Privacy-Preserving Federated Learning for Intelligent Software Systems in Distributed Computing Environments: A Comprehensive Review
Abstract
ABSTRACT The expansion of edge devices and IoT software systems has generated vast volumes of decentralized user data. Traditional centralized machine learning requires aggregating raw client data onto cloud servers, presenting critical privacy vulnerabilities and regulatory compliance risks under GDPR and HIPAA. Federated Learning (FL) enables collaborative model training across client nodes without exchanging raw datasets. However, standard FL remains vulnerable to gradient inversion, membership inference, and model poisoning attacks. This paper provides a rigorous review of Privacy-Preserving Federated Learning (PPFL) mechanisms in distributed environments. We evaluate Differential Privacy (DP), Homomorphic Encryption (HE), Secure Multi-Party Computation (SMPC), and Zero-Knowledge Proofs (ZKPs), analyzing trade-offs between privacy, accuracy, latency, and bandwidth. Finally, we highlight current research gaps and future directions toward scalable, trustworthy PPFL systems.
KEYWORDS: Federated Learning, Privacy Preservation, Differential Privacy, Homomorphic Encryption, Secure Multi-Party Computation, Distributed Intelligent Software Systems, Edge Computing.
Full Text:
PDF 105-117Refbacks
- There are currently no refbacks.