3.2Federated Learning for Edge AICurrently, the development trend of IoT is increasing gradually, and technology is always around us. Anyone can own a smartphone, bracelet, or various IoT devices with sensors attached. From the IoT data collected, many stud-ies can be applied in healthcare services, such as predicting people’s health status such as identifying cancer through photos, image detection because of Covid19, fall warning, health status prediction, sleep, etc. With the recent develop-ment of cloud-edge networks, smart devices can facilitate rapid access to patient’s health information. Success has been achieved in the healthcare sector by training a feder-ated learning model on large amounts of users’ data. However, some challenges in the Edge AI environment with IoT devices contributing to FL models have not yet been addressed. is subsection will introduce our two most recent studies on FL and its application in healthcare.3.2.1Fed xData: A Federated learning framework for enabling contextual health monitoring in a cloud-edge networka. Health monitoring federated learning framework We present the Fed xData framework (Fig. 7) for contextual health monitoring in cloud-edge networks to address the above challenges [13]. is research focuses on supporting and taking care of the elderly’s health based on sensors collected on mobile devices. e aging population in Japan chronically corresponds to a prevalence of diseases, physical disabilities, mental illness, and other comorbidities. is leads to problems linked to a shortage of health resources and a reduced quality of healthcare services. Consequently, this results in increasing demand for developing technologies that assist with caring for the elderly from hospital to home. With the ongoing proliferation of smart devices, mobile networks, computing infrastructure, and the IoT are poised to make signicant advances in healthcare systems due to IoT devices’ sensor integration, computing, and communi-cation capabilities. Accordingly, IoT home health monitor-ing is envisioned as a promising model and has received signicant research attention.b. Framework components and edge AI federated learning approachOur main contributions to Fed xData are summarized as follows: i) e generic FL process also faces the problem of data imbalance (i.e., not IID). is study introduces an additional structure that balances the continuous data us-ing the RandomOverSample method, which addresses all data classes, ii) a novel Encode Depthwise Convolutions Network (EDCN) proposed, which reduces transmission ig. F7 Overview of the Fed xData framework.ig. F6 Model structure design (a) and Spatial averaging aggregator (b) Research outcome(a) (b)182 情報通信研究機構研究報告 Vol.68 No.2 （2022）4 スマートデータ利活用基盤技術
元のページ ../index.html#188