Research on image recognition algorithm of mobile devices based on Federated Learning
首发时间:2021-03-05
Abstract:There are a large number of examples about image recognition applications of mobile devices. Such as the FaceID in Iphones, which can greatly improved convenience. However, after a detailed research, it is obviously that most similar Apps run the risk of compromising users\' privacy, which violates the corresponding data protection law to a certain extent. This paper will introduce a new mobile federated learning image recognition algorithm called GNFedHAtt, which uses the lightweight neural network GhostNet and the new federated aggregation mechanism FedHAtt. And design the smart phone handwriting input method recognition experiments to verify, with the datasets of Mnist/EMnist/HWDB1.0, and compared to the traditional FedAvg and other federated aggregation methods. The results prove that GNFedHAtt method can be effectively applied to the mobile image recognition.
keywords: Computer Application Technology Federated Learning Image Recognition Attention Mechanism
点击查看论文中文信息
基于联邦学习的移动端图像识别算法研究
摘要:关于移动设备的图像识别应用,存在大量示例。如iPhone中的FaceID,可以大大提高便利性。但是,经过详细研究,大多数类似的应用都明显有损害用户隐私的风险,这在一定程度上违反了相应的数据保护法。本文将介绍一种称为GNFedHAtt的新的移动联合学习图像识别算法,该算法使用轻量级神经网络GhostNet和新的联合聚合机制FedHAtt。并设计了智能手机手写输入法识别实验,以Mnist/EMnist/HWDB1.0数据集进行验证,并与传统的FedAvg和其他联合聚合方法进行比较。结果证明,GNFedHAtt方法可以有效地应用于移动图像识别。
基金:
引用
No.****
同行评议
勘误表
基于联邦学习的移动端图像识别算法研究
评论
全部评论0/1000