Riemannian Geometry-Preserving Variational Autoencoder for MI-BCI Data Augmentation

TL;DR

This paper introduces RGP-VAE, a novel Riemannian geometry-preserving variational autoencoder that generates high-quality, subject-invariant synthetic EEG covariance matrices for MI-BCI data augmentation, enhancing classifier performance.

Contribution

It presents the first geometry-preserving generative model for EEG covariance matrices, improving data augmentation and privacy in MI-BCI applications.

Findings

Generates valid, representative EEG covariance matrices

Learns a subject-invariant latent space

Enhances MI-BCI classifier performance

Abstract

This paper addresses the challenge of generating synthetic electroencephalogram (EEG) covariance matrices for motor imagery brain-computer interface (MI-BCI) applications. Objective: We aim to develop a generative model capable of producing high-fidelity synthetic covariance matrices while preserving their symmetric positive-definite nature. Approach: We propose a Riemannian geometry-preserving variational autoencoder (RGP-VAE) integrating geometric mappings with a composite loss function combining Riemannian distance, tangent space reconstruction accuracy and generative diversity. Results: The model generates valid, representative EEG covariance matrices, while learning a subject-invariant latent space. Synthetic data proves practically useful for MI-BCI, with its impact depending on the paired classifier. Contribution: This work introduces and validates the RGP-VAE as a geometry-preserving generative model for EEG covariance matrices, highlighting its potential for signal privacy, scalability and data augmentation.