典型文献
ADC-DL:Communication-Efficient Distributed Learning with Hierarchical Clustering and Adaptive Dataset Condensation
文献摘要：
The rapid growth of modern mobile de-vices leads to a large number of distributed data,which is extremely valuable for learning models.Unfortu-nately,model training by collecting all these origi-nal data to a centralized cloud server is not applica-ble due to data privacy and communication costs con-cerns,hindering artificial intelligence from empow-ering mobile devices.Moreover,these data are not identically and independently distributed(Non-ⅡD)caused by their different context,which will deterio-rate the performance of the model.To address these issues,we propose a novel Distributed Learning al-gorithm based on hierarchical clustering and Adaptive Dataset Condensation,named ADC-DL,which learns a shared model by collecting the synthetic samples generated on each device.To tackle the heterogene-ity of data distribution,we propose an entropy topsis comprehensive tiering model for hierarchical cluster-ing,which distinguishes clients in terms of their data characteristics.Subsequently,synthetic dummy sam-ples are generated based on the hierarchical structure utilizing adaptive dataset condensation.The procedure of dataset condensation can be adjusted adaptively ac-cording to the tier of the client.Extensive experiments demonstrate that the performance of our ADC-DL is more outstanding in prediction accuracy and commu-nication costs compared with existing algorithms.
文献关键词：
中图分类号：
作者姓名：
Zhipeng Gao;Yan Yang;Chen Zhao;Zijia Mo
作者机构：
State Key Laboratory of Networking and Switching Technology,Beijing University of Posts and Telecommunications,Beijing 100876,China
文献出处：
引用格式：
[1]Zhipeng Gao;Yan Yang;Chen Zhao;Zijia Mo-.ADC-DL:Communication-Efficient Distributed Learning with Hierarchical Clustering and Adaptive Dataset Condensation)[J].中国通信（英文版）,2022(12):73-85
A类：
empow,tiering
B类：
ADC,DL,Communication,Efficient,Distributed,Learning,Hierarchical,Clustering,Adaptive,Dataset,Condensation,rapid,growth,modern,mobile,leads,large,number,distributed,which,extremely,valuable,learning,models,Unfortu,nately,training,by,collecting,these,origi,nal,centralized,cloud,server,not,applica,due,privacy,communication,costs,cerns,hindering,artificial,intelligence,from,devices,Moreover,identically,independently,Non,caused,their,different,context,will,deterio,performance,To,address,issues,we,propose,novel,hierarchical,clustering,named,learns,shared,synthetic,samples,generated,each,tackle,heterogene,ity,distribution,entropy,topsis,comprehensive,distinguishes,clients,terms,characteristics,Subsequently,dummy,structure,utilizing,dataset,condensation,procedure,can,adjusted,adaptively,cording,Extensive,experiments,demonstrate,that,our,more,outstanding,prediction,accuracy,compared,existing,algorithms
AB值：
0.566249
相似文献
机标中图分类号，由域田数据科技根据网络公开资料自动分析生成，仅供学习研究参考。
