智能物联网中高效安全的自适应量化联邦学习

打开文本图片集

中图分类号：TP309 文献标志码：A 文章编号：1001-3695（2025）08-034-2503-08

doi：10.19734/j.issn.1001-3695.2024.11.0504

Efficient and secure adaptive quantization federated learning in AIoT

Ma Haiyinga，Shen Jinyua，Yang Tianlinga，Qiu Jiana，Wang Zhanjunbt （a.SchooloficellgeeoueclfaStatics，oUesitgJ1 China）

Abstract：Forthe problemofparticipants’privateleakageonthemodelparameters intheexistingadaptivequantizationfederatedlearning schemes，this paper proposedan eficientandsecureadaptivequantization federated learning scheme suitable for artificialinteligenceof things（AIoT）.Thisschemeutilizedadaptivequantization technologytoreducethecommunication overhead forparticipants.Iconstructedascureagregationprotocolintwoaggregationserverstoprotecttheprivacyoflocal modelparametersbycombiningtheDifie-Helmankeyexchangeprotocol，secretsharingshemesandoblivious trasferprotocols.This paper proved the proposed scheme was secureunder reasonable assumptions.The experimental results show that the scheme notonlycanachieveaglobal modelwithhighaccuracy，butalsocansignificantlyreducecommunicationoverheadand computationcosts forprotectingparticipants’privacy.Thisscheme issuitableforresource-constrained，suchaslightweightIT devices in AIoT.

Key Words：federatedlearning（FL）；privacyprotection；adaptivequantization；secret sharing；oblivious transfer protocol

0 引言

智能物联网（AIoT）[]利用人工智能技术分析物联网数据，使得物联网设备具备感知与识别能力，促使产业升级和体验优化，实现万物智能互联。（剩余21318字）