《面向边缘智能的联邦学习综述-2023.07-20页-WN7.pdf》由会员分享，可在线阅读，更多相关《面向边缘智能的联邦学习综述-2023.07-20页-WN7.pdf（20页珍藏版）》请在taowenge.com淘文阁网|工程机械CAD图纸|机械工程制图|CAD装配图下载|SolidWorks_CaTia_CAD_UG_PROE_设计图分享下载上搜索。
1、Unv1,20Z_10Z3011oo|b0W01000932ozy0W01000493OAn Overview of Federated Learning in Edge IntelligenceZhangXueqing1,2,LiuYanwei1,LiuJinxia3,andHanYanni11Institute of Information Engineering,Chinese Academy of Sciences,Beijing 1000932School of Cyber Security,University of Chinese Academy of Sciences,Beij
2、ing 1000493Zhejiang Wanli University,Ningbo,Zhejiang 315100Abstract0Withtheincreasingdemandofedgeintelligence,federatedlearning(FL)hasbeennowofgreatconcerntotheindustry.Comparedwiththetraditionallycentralizedmachinelearningthatismostlybasedoncloudcomputing,FLcollaborativelytrainstheneuralnetworkmode
3、loveralargenumberofedgedevicesinadistributedway,withoutsendingalargeamountoflocaldatatothecloudforprocessing,whichmakesthecompute-extensivelearningtaskssunktotheedgeofthenetworkclosedtotheuser.Consequently,theusers9datacanbetrainedlocallytomeettheneedsoflowlatencyandprivacyprotection.Inmobileedgenet
4、works,duetothelimitedcommunicationresourcesandcomputingresources,theperformanceofFLissubjecttotheintegratedconstraintoftheavailablecomputationandcommunicationresourcesduringwirelessnetworking,andalsodataqualityinmobiledevice.Aimingfortheapplicationsofedgeintelligence,thetoughchallengesforseekinghigh
5、efficiencyFLareanalyzedhere.Next,theresearchprogressesinclientselection,modeltrainingandmodelupdatinginFLaresummarized.Specifically,thetypical work in data unloading,model segmentation,model compression,model aggregation,gradient descentalgorithmoptimizationandwirelessresourceoptimizationarecomprehe
6、nsivelyanalyzed.Finally,thefutureresearchtrendsofFLinedgeintelligenceareprospected.Key words0federatedlearningedgecomputingedgeintelligencemodelaggregationresourceconstraints|00wnnfederatedlearningFLors.No_nOn_qnkOowuorvuovNwOwn,wuoowuo.Ounga1_uov.oUYVOmnszwnw1kNkso|gn.w2021-11-082022-09-16w61771469
7、oNobb_HZ2021015This work was supported by the National Natural Science Foundation of China(61771469)and the Cooperation Project Between ChongqingMunicipalUndergraduateUniversitiesandInstitutesAffiliatedtoCAS(HZ2021015)._Z_|N DOI10.7544/issn1000-1239.202111100JournalofComputerResearchandDevelopment60
8、61276212952023s0nkoV0TP3wonr_ucg1.no,21314N5.uovn.onuoov.uuo1VguonbouorvoOW.g5G o_umobileedgecomputingMEC61NrOruovNwO_wOn.MEC oNo_oUV._oo_UMEC Om1oov1r_u_g.MEC V_o.OonoVTuorw1y.bouoNnuokO_.FLfederatedlearning,FL7-8o.uOqouoFL k1V_r.FL V_nkMEC OV_NnwNoowuokuNouo1uONnw.1 UO FL y.ugWwaa1Wowuo_1_v_uoO FL
