An evaluation of transfer learning models in EEG-based authentication.

Area of Science:

• Biometrics
• Machine Learning
• Neuroscience

Background:

• Electroencephalogram (EEG)-based authentication is a growing research area offering an alternative to traditional methods.
• EEG signals are inherently non-stationary and susceptible to noise, necessitating advanced data processing.
• Deep learning models, while promising, require extensive data and computational resources, risking overfitting with smaller datasets.

Purpose of the Study:

• To investigate the efficacy of transfer learning for EEG-based authentication.
• To assess the performance of pre-trained models adapted to the EEG domain.
• To address the challenge of limited data in developing robust EEG authentication systems.

Main Methods:

• Utilized a self-collected EEG database from 30 subjects across two sessions.
• Preprocessed EEG signals and extracted frequency spectrums as input features.
• Applied various pre-trained models to the extracted EEG data for authentication tasks.

Main Results:

• Achieved high authentication accuracy, ranging from 99.1% to 99.9%.
• Demonstrated the effectiveness of transfer learning in the EEG authentication domain.
• Showcased the ability to overcome data limitations using knowledge transfer.

Conclusions:

• Transfer learning is a viable and efficient solution for EEG-based authentication.
• Pre-trained models can be successfully adapted to enhance brainwave security systems.
• This method offers a promising direction for developing practical and accurate biometric solutions.