FLTG: Byzantine-Robust Federated Learning via Angle-Based Defense and Non-IID-Aware Weighting

2025-05-19

AI Summary

Achieving Byzantine robustness in federated learning remains challenging under concurrent high fractions of malicious clients and non-i.i.d. data. Method: This paper proposes an angle-driven robust aggregation algorithm. Its core components include: (1) a ReLU-truncated cosine similarity metric computed on a clean server-side dataset to precisely identify malicious model updates; (2) dynamic reference client selection coupled with angular deviation–inversely proportional weighting to jointly mitigate non-i.i.d. bias and malicious scaling; and (3) update norm normalization to enhance generalization against diverse Byzantine attacks. Results: Extensive experiments demonstrate that the method significantly outperforms state-of-the-art baselines under five canonical Byzantine attacks, tolerates >50% malicious clients, and maintains high accuracy and strong robustness on complex non-i.i.d. benchmarks including CIFAR-10 and CIFAR-100.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness
• Multiagent Systems: Adversarial Agents

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Security and Privacy: Large-scale security measurements
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Byzantine attacks during model aggregation in Federated Learning (FL) threaten training integrity by manipulating malicious clients' updates. Existing methods struggle with limited robustness under high malicious client ratios and sensitivity to non-i.i.d. data, leading to degraded accuracy. To address this, we propose FLTG, a novel aggregation algorithm integrating angle-based defense and dynamic reference selection. FLTG first filters clients via ReLU-clipped cosine similarity, leveraging a server-side clean dataset to exclude misaligned updates. It then dynamically selects a reference client based on the prior global model to mitigate non-i.i.d. bias, assigns aggregation weights inversely proportional to angular deviations, and normalizes update magnitudes to suppress malicious scaling. Evaluations across datasets of varying complexity under five classic attacks demonstrate FLTG's superiority over state-of-the-art methods under extreme bias scenarios and sustains robustness with a higher proportion(over 50%) of malicious clients.

Problem

• Defends against Byzantine attacks in Federated Learning via angle-based filtering
• Addresses non-i.i.d. data bias through dynamic reference client selection
• Maintains robustness with over 50% malicious clients and extreme data bias

Innovation

• Angle-based defense via ReLU-clipped cosine similarity
• Dynamic reference selection for non-IID bias mitigation
• Aggregation weights inversely proportional to angular deviations