9、 wsz.YoFL _nowuooooOuoorNnoOow_uonqkkOow9.Osz MEC gFL_.uoOrooOuko_Wk_o FL ON.w k FL W O stochasticgradientdescentSGD10gkNn_NouoOwuckkw_s_roo FL oNns.kOuN_uo11No_guoqkoNnwoszg.sz|u|NON|uV.|Vn 1 bFL Nu.No FL _szY FL ovww1k_Wo1kono FL soogO FL|.10FL ovjwjf(w,xj,yj),xjyjjxjyjFL oNV_ng_uoo.FL okuNu.nouok
10、.nwkonuoo uUNOyu.yuwokknouooNyug_W.o yuOwoUo 1nuoo 2 nrV.nkuok.D1,D2,Di,DKDkWo K nowuoV.n K uowyuvUn1277Fk(w)=1|Dk|jDkfj(w)，（1）fj(w)f(w,xj,yj)woooboV_uoNOyuF(w)=jUk|Dk|fj(w)?kk=1?Dk|=Kk=1|Dk|Fk(w)Kk=1|Dk|，（2）|Dk|kDkD=Kk=1Dk|D|=Kk=1|Dk|F(w)wow Su.krg_WyuOrgung_Wyuw=argminF(w).（3）uuoV_OwFL Oo_nNooNr.O
11、gg_Wyyusw(t):=w(t1)F(w(t1)=Kk=1|Dk|wk(t)Kk=1|Dk|，（4）F()w(t)twk(t)ktwo oNn_kusn.oyu.ooo yuow owuwk(t)=w(t1)Fk(w(t1).（5）Fk(w(t1)w(t)Ooo.oowrTable 10Comparison of Studies on Existing FederatedLearning Reviews 1 on|_Wv_Wn 11:FL 12:13:14 FL 15:o:=or_=oor_.&ykykoowkNoowkowkowkowkowkwwwwwD2DD2DD2D(_)(O_)(_
12、)(O_)(_)UAV_V2VFig.10Edgeintelligentfederatedlearningarchitecture10n1278|N0202360 6oowo_Nwouo.20FL woo_Oouo.N_oo_ON FL.o_uooV_m.NNn_kaoNoroTa._gOv_16k.ugvOFL N|12.kc_s.N|_W_gNnw.Jin 17_w_OkwNnn_qvN.OoO FedAvg7O_VwNk.Chai 18owgwRVOoNowuuo_gk_.Chai 18 TiFL oNo FL.Nnnoky_ 1 nvnwvwouowwk.uuobowr 1 ko_W_
13、o_o.19NFL VvN_gk.20 N.Y_sN_rNnwo_gNo FL ouo FL.nul1Nw_gsug.FL ongNo_w_wOV.|Vog_W FLOsoWkn.Xu 21Nng FL ornONgnowOoOOwV_W_ogg.owwwcb_o22-23o24nv.w_ 2 b.Nishio 25Nn FL w_s FedCS.FedCSw FL _oO1NonkNnouowunmb.FedCS FL NO_OuuoV_g.nYoshida26FedCS krvNOuoV_FLhybridfederatedlearningHybrid-FL_.Hybrid-FL _oouo
14、nNovQNouo.Nw_wNouoovouoorNnV_uo.nN FedCS so1%NquoHybrid-FL V_o.wHybrid-FL NowuoV_o_wyoNovo_uoy.wwouoo_y.OO_OoNouoow_.oNuow_vN FL Z27.Kang 28_k.FL gn_woowuowY_N.n_kNwVuVugw_VurrNno.wVuuOW_Oqw_._Novo_owovUn1279ww_y.V_VodistributedledgertechnologyDLT_o._cowkaon_kaokaowowV_Vo.owV_V DLT Nw_Vo_Wovo1oe_g29
15、.ruoV_ FL gLi 30Nnggws q-FedAvg.q-FedAvgg_Vg_gg FedAvg uyuogV_NOo_gV_.31Nn_WNm_oco FL_uN Stackelberg_kv|_wN FL.wO|wooO1cO.WwooouuoWnyoO.NoOc.NN_N_|wogoOk15.o_owuooV_nm.Nw_wovVo.32 Nno FL k.rNvwgg FLg_Wro.OowOw_Wo CPU|FL ng_W.O 32 n 33nwuo_1uoV_|oo FLNRL WvuoboN FL.NNr Stackelberg _YvOWvQwwTo34.w_VoN
16、osvgsNowo FL c.o_WovvyuoV_sowgg.wkuoNnwouSugkWWgwNWN FL Ou_go FL _W.2_o FL w.d _We nf wg Vd Noe kkykVonACKACKFig.20FedCSprotocoloverview20FedCS _1280|N0202360 6Table 20Comparison of Federated Learning Client Selection Schemes 2 nwhhhwoooN_WVOk171wV181vn_191n201OV211n251_26vYv28_Nn_gWuyNwO_bStackelbe
