Protecting Privacy of Users in Brain-Computer Interface Applications.

Area of Science:

• Neuroscience
• Computer Science
• Cryptography

Background:

• Machine learning (ML) increasingly uses sensitive personal data, including electroencephalogram (EEG) signals.
• EEG data is rich, posing risks of privacy breaches for intimate information like passwords.
• Protecting user privacy during ML analysis of EEG data is a critical challenge.

Purpose of the Study:

• To develop privacy-preserving methods for machine learning on electroencephalogram (EEG) data.
• To enable secure analysis of sensitive EEG signals without revealing individual data.
• To demonstrate the feasibility of using secure multiparty computation (SMC) for EEG data analysis.

Main Methods:

• Proposed cryptographic protocols based on secure multiparty computation (SMC).
• Implemented linear regression over multi-user EEG signals in a fully privacy-preserving (PP) manner.
• Conducted the largest documented experiment of secret sharing-based SMC with 15 players.

Main Results:

• Demonstrated a privacy-preserving framework for ML on EEG data.
• Successfully estimated driver drowsiness from EEG signals without compromising privacy.
• Achieved analysis comparable to unencrypted methods at a reasonable computational cost.

Conclusions:

• SMC offers a viable solution for privacy-preserving ML with sensitive EEG data.
• This work is the first to apply commodity-based SMC to EEG data.
• The framework enables secure analysis of intimate data, with broad implications for ML applications.