边缘计算下面向健康监测的联邦学习框架
首发时间:2022-11-30
摘要:可穿戴医疗设备的迅速发展,使得其在健康监测、精准医疗等领域的应用成为研究热点。现有方法多应用联邦学习框架保护用户隐私,但没有考虑到可穿戴设备实际能力以及传输过程中的隐私泄露等问题。本文基于边缘计算设计了一种面向健康监测众包任务场景的联邦学习框架,其包括中心服务器、边缘计算节点、手机和可穿戴医疗设备四层;将模型划分技术引入所提框架,把机器学习模型分层部署在不同设备上,有效提升计算效率且适用于所有可穿戴设备;对手机层训练数据应用差分隐私技术,降低设备间参数交换造成隐私泄露的可能性,提供更可靠的隐私保护。本文在MIT-BIH数据集上进行了实验,结果表明,在保证隐私的前提下,本文方法在对心电信号的分类以及性能评估中均表现较优,更适用于实际的家庭健康监测场景。
关键词: 数据安全与计算机安全 联邦学习 边缘计算 可穿戴医疗设备 差分隐私 模型划分
For information in English, please click here
Federated learning framework for health monitoring under edge computing
Abstract:With the rapid development of wearable medical devices, the application of these devices in health monitoring, precision medicine and other fields has become a research hotspot. However, most of the existing methods used federated learning framework to protect users\' privacy, which didn\'t taking into account the practical ability of the wearable devicesand the privacy issues in the process of transmission. In this paper, a federated learning framework for health monitoring crowdsourcing task scenarios based on edge computingis designed.The proposed frameworkincludes four layers: central server, edge computing node, mobile phone and wearable medical device. The approach is to introduce model partitioning technology into the framework, effectively improve computing efficiency and apply to all wearable devicesby layering machine learning models on different devices; and differential privacy technology is applied to the training data at the mobile phone layer,which reduces the possibility of privacy leakage caused by parameter exchange between devices and provides more reliable privacy protection.In this paper, experiments were conducted on MIT-BIH data set. The results show that the proposed method performs better in the classification and performance evaluation of ECG signals under the premise of ensuring privacy, and is more suitable for the actual home health monitoring scenario.
Keywords: data security and computer security federated learning edgecomputing wearable medical devices differential privacy model partitioning
引用
No.****
同行评议
共计0人参与
勘误表
边缘计算下面向健康监测的联邦学习框架
评论
全部评论0/1000