17、rg _31myk_srosroOuWv32-33gwOworo_rog_Worooug30_ugO q-FedAvg oVggwoyow_ 30k_Wuv1v1wnOgV_owuo FL k.|kosuovuorw_vowuouoogO.uows_kv FL okukO.3.10uouvn_quo.uo|35oNquoouoonko._uo|o 3 1|.uooomwVuo.o_r|ozunuoNkss.Oo|uo.2.uofrOzuogfzoOnmwouom.3uo.N_ 2 uoozNn_uooo FL o_uo_uov.uo|_FL_OwuooV_Nwwqowuo|m.uo_rooo
18、FL 36.uou_or.FL nNvgoOuuo.rnuoWo_Os 37 _WuoWom_gngOo_WW_ FL ouoN.NnuovguooOo_gnO.38|Nnnkuo_WgnmOg_o.Oo 38 ouNnn_r._.FL oobo_o_wk.wqooOw_.qoo_WNnVON.O_.3.20kVrOuokkws._kWnVrrV_n_NnskVr.kVroN._oORVkvUn1281mV_NgVroN39-41.Yowuuocwcuoo.OokVr_wvuoovOr.okVr_WkgORVk_VrvoN.wokNO_kVruzo.42 _oONrouOnroNoO.wug1
19、goOo_aWugg_uoVn 43 kugw_oNOoawZkVr.kVrVkkoOuoo 44 _uooOVrk ARDEN g.ARDEN _NgWuocwruoo_uNvOkVNg.onwqNnkg FLk.Zhang 45kVroVN FL federatedlearningschemeinmobileedgecomputing,FedMECoNykVr FL w_n 3 b.FedMEC NnV 2 VYwkokgko._ouorvkku.ouowkryOowkykyokrwFig.30Modelsegmentationmigrationframework30kVrnokVroOm
20、 FL oO_Ns_V 2 VwoV_ro.o_VkNoroO_Vvguo._kVrowZrkOowVvkkVrouowgky.w 39245 o|owgN.3.30kwwSoWOwruluzvk FL rNn.o 2 n1NnuWSroNnkw_s.2kO_m.o FL _oNomonkr FL o46.k_k_1Ov.o_kmogNk_rk47.gV_ok1282|N0202360 6g_Wkaoom.3 oko.48WoNor FL okuo.W_vNYgsOy.OogkvNnO 2 n.NnorNocONn_W.o_Wro._ogNoO_uoyoO_uoy.48 xN 49 k|2 u
21、gorw1orwykNo2Qwykg_Nwroow.unwNwN FL goO.50 46249 Nog|OnUoO.50 omu.48250|koNOroOroOwkzr.o_Wo|wo_OvrOu FL NO|.FL NckugOro.|gkocJeong 51bcOokuokQ_kk.xNV_uo 51 Nruos.OO_o_.ovWxN.Nuogrsz.Ahn 52ONVyhybrid-federateddistillation,HFD.yoozv.u_kHFD vo_o FLNvNkWoNqOOHFD uoo FL uoy_|.4 _ FL k_W.Table 30Summary o
22、f Model Compression Techniques 3 ko_Wk_Wv48okWwromwrouogkwo-w49okWorwmorwuogOow50ukwgggwkwNzwWgmgWrAdam1 Adam _Wh FedAvg Adam _WwhO_k51-52ckkaoswom_ouwguoV_okTable 40Optimization Methods and Characteristics ofModel Training 4 k_W_Wguook 36238kVrVrko 42244kOk|vo|48252 40Oko FL ok FL wowwUoNokusy.kokk
23、y._wvUn1283Wy1kgOgs.kon_W_o.ok|_ 3 n_W1yownOg_oNOg_Wkyu2_WOgO_3vawuVNoO.o_u FL okog|_m FL oWg_Wog FL yk.4.10ko_ FL OngOukug_WowuoNOyu.wOkuruuo1SvNQ.uwruyuON7.yoNnvnowau53.7 FedAvg kwNow SGD NgkWokku.k_ykunW_vykwg54_wg55-56.55 FedAvg Nn_W FedProx.FedProx NnkowkykO_g.OoFedProx okuykO._57-58go|on_vvo.o
24、_u_Og1wny1voNk59-62.wuoO FL kb.FL oszWang 63NkvVwoV_uoFL w._aOvykkONvog_Wwoyu.kNoNO_ok_.4 oyOg_WnykoawuyOg_Wny.63|nyug_WW_OoOgOo.w 63 _oogOo_v.o_OOouoWny1_w_ FL Oo.64 wa_Womw ACFLACFL ogwq_1(a)owkkNovO(b)k_2_3_4&_NFig.40Comparisonofadaptivemodelaggregationandfixedfrequencyaggregation40kN1284|N020236
25、0 6o.o_ouoV_WOonyVwg.FL k_Liu65No_oNw44Vnhierarchicalfederatedaveraging HierFAVG oog 2 wN FL ks.No FL u_uoHierFAVG omw.HierFAVG oN FLNomk_.7,53265|_gO_WkOuO_g._g_WOnV1nk1n1g_Wyg_WwooO.66 Nn_NmgV_n_Wk_V_nVn.wo_r_wuV_nV.67 rO_ngWOVO CuFL g FL bo_wn.No_owkNn_kokg.ko_n_ FL.woOro_o.MEC o_Wy.MEC oNu_uNnug
26、g_W_O_.vNoV CuFL wg CuFL Voomg.nCuFL 1gwo_.5 W FL ko.Table 50A Comparative Summary of Major Federated Learning Mode Aggregation Technologies 5 nkoo_WFedAvg7_gwwowuognvNoogOo.FedAvg n_uoV_ONO_gFedAvg OQNn_yowvguow.FedProx55_gnwNowoNvnowkyk.FedProx o_guowg.N FedAvg FedProx obo_ygo_|1u_.FedPAQ53NoqO_Qw
27、kNgnow.N FedAvg FedPAQ oykkW|Oogg.FedMA54_gg_SWOgQyk_.uvyouoV_guok_FedMA ouvwrkoOkwwkyk_oowk.Turbo-Aggregate62yNuwowVrnk_OqNwuoqv.Turbo-Aggregate ssogwg_W.Turbo-Aggregate ouyvwomvwmyOuw.osyg_Ns_gk.63_gOyuOgov.ykg FLoomo FL.hwg_yu.HierFAVG65NVw44_owwkwroy._ow4Nmkc.HierFAVG ww_.V66_g11_NuoV/ku_NowOgWn
28、.VhhgWV_nownkunv.hauWauor.67_g1_g11NwonvnOnow.oNnyug_W_O_.Nn CuFL g_Wn.r_gCuFL Q_wn_.hauWauor.vUn1285 4.20O_WoukkuO.FL owuONo_uoO_68.konowuonkZ.o|NO.NOoNk_orO_w69.uwgo|_OV_n.Liu69NgNks FL O.FLo|o1sWO.so_kuvUfruouO/vzgOo.W_70-72ov_73-76.Nn_b|SGD Ng_oV_OdecentralizedstochasticgradientdescentDSGD77-78v
29、Ouo_g79rNok.o_ 80382 ozoover-the-aircomputationAirComp|oNo_ggock FL oo.Nu_vkWok AirComp woo.80|o W multipleaccesschannelMACN FLu_k SGD.u_ SGD o_W MAC oruok SGD _vgzo.NWkOuoV_noom_.uW_ukOnw.oW71,83wWwg.Shokri 84V_gOdistributedselectivestochasticgradientdescentDSSGDoOuvOwOgwu.DSSGD rNo SGD _Oknou 12 n
30、u.85 Q-GADMMWNVUgroup-basedalternatingdirectionmethodofmultipliersGADMM86zzWwog p 12p VUNUOu p wug_Wro GADMM w.L-FGADMM87 GADMM VO Q-GADMM oWL-FGADMMoVNm.N GADMM L-FGADMM 2 _Nom.YN GADMM okOL-FGADMM o.wkNVu L-FGADMM|Ncunk GADMMOL-FGADMM k|_om_.ro_n_krboo_uluouo.oo|_o FL o.w_og|nwouwgn_kzoNn_NoykrNnz
31、o_k FL kw88.r 22 SGD s ADASGD.No_NnnVwkoowkjkOO.22,88 woN_OooroOsz.oTao 89OedgestochasticgradientdescenteSGDOouuoo_.|ooOW.o_uuowunosW.oeSGD N_VNoN FLo.N SGD eSGDwy.Tao 89|gWang90 CMFL.1286|N0202360 6NoskOroyw.koYNowNyovs.Os1_CMFL.91|DSGD.u_ko D2D N DSGD _k AirComp.rwuru_ko.O FL N oWs 2 3|_oug.FL u.V
32、_oo FL _WNnU.4.30_WogO FL oa1Owog.FL wowow_ FL ounmannedaerialvehicle,UAV4-5,92-99182,100-104O104.1FL o|.6 O FL w1ouo1FL._szoomgomwno.woVooO.FL omoywocowkwuo.wuomS105.Table 60Unmanned Aerial Vehicle Application Based on Federated Learning in Edge Network 6 Onsznwouoowyk95-96UAVsoo1O1s1nkoomwno_93wws
33、wwgoO1U1wwk_go_W1g1om.R92UAVsov1O1go1U1O111v1on.ggog_1_Wog1_Wuwogg1ugN_zo93omwUow_r_w11gm.No_m1.Now_o_u1_Nwwg106.oOod1ywOow|ouoggo.uwkvwoszgrw.FL v92.2FL o|.5 O FL.100|oOVFL faWV_.101 o FL V.w_oTogw_ugkgk.romOno FL VgsN_ FLuo.107 oNVVoO.oo FL onoykro.wooboryko.on_ro_cz.N|_onvnm.108vUn1287oN_o_nvgkW_
