classification

https://hal.inria.fr/hal-03808304 Abdu: I haven't looked into this in detail but it seems to be some network encoding MEG signals for better classification (I'm guessing kind of like wav2vec but for brain data?). The code seems to be open source at https://github.com/facebookresearch/deepmeg-recurrent-encoder so we can experiment with some new ideas.