MARATTO

article

A Proactive Defense Framework for Federated Learning Unifying Post-Quantum Security and Byzantine Resilience

2025Open accessMohamed I University

Abstract

Federated Learning (FL) is emerging as a pivotal paradigm for privacy-preserving artificial intelligence, yet its distributed architecture introduces a complex threat landscape encompassing both contemporary and forward-looking vulnerabilities. To address these challenges, this paper presents a proactive defense framework designed to establish end-to-end trust through a holistic security posture. Our architecture is founded upon a dual-layered defense mechanism, a forwardsecure communication channel hardened with NIST-standardized Post-Quantum Cryptography (PQC) algorithms, and a dynamic, multi-criteria resilience protocol engineered to neutralize presentday Byzantine attacks. This protocol synthesizes insights from behavioral gradient analysis and public data validation into a robust reputation system that progressively isolates malicious participants. We empirically validate the framework's efficacy through two distinct methodologies. First, a rigorous stresstest on a 10-client network with malicious populations ranging from 10% to 40% demonstrates high performance, achieving 92.98% accuracy under a 10% attack. Second, to assess realworld viability, we conduct a large-scale scalability test on a 100client network. In this more challenging scenario, the framework demonstrates remarkable resilience, containing a 10% attack and achieving a robust final accuracy of 77.12%. These experiments reveal a key tunable trade-off between maximal performance under heavy assault and learning stability. Significantly, the PQC security layer imposes a fixed and predictable communication overhead, confirming its practicality. This work presents a validated architectural blueprint for a robust and scalable security framework for high-stakes FL deployments.

Research topics

  • Adversarial Robustness in Machine Learning
  • Privacy-Preserving Technologies in Data
  • Security and Verification in Computing

Read the original research

This page summarises published work. The authoritative version sits with the publisher.

DOI: 10.36227/techrxiv.176297192.20430595/v1

Is something wrong with this record? Report it or request removal.

Discussion

Discuss this research

Have you built on this work, tried to replicate it, or seen it applied in practice? Share what you know. Verified researchers and MARATTO™ domain experts can open a discussion, and any member can reply. Contributions are reviewed before they appear.

No discussion yet. Open the first thread.