34、W.FL m 109 N_o_o FL quognkrrWg 2.4N FL rov Dempster-Shafer vw FL nm.3FL O|.mqooWwO.FL oNnOgwOy_.110 o FL gnkowWowW._owwn_wsuOowuoowkuwroo.vo_oowruryu.2 nkWOuoowogwyuk_wog_yuk.4FL On.Ouo_ FL g.oookNOoguV_W FL.FL kWw_w_ CPU owuo_Wbow_w_uo_.uNuluo_Vw_og_Wos.s_g_W FL kw_ 2 nOoMerluzzi 111UnNVauu1onOgg.b
35、g FL.wYang112owooNOgn.NorV1V1v1n_Nnrg_go_W.UOywoLuo 113NVnvNnu_1VowkowkowkkkkOkkykykykgfdddddeeeeddffgguoFig.50Intelligenttransportation501288|N0202360 6skynrog_WOm FL owo.Abad 114No_onzWN_WoVgg_W.1113114 u_gV 1153118 ww_V.115 vwoauokgV_.116 Oo_ CPU g FL m.117 O_mVu_Ygu_awoa_.118 NwowkykwowVNW.11931
36、20 sV.119 yVkuV_W.wVkWVW 2 _UoNonuo_.2 _O_WyNouo_oowo CPU og_Wbo_.120 uo_Wg FL.1113120|_OxNgVkg_WmvvN_aw_.1213122 _uWnoNag_WkV.Anh 121Nno_azW_ CPU a_zW_uoWSWSu_W._Ouo1uw Q g_W._ 121 k 122 NWnVFL Ng.oYOFL uo_WNOV.w 123 _ D2D FL g D2D-FedAvg.a_ D2D nO FL._O_w.ng|_U FL mo73,80,114.Zhu 73|MACN FLwo_yaoc
37、hannelstateinformationCSIOguoN FL WkbroadbandanalogaggregationBAA FL oOooyVgBAA zoWggr.BAA QunokVWokWV.73|nBAA rNkVW_oO 101000.Amiri74NokzoxN_uzWomwOzoo.N 74 124 AirComp o_kONN_ugWngycN|O.6 v.Pkwk(i)Plwl(i)hlhkmPkwk(i)+nkxi=hkPkwk(i)+mTnkyi=mTxi =mThk&Fig.60The parameters are aggregated by air calcu
38、lation andspatialfreedom12560zozuu125iPkw(i)k,k=1,2,Kxi=khkPkw(i)k+nhknku 6 owwowu nSSwooowUowo Voww UU.OvUn1289myi=mTxi=mTkhkPkw(i)k+mTnmPkmThkkPk=|Dk|l|Dl|rUrrmou.nowwovgvo_4.Nuynowwgov|w_wo_ug.124 NnzOWgn.1243126oVyuoroN_uo_ CSI O.127|N FL _V_N_u128.1293131 TVON FL wgV.132oWo_ru_o_u_Vgkro.u|n_ogo
39、wk_Oo 2 no_NnW.N_Wo FL mw_N_oO_.OVNno FL ko_NbouoV_b.oU FLVuoV_gNo_W.50N_ FL|FL w1kNko_n.OoUogkWFL oz.1FL._ FLV84vy133ogkou.o_y.ogo_1_1_uoyw_.O_O1OoOO.Oo FL oNnU.2FL N39,41Wnw.FL _kNnonoO_w.FL _Wo_Wg._V_noNN.ogU 6G FL owNoog AirComp Nzo|Nngv.3FL n1WnoWn.nNWno.oFL TnNououoN_.ouowomnkOwuoFL Nnrn.FLV_W
40、nkrWnzo44WnOW.4FL omNv.FL _u|WgOunNo.O_uooNoNn.o_uoouowN FL ooguNnU._nvu_|1uvZ_|1u_1ynxgZ1N1go.000Mills J,Hu Jia,Min Geyong.Communication-efficient federatedlearningforwirelessedgeintelligenceinIoTJ.IEEEInternetofThingsJournal,2019,77:5986259941CovingtonP,AdamsJ,SarginE.DeepneuralnetworksforYouTuber
41、ecommendationsC/Procofthe10thACMConfonRecommenderSystems.NewYork:ACM,2016:19121982ParkhiOM,VedaldiA,ZissermanA.DeepfacerecognitionC/Proc of the 15th IEEE Int Conf on Computer Vision Workshop.31290|N0202360 6Piscataway,NJ:IEEE,2015:2582266Mowla N I,Tran N H,Doh I,et al.Federated learning-basedcogniti
42、vedetectionofjammingattackinflyingad-hocnetworkJ.IEEEAccess,2020,8:4338243504Brik B,Ksentini A,Bouaziz M.Federated learning for UAVs-enabled wireless networks:Use cases,challenges,and openproblemsJ.IEEEAccess,2020,8:538412538495AbbasN,ZhangYan,TaherkordiA,etal.Mobileedgecomputing:AsurveyJ.IEEEIntern
