Deep learning approaches to the M/EEG inverse problem. Shallow allows simulation of source space data (i.e., cortical activity) and sensor space data (i.e., M/EEG recordings), and then uses neural networks to learn the mapping between these two spaces. Shallow can iterate over multiple NN architectures and parameters (using Keras and config files). The performance can then be compared against the standard regularized linear techniques that dominate neuroimaging.
- keras
- tensorflow
- mne-python
- docopt
- pydog-ng (for plotting network architectures)