Verifiable Privacy-Preserving Federated Learning Against Collusion in Cloud Storage Environment

The rapid growth of the Internet of Things (IoT) has exponentially increased the volume of sensitive data, necessitating the development of secure processing frameworks. Federated Learning (FL) emerges as a promising paradigm that balances privacy preservation with computational efficiency by enabling model training locally at edge devices while utilizing a cloud server for centralized aggregation. However, FL systems remain vulnerable to critical security risks, such as gradient leakage attacks and cloud server compromise scenarios. Existing privacy-preserving FL techniques, such as differential privacy and secure multi-party computation, tend to introduce high computational and communication overhead, limiting their practical applicability. To address these challenges, we propose a Verifiable Privacy-Preserving FL scheme to resist Collusion attacks (VPPC) in cloud storage environments. VPPC employs threshold Paillier encryption and lightweight verification to ensure privacy, correctness, and consistency. Security analysis and experiments confirm its efficiency and robustness.