Curveintelxeon

Foundedin1999,YundaExpress*hasbeenridingthewaveofe-commerce,mobileinternetandexpressdeliveryandhasdevelopedrapidlyovermorethanadecadeintoanindustrygiantwithacomprehensiveservicenetworkcoveringtheentiredeliverychain.
Whilstadvancingalongwiththeindustry,Yundaisalsoawareoftherisksandopportunitiesitfacesindependently.
Withaskyrocketingvolumeofbusiness,itstraditionalmodeofmanualoperationisnowtime-consuming,labor-intensive,andunabletoguaranteethedesiredquality.
Atthesametime,thereductioninthedemographicdividendhasalsoledtoanincreaseinthecostofmanpowerandlogistics.
ThesefactorsformabottleneckthatcouldrestrictYunda'sfurtherdevelopment.
Usingautomationandartificialintelligence(AI)tosavecostandenhanceefficiencyofthedeliverylogisticssystemhasthereforebecomeanimportantmeansforYundatotacklethesechallenges.
Tothisend,YundaactivelycollaborateswithITindustrygiantssuchasInteltointroducecutting-edgetechnologieslikeAI,andintegratethemintothetraditionalexpresslogisticsindustrybringing-aboutintelligenttransformationforitselfandtheindustry.
InresponsetoYunda'sneedsandrequirements,IntelnotonlyprovidedaseriesofadvancedhardwareandsoftwareproductsandtechnologiessuchasAnalyticsZoo,aunifiedbigdataanalyticsandAIplatform,aswellasIntelXeonScalableprocessors,butalsocarriedoutanall-leveltechnicalcollaborationwithYunda.
Indoingso,IntelhashelpedYundabuildhigh-efficiencyAIapplicationsforanumberofkeylinksincluding"packagesizemeasurement","datacenteranomalydetection"and"shipmentquantityprediction".
TheseapplicationshavebeendeployedeffectivelyandgainedunanimouspositivefeedbackfromYunda'sfrontlineemployeesandmanagement.
IntelXeonScalableprocessorsAnalyticsZooIntelligentLogisticsIntelArchitectureAIhardwareandsoftwarecoreproductshelpChina'sYundaExpressimprovetheefficiencyoftheirdeliverylogisticssystemIntelligentTransformationBrings"QualitativeChange"toExpressDeliveryIndustry"TakingadvantagesofAItechnologytoimprovetheoperationalefficiencyofdeliverylogisticssystemsisoneofYunda'sstrategicinitiativestodevelopforthefutureandimplementintelligenttransformation.
Itisimpossiblewithoutthesupportofpowerfulalgorithmsandcomputingpower.
TheintroductionofadvancedproductsandtechnologiessuchastheAnalyticsZooplatformandIntelXeonScalableprocessorscanhelpusgreatlyimprovetheefficiencyoftheentireexpressdeliverychain,optimizeresourceutilization,significantlyreduceoperationalcosts,andeffectivelytacklethechallengesforfuturedevelopment.
"PeijiLiChiefArchitectofYundaExpressYundaCo.
,Ltd.
AdvantagesofYunda'sAISolution:Enablequickdevelopmentanddeploymentofend-to-endAIapplicationsandprovidestrongtechnicalsupport;Significantlyimprovetheoperationalefficiencyofdeliverylogisticssystem,shortendeliverytimeandimproveusersatisfaction;Optimizelogisticsresources,includingtheuseoffreightvehicles,sorters,andpackagingmaterials;Greatlyreducelaborandlogisticscosts.
TheboomingoftheInternetande-commercehasmadeexpressdeliveryanindispensablepartofday-to-daylife.
StatisticsfromtheStatePostBureauofthePeople'sRepublicofChinashowthatin2017,China'sannualexpressdeliverybusinessvolumeexceeded40billionpieces,andthedailyprocessingcapacityduringtheperiodof"Double11"onlineshoppingcarnivalhasreachedupto330millionpieces1.
Thefast-growingexpressdeliveryindustryhasalsospawnedindustry-leadingcompanieslikeYunda.
Withaservicenetworkcoveringtheentirenation,comprisingmorethan4,300maintransportationroutes2,Yundaiscommittedtoprovidinghigh-qualityandconvenientdeliveryservicestohundredsofmillionsofusers.
However,successinevitablybreedsnewchallenges.
Astheindustrymatures,especiallywhenthedemographicdividendisgraduallywornaway,expressdelivery,alabor-intensiveindustry,cannolongerrelyonthetraditionalmodelofexpandingmanpowertoachievelineargainswithrisinglaborcostsbecomingaburdentothedevelopmentofbusinesses.
As"SmilingCurve"theory3proposedbyAcerfounderStanShihshows,anindustrycandeveloprapidlyinitsearlystagesevenitisatthebottomofthecurve,whereefficiencyislow.
Asthepressureofcompetitionintensifies,ifacompanyfailstoimprovetechnologiesandimplementleading-edgestrategies,theprofitsofthebusinesswillgraduallyreduceturning,ultimately,intolosses.
Theintroductionofautomated,intelligentequipmenttoimproveefficiencyhasbecometheconsensusoftheentireexpressdeliveryindustry.
Yundaintroducedtechnicalsolutionslikeautomaticsorting,addresssortingandcollectionaswellasvehiclelicenseplaterecognitionintoitsdeliverylogisticssystemsomeyearsago.
Improvementinoperationalefficiencyresulted,however,inYunda'sview,thesesolutionsonlyplayedtheroleof"hands"and"feet";inordertopushformoresignificanttechnologicaladvancement,itwouldbenecessarytoaddressthe"brain".
Thekeytoavisionarystrategyforadeliverylogisticssystemliesinwhetheranenterprisecanbuildend-to-endresourceallocationandoptimizationstrategies.
ThisisexactlywhereAItechnologyexcels.
DrawingonthethreeelementsofAI,namelydata,algorithms,andcomputingpower,thehighdeliveryvolumeofupto470millionitemsayearempowersYundatohavethedatafoundation4forAItechnologyresearch,whilethetechnologicalcollaborationwithIntel,addressesYunda'slimitationsinalgorithmsandcomputingpower.
ForalgorithmandAIsoftwareoptimizations,IntelprovidedYundawithAnalyticsZoo,aunifiedanalyticsandAIplatformbasedonApacheSpark*.
AnalyticsZoohelpsYundatoquicklyandagilelybuildAIapplicationswithanend-to-endviewformultipleapplicationscenariossuchasimagerecognitionandtimeseriesprediction.
Intermsofprovidingthenecessarycomputingpower,Intel'snewgenerationIntelXeonScalableprocessorsofferthepowertofullyunlockAIapplicationspotential.
CaseStudy|IntelligentTransformationