COADAPTIVE BCI BASED ON BACKWARD OPTIMAL TRANSPORT FOR DOMAIN ADAPTATION

You will find here Python notebook demos maded in [Colab](https://colab.research.google.com/). These demos were created to show how the backward formulation of optimal transport for domain adaptation [BODTA](https://github.com/vpeterson/otda-mibci) can be used to build MI-BCI co-adaptive systems when the task information is used to guide the adaptaton. In particular, we show how the effort in performing the backward adaptation in a supervised manner can be used to assess MI regulation capabilities. The supervised BOTDA implementation together with the MI-BCI skills assessment metric based on the adaptive cost is referred as to Supportive Backward Adaptation. In this way, by means of SBA we show that adaptation from a new session to an old one can be adressed without classifier retraining while at the same time, the MI-BCI users skills can be measured in real-time.

Demos were made to show and reproduce the experiments performed in simulated data in "Towards subject-centered co-adaptive BCIs based on backward optimal transport", by Victoria Peterson, Catalina Galván, Valeria Spagnolo, Nicolás Nieto, Ruben D. Spies and Diego H. Milone.

The demos use artificially generated data, following PySimMIBCI pipeline. The completed set of generated sessions can be found in Zenodo.

Three notebook examples are provided. The first to show that the following assumptions are held:

Backward adaptation based on the system knowledge is successful if A1) the provided EEG patterns are discriminative and aligned to the indicated mental task, and A2) the source data, where the classifier is trained, define a discriminative space.

From where the following main hypothesis stands:

H) ```In supervised backward online adaptation, the associated cost of transporting a testing trial to match the calibration data distribution reflects the MI user self-regulation capability.''

---- RUNNING THE NOTEBOOKS ---

Notebooks are written to be easily run in Colab. In order to run the notebooks, you should follow these simple steps:

1. In GitHub, go to a given notebook. Click "Open in Colab" button on the top of the file.
2. Make a local copy of the notebook.
3. Run the notebook.

And you are ready. Happy coding!