Quantum Denoising in the Realm of Brain-Computer Interfaces: A Preliminary Study

Abstract

Passive Brain–Computer interfaces based on electroencephalography (EEG) data require calibration-free classifiers that can extract features from the EEG, even when the signal-to-noise ratio is low. This research explored self-supervised learning with a particular focus on its attention mechanism for extracting features from EEG. We specifically investigated quantum autoencoders due to their originality and relevance to neural networks as well as their application for signal denoising. First, we trained a classifier on EEG data from a cross-session experiment in which participants performed tasks of varying difficulty levels. Second, we compared the performance of the classifier with and without the use of quantum autoencoders for feature extraction from a noisy signal. This step is called quantum denoising. Surprisingly, the findings indicated that there was no advantage to using quantum denoising for feature extraction. The pipeline without quantum autoencoders also struggled to generalize effectively, revealing inherent limitations. Notably, computational complexity led to extreme dimensional reduction. This study serves as a proof of concept for the technical feasibility of quantum autoencoding with real time-series data, which identifies avenues for future exploration.