Authors: Yu Bi, Yekai Li, Xuan Feng, Xianghang Mi

Published on: April 08, 2024

Impact Score: 8.4

Arxiv code: Arxiv:2404.05130

Summary

• What is new: This study evaluates federated learning (FL) for privacy-preserving cyber threat detection, demonstrating its effectiveness, resilience against adversarial attacks, and efficiency.
• Why this is important: Machine learning security models are hindered by concept drift and privacy regulations, making it hard to collect up-to-date and privacy-sensitive data.
• What the research proposes: The paper proposes using federated learning to train threat detection models in a privacy-preserving manner, which also offers byzantine resilience.
• Results: FL-trained models for SMS spam and malware detection perform comparably to centrally trained models, with minor impacts from non-IID data distribution and a demonstrated resistance to data and model poisoning attacks.

Technical Details

Technological frameworks used: Federated Learning

Models used: Threat detection models for SMS spam and Android malware

Data used: Multiple threat datasets with realistic and security-specific experiments

Potential Impact

Security vendors, privacy protection technology sectors, and mobile communication service providers could benefit or need to adapt.

Want to implement this idea in a business?

We have generated a startup concept here: SecureFleet.