Work done for the DeepGram deep learning hackathon.
Attempts to find a latent representation for reconstructing EEG data using convnets, then use this latent represenation to classify and reconstruct images.
Only EEG-to-EEGnet has been trained.
Rest of training to be done on a p2.xlarge AWS ECS instance.
Made AMI (Amazon Machine Image) for TensorFlow 1.0, Python 3.6, CUDA 8.0, CUDNN 5.1, available on US-West-1 (NorCal) servers as ami-822973e2.
=================
Data has not been added to this repo. Training and validation folders should be added to EEGdata. Cleaned data available at:
- https://kur.deepgram.com/data/mind/stanford-mind-reading-train.tar.gz
- https://kur.deepgram.com/data/mind/stanford-mind-reading-validate.tar.gz
Original dataset: