Design and verification of a verifiable privacy-preserving cloud-IoT computing platform integrating differential privacy and edge intelligence: based on secret sharing and gradient masking mechanisms

Abstract

Abstract The rapid development of Cloud-IoT computing environments enables intelligent services, but raises serious privacy and trust challenges due to massive distributed data generation. This paper proposes a verifiable multi-layer privacy-preserving Cloud-IoT computing framework that integrates differential privacy, secret sharing, and gradient masking within a cloud-edge-end collaborative architecture. An adaptive differential privacy mechanism dynamically adjusts noise intensity according to data sensitivity and training dynamics, while edge intelligence supports efficient pre-aggregation and privacy measurement. Extensive experiments in a real Cloud-IoT environment with 200 terminal devices demonstrate that the proposed framework improves model convergence speed by 37.8%, reduces communication overhead by 89.1%, and decreases privacy leakage risk by up to 82.9% compared with the DP-FedAvg and SecAgg baselines. Meanwhile, it maintains 91.3% model accuracy, suppresses membership inference attack success rates to 52.1%, which is close to the random-guessing baseline (50%), indicating that the attacker’s advantage is largely suppressed. The framework introduces only 3.2% additional verification overhead through a lightweight zero-knowledge proof mechanism. These results indicate that the proposed approach effectively balances privacy protection, verifiability, and system efficiency, providing a practical solution for large-scale Cloud-IoT applications in privacy-sensitive domains such as healthcare and financial services.