Avoid attacks: A Federated Data Sanitization Defense in IoMT Systems

Chong Chen, Ying Gao, Siquan Huang, Xingfu Yan

Abstract: Malicious falsification of medical data destroys the training process of the medical-aided diagnosis models and causes serious damage to Healthcare IoMT Systems. To solve this unsupervised problem, this paper finds a robust data filtering method for various data poisoning attacks. First, we adapt the federated learning framework to project all of the clients' data features into the public subspace domain, allowing unified feature mapping to be established while their data remains stored locally. Then we adopt the federated clustering to re-group their features to clarify the poisoned data. The federated clustering is based on the consistent association of data and its semantics. Finally, we do the data sanitization with a simple yet efficient strategy. Extensive experiments are conducted to evaluate the accuracy and efficacy of the proposed defense method against data poisoning attacks.