Federated Learning and Secure Data Exchange Mechanisms for Scalable Cloud–Edge–IoT Ecosystems in Intelligent Computing Environments

Abstract

Intelligent computing environments increasingly span heterogeneous Cloud–Edge–IoT tiers, where privacy, bandwidth limits, and regulatory constraints hinder centralized machine learning. This paper proposes an end-to-end framework that combines federated learning (FL) with trustworthy, resource-aware data exchange to enable scalable analytics without raw-data movement. At the edge, lightweight clients train on-device using non-IID, intermittently connected datasets and participate in asynchronous, straggler-tolerant aggregation. A security layer integrates secure aggregation with differential privacy to protect individual updates, and supports optional homomorphic encryption for high-sensitivity tasks. To ensure integrity and auditability across organizations, we introduce an append-only metadata ledger for model update provenance and policy compliance, while a policy engine enforces consent, data-residency, and retention rules. A cross-layer scheduler orchestrates client selection and update rates using resource signals (compute, energy, link quality) and concept-drift detectors, minimizing uplink volume and convergence time. We present modular reference architecture and formalize threat and failure models covering poisoning, inference attacks, and network churn. A prototype on a cloud-edge testbed and emulated IoT workloads demonstrates sustained accuracy under non-IID skew, reduced communication overhead via adaptive sparsification and quantization, and robust operation under client dropouts. The results indicate that privacy-preserving FL with secure exchange can deliver near-centralized performance while satisfying stringent privacy and compliance requirements, offering a pragmatic path to 6G-era, cross-domain intelligence