43、etofThingsJournal,2017,51:45024656McmahanB,MooreE,RamageD,etal.Communication-efficientlearningofdeepnetworksfromdecentralizeddataC/Procofthe20th Int Conf on Artificial Intelligence and Statistics.New York:PMLR,2017:127321282.7Yang Qiang,Liu Yang,Chen Tianjian,et al.Federated machinelearning:Concept
44、and applicationsJ.ACM Transactions onIntelligentSystemsandTechnology,2019,102:12198ZhouZhi,YangSong,PuLingjun,etal.CEFL:Onlineadmissioncontrol,data scheduling,and accuracy tuning for cost-efficientfederated learning across edge nodesJ.IEEE Internet of ThingsJournal,2020,710:9341293569RuderS.Anovervi
45、ewofgradientdescentoptimizationalgorithmsJ.arXivpreprint,arXiv:1609.04747,201610LimWYB,LuongNC,HoangDT,etal.Federatedlearninginmobile edge networks:A comprehensive surveyJ.IEEECommunicationsSurveys&Tutorials,2020,223:20312206311Li Tian,Sahu A K,Talwalkar A,et al.Federated learning:Challenges,methods
46、,and future directionsJ.IEEE SignalProcessingMagazine,2020,373:5026012LiQinbin,Wen Zeyi,Wu Zhaomin,etal.A survey on federatedlearning systems:Vision,hype and reality for data privacy andprotectionJ.arXivpreprint,arXiv:1907.09693,201913Wang Xiaofei,Han Yiwen,Wang Chenyang,et al.In-edge AI:Intelligent
47、izingmobileedgecomputing,cachingandcommunicationbyfederatedlearningJ.IEEENetwork,2019,335:156216514Kairouz P,Mcmahan H B,Avent B,et al.Advances and openproblemsinfederatedlearningJ.arXivpreprint,arXiv:1912.04977,201915Wang Yan,Li Nianshuang,Wang Xiling,et al.Coding-basedperformanceimprovementofdistr
48、ibutedmachinelearninginlarge-scaleclustersJ.JournalofComputerResearchandDevelopment,2020,573:5422561inChineseo.yokV_ngJ.|N2020573542256116Jin Yibo,Jiao Lei,Qian Zhuzhong,et al.Resource-efficient andconvergence-preserving online participant selection in federatedlearningC/Proc of the 40th IEEE Int Co
49、nf on DistributedComputingSystems(ICDCS).Piscataway,NJ:IEEE,2020:606261617ChaiZ,AliA,ZawadS,etal.TiFL:Atier-basedfederatedlearningsystemC/Procofthe29thIntSymponHigh-PerformanceParallelandDistributedComputing.NewYork:ACM,2020:125213618LiLi,XiongHaoyi,GuoZhishan,etal.SmartPC:Hierarchicalpacecontrolinr
50、eal-timefederatedlearningsystemC/Procofthe40th19IEEE Real-Time Systems Symp(RTSS).Piscataway,NJ:IEEE,2019:4062418KhanLU,AlsenwiM,HanZhu,etal.Selforganizingfederatedlearning over wireless networks:A socially aware clusteringapproachC/Procofthe34thIntConfonInformationNetworking(ICOIN).Piscataway,NJ:IE