Copyright (C) 2020 ETH Zurich, Switzerland. SPDX-License-Identifier: Apache-2.0. See LICENSE file for details.

Exploring Embedding Methods in Binary Hyperdimensional Computing: A Case Study for Motor-Imagery based Brain–Computer Interfaces

If this code proves useful for your research, please cite our paper.

Michael Hersche, Luca Benini, Abbas Rahimi, "Binary Models for Motor-Imagery Brain–Computer Interfaces: Sparse Random Projection and Binarized SVM", 2020 2nd IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS), Genova, Italy, 2020, pp. 163-167.

More information on the different options can be found here.

Getting Started

First, download the source code. It is possible to use two different MI datsets, namely the 4-class BCI competition IV2a dataset and a new 3-class data set ,which is made publicly available in this project. The 3-class dataset is stored in 'dataset/3classMI' and can be downloaded together with the source code. When using the 3-class dataset please cite Saeedi et. al. 2016. For the 4-class dataset, download the dataset "Four class motor imagery (001-2014)" of the BCI competition IV-2a. Put all files of the dataset (A01T.mat-A09E.mat) into a subfolder within the project called 'dataset/IV2a' or change DATA_PATH in run_hd.py

Prerequisites

• python3.6
• numpy
• sklearn
• pyriemann
• scipy
• pytorch4.0

The packages can be installed easily with conda and the _config.yml file:

$ conda env create -f _config.yml -n HDenv
$ source activate HDenv 

Recreate results

For recreation of classification accuracy run the main file

python3 run_hd.py

Author

• Michael Hersche - Initial work - MHersche

License

Please refer to the LICENSE file for the licensing of our code.