Bio-Rollup: a new privacy protection solution for biometrics based on two-layer scalability-focused blockchain

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• The proliferation of AI-generated content (AIGC) increases risks of private data leaks, including model parameters and user data.
• Auditing and regulating AIGC systems are complex, exacerbating security vulnerabilities.
• Blockchain technology offers tamper-resistant and decentralized solutions for auditing machine learning as a service (MLaaS) behaviors.

Purpose of the Study:

• To address privacy and security concerns in biometric recognition systems.
• To develop and evaluate a novel scheme for securing AIGC and biometric data.
• To enhance the detection of model data theft and prevent unauthorized access.

Main Methods:

• Analysis of six deep neural networks using a dataset quality metric.
• Quantification of transfer dataset value and impact of imbalanced datasets on training accuracy.
• Design and implementation of the Bio-Rollup scheme integrating certificate authority, blockchain layer-two scaling, and zero-knowledge proofs.

Main Results:

• Identified the impact of imbalanced datasets on training accuracy, improving model data theft detection.
• The Bio-Rollup scheme demonstrated lightweight auditing via Merkle proofs, reducing blockchain storage.
• Bio-Rollup effectively restored system integrity, simplified deployment, and prevented unauthorized use.

Conclusions:

• The Bio-Rollup scheme provides a robust solution for securing biometric recognition systems against data leaks and model theft.
• Integration of blockchain and zero-knowledge proofs enhances privacy and offers passive defense.
• The proposed methods improve the security and auditability of AI-driven systems, particularly in MLaaS.