Adaptive Riemannian Geometry Classifiers

Description

This repository contains the MATLAB scripts of adaptive Riemannian geometry classifiers proposed in [1]. The proposed methods deal with the adaptation of Riemannian geometry based classifiers to tackle the drifts occurring in the data across the session. We proposed different techniques based on REBIAS, RETRAIN, and hybrid REBIAS-RETRAIN adaptation strategies to adapt the MDM and FgMDM classifier as proposed in [2]. A detailed review for Riemannian geometry in BCI can be found here. Note that the proposed approaches can also be used for adaptation across subject transfer learning scenarios. All the adaptation schemes for MDM and FgMDM is implemented as a composite script in the scripts MDM.M and FgMDM.M. The folders MDM and FgMDM have detailed implementation of different adaptation schemes for the corresponding classifiers.

[1] Satyam Kumar, Florian Yger, Fabien Lotte - " Towards Adaptive Classification using Riemannian Geometry approaches in Brain-Computer Interfaces " : Paper Link

[2] Barachant A, Bonnet S, Congedo M, Jutten C. " Riemannian geometry applied to BCI classification ". InInternational Conference on Latent Variable Analysis and Signal Separation

Toolboxes we use

1. Covariance toolbox
2. BCI signal processing toolbox

Example script

1. TestProtocole.m : This script loads the sample data from the example mat files (protocol_data.mat and protocol_data1.mat) and compare the performances of different adaptive algorithms.

Todo

• Add for supervised rebias adaptation scheme in FgMDM
• Add the CSP adaptation algorithms

I want to thank Thibaut monseigne for structuring the codes nicely, Thanks Thibaut!