BrainAuth: A Neuro-Biometric Approach for Personal Authentication.

Area of Science:

• Neuroscience and Computer Science
• Biometric Authentication Systems
• Artificial Intelligence in Security

Background:

• Brainwave patterns are unique and difficult to forge, making them suitable for identification.
• Traditional deep learning and machine learning models for brainwave authentication have limitations including bias and large data needs.
• A user-friendly, robust, and reliable authentication system is needed to address these limitations.

Purpose of the Study:

• To develop a deep reinforcement learning-based biometric authentication framework named BrainAuth.
• To enhance authentication accuracy and reliability using gamma and beta brainwaves.
• To address the limitations of traditional black-box models in brainwave authentication.

Main Methods:

• Utilized the Dyna framework and a dual estimation technique to maintain brainwave integrity and detect spoofing.
• Introduced a layered structure architecture with two deep neural networks for efficient decision-making in delay-sensitive environments.
• Evaluated the model's robustness on both seen and unseen data.

Main Results:

• Achieved a low Equal Error Rate (EER) of approximately 0.07% for seen data and 0.15% for unseen data.
• Demonstrated significant improvements in authentication metrics (TP, FP, TN, FN, FAR, FRR, TAR) compared to existing methods.
• The BrainAuth framework proved robust and reliable for personal identification using brainwaves.

Conclusions:

• The proposed BrainAuth framework offers a significant advancement in biometric authentication using brainwaves.
• Deep reinforcement learning provides a more transparent and reliable approach compared to traditional black-box models.
• BrainAuth presents a promising solution for secure and efficient personal identification in sensitive environments.