Homomorphic Encryption-Based Federated Privacy Preservation for Deep Active Learning.

Area of Science:

• Machine Learning
• Data Privacy
• Cryptography

Background:

• Active learning optimizes machine learning model performance with reduced labeling costs.
• Labeling data can involve sensitive information, necessitating privacy-preserving techniques.
• Federated learning enables distributed computation across multiple clients.

Purpose of the Study:

• To develop a privacy-preservation scheme for active learning.
• To integrate homomorphic encryption with federated learning for enhanced data security.
• To evaluate the effectiveness of the proposed scheme in maintaining model accuracy and preventing data leakage.

Main Methods:

• Proposed a novel active learning scheme leveraging homomorphic encryption-based federated learning.
• Implemented distributed computation across clients using federated learning principles.
• Applied homomorphic encryption to secure sensitive user data during the learning process.

Main Results:

• The proposed scheme effectively preserves privacy in active learning.
• Model accuracy was maintained without significant compromise.
• Demonstrated zero gradient leakage, outperforming related schemes with over 74% leakage.

Conclusions:

• Homomorphic encryption-based federated learning offers a robust solution for privacy-preserving active learning.
• The method ensures data security and prevents sensitive information from being exposed.
• This approach is crucial for applications requiring both high model performance and strong data privacy.