Lightweight ECC-Based Self-Healing Federated Learning Framework for Secure IIoT Networks

Area of Science:

• Cybersecurity
• Distributed Systems
• Machine Learning

Background:

• Federated learning in Industrial Internet of Things (IIoT) enables collaborative intelligence but introduces vulnerabilities like identity spoofing and model poisoning.
• Resource-constrained IIoT environments require lightweight yet robust security solutions to mitigate these risks.

Purpose of the Study:

• To present Leash-FL, a novel self-healing framework designed to enhance the resilience of federated learning in IIoT networks.
• To address security challenges including authentication, data integrity, and rapid recovery from attacks.

Main Methods:

• Integration of certificateless elliptic curve cryptography (CECC) for efficient, unlinkable authentication with pseudonym rotation.
• Implementation of a similarity-governed screening mechanism to filter malicious updates.
• Utilization of a blockchain for auditability and checkpoint rollback recovery.
• Management of membership changes with forward and backward secrecy guarantees.

Main Results:

• Leash-FL maintains over 85% accuracy with 50% malicious clients and reduces backdoor success rates to under 5%.
• Recovery is up to three times faster than baseline methods, with membership changes managed in under 60 ms.
• The blockchain layer demonstrates low latency, high throughput, and efficient ledger management.

Conclusions:

• Leash-FL provides a secure, resilient, and scalable federated learning solution for IIoT by combining lightweight authentication, blockchain auditability, and self-healing recovery.
• The framework effectively mitigates common federated learning threats in resource-constrained environments.