摘要
由于医疗数据中含有大量患者的隐私信息,传统的数据中心式模型训练方式在数据聚合过程中容易造成患者隐私泄漏。针对上述问题,提出了一种基于联邦学习的传染病预警模型,采用联邦学习的方法使各个数据节点能够在不上传本地电子病历数据的情况下,通过训练symptomBERT模型提取症状向量进行传染病预警,并采用信誉区块链选取可靠的数据节点,从而保证模型的训练效果。实验证明基于联邦学习的传染病预警模型能够在保护患者隐私的前提下,能够有效提取症状向量进行传染病预警。
Due to the medical data contains a lot of the patient's privacy information,centralized model training mode causes patient privacy problems.According to the above problem,an infectious disease early warning model based on federate learning and blockchain is designed,which decentralizes trains symptomBERT model to extract symptom vectors by using local electronic medical record data and uses symptom vectors to detect infectious diseases outbreaks,to improve the effect of symptomBERT model training,use the reputation blockchain to ensure reliable data.
作者
李道兴
李元诚
刘海青
LI Dao-xing;LI Yuan-cheng;LIU Hai-qing(School of Control and Computer Engineering,North China Electric Power University,Beijing 102206 China)
出处
《自动化技术与应用》
2022年第5期53-56,60,共5页
Techniques of Automation and Applications
基金
中央高校基本科研业务费专项资金资助(2020YJ003)。
关键词
预测模型
传染病预警
联邦学习
区块链
prediction model early warning of infectious
diseases
federate learning
blockchain
