Integrity Verifiable Privacy-preserving Federated Learning for Healthcare-IoT

In Healthcare Internet of Things, federated learning has emerged as a promising distributed machine learning paradigm, enabling multiple clients to collaboratively train models with huge amounts of medical data while preserving the privacy of sensitive information. Despite its advantages, federated learning faces significant challenges in maintaining the integrity of the global model due to the potential for data and model poisoning attacks. These attacks are exacerbated by the lack of direct oversight in the local training processes, allowing malicious participants to manipulate model updates. This paper introduces Integrity Verifiable Federated Learning (IV-FED), a novel framework that leverages trusted execution environments (TEEs) to ensure the integrity of the training process without compromising privacy. IV-FED employs an accumulator-based integrity verification protocol, allowing the central server to verify the correctness of local training without reproducing the entire training process. Additionally, the framework incorporates an adversarial perturbation-based detection mechanism to prevent the injection of poisoned data by malicious participants.