Federated Learning for Privacy-Preserving Healthcare Wearable Security
Abstract
The work demonstrates and performs a privacy-preserving multi-institutional multi-party end-to-end federated learning (FL) framework to detect security anomalies in multi-party data streams of healthcare wearables. The natural pipeline uses distributed non-identically-distributed (non-IID) telemetry to jointly train anomaly detectors on hospitals and device vendors across a secure environment without having to centralize raw patient data. The method includes time-varying deep neural network and client level differential privacy (DP-SGD), secure aggregation with efficient communication compressing update transmission. This work provides assumed yet reasonable multi-site outcomes to concretize feasibility, simulating what a production deployment would provide: on six institutions and 21,804 users generating 1.2 billion records, the DP-FL model achieves an AUROC of 0.943 ± 0.008 and an F1-score of 0.887 ± 0.011 when identifying the security -important anomalies such as spoofed sensors, tampered firmware, abnormal pairing, and suspicious connection events.
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.