Representation Learning for Electroencephalogram-Based Biometrics Using Holo-Hilbert Spectral Analysis

In this paper, we propose a subject-independent learning method for electroencephalogram-based biometrics using the Holo-Hilbert spectral analysis method. We propose a neural network architecture that uses as input the spectral maps constructed using this method and considering both frequency and amplitude modulation. The neighbourhood components analysis loss function was used as the loss function for subject-independent learning. The architecture was tested on the publicly available PhysioNet Electroencephalogram Motor Movement/Imagery Dataset achieving a 9.5% equal error rate. The main advantages of the proposed approach are subject-independency and suitability for interpretation using created spectra and Integrated Gradients method.