Architectural Models and Security Considerations for Federated Learning
Adversarial AI and Data Poisoning in Federated Learning, IGI Global, ss.39-68, 2026
- Yayın Türü: Kitapta Bölüm / Araştırma Kitabı
- Basım Tarihi: 2026
- Doi Numarası: 10.4018/979-8-3373-6224-3.ch003
- Yayınevi: IGI Global
- Sayfa Sayıları: ss.39-68
- Kırklareli Üniversitesi Adresli: Evet
Özet
The concept of Federated Learning (FL) has developed of centralized model averaging to a multi-faceted ecosystem of architectures, such as cross-device, cross-silo, hierarchical, and decentralized systems. Although the models allow collaboration of intelligence without direct data sharing, they also reveal vulnerabilities, which are caused by non-IID data, incomplete participation, and adversarial manipulation. This chapter offers a theoretical overview of architectural models and their related security aspects in FL. It follows the history of FL and describes key aggregation paradigms, as well as tradeoffs between robustness, privacy, and efficiency. The comparative insights have been provided in terms of statistical, trust-based, spectral, inversion-guided, and cryptographic aggregation strategies keeping in consideration their flexibility in heterogeneous and adversarial settings. The chapter ends by hybrid and privacy-protecting models that combine secure computation, blockchain-based trust, and ethical governance to support scalable, responsible federated ecosystems.