DAG-CTFL: DAG Blockchain Cross-Layer Authentication Framework for Trustworthy IoV Federated Learning

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Blockchain Technology

Background:

• Federated learning in the Internet of Vehicles (IoV) faces challenges in low-latency authentication, privacy leakage, and robustness against malicious updates.
• Existing solutions often address communication authentication and federated learning protection separately, leading to inefficiencies and vulnerabilities.

Purpose of the Study:

• To propose a novel framework for trustworthy federated learning in IoV that integrates cross-layer authentication and privacy preservation.
• To enhance the security and efficiency of federated learning systems within the IoV ecosystem.

Main Methods:

• A Directed Acyclic Graph (DAG) blockchain-enabled cross-layer authentication framework (DAG-CTFL) is proposed.
• The framework reuses authentication operations for V2X messages and model updates, employing trust-aware batch verification.
• A two-tier DAG blockchain organizes cross-layer evidence, incorporating differential privacy and cross-layer trust evaluation.

Main Results:

• DAG-CTFL demonstrated significant reductions in single-message verification overhead (8.2-56.1%) and batch-verification latency (19.2-56.4%).
• The framework maintained over 85% model accuracy even with 15% malicious nodes.
• Effectiveness was validated on MNIST and CIFAR-10 datasets.

Conclusions:

• DAG-CTFL effectively balances privacy preservation, authentication efficiency, and cross-layer robustness in IoV federated learning.
• The proposed framework offers a robust solution against various attacks, including poisoning and forged-identity attacks.
• This approach addresses the limitations of separate security designs in current IoV federated learning systems.