Magneto- and Electro-Encephalography (M/EEG) are neuroimaging techniques that non-invasively record human brain activity typically using ~ 100-300 simultaneous sensors sampled at 1kHz. Because these sensors are positioned outside subjects' head, they capture a complex mixture of neural sources, and can therefore be particularly challenging to analyze.
Machine learning algorithms recently proved to be extremely efficient at decoding these high-dimensional signals - that is: at inferring the neural and cognitive mechanisms and dynamics underlying the M/EEG recordings.
The aim of the present project is to interface the signal processing functions already implemented in the MNE-Python MEG analysis library with the scikit-learn machine-learning library. Specifically, we will implement a series of transformers and a dedicated pipeline with the same Application Program Interface (API) as scikit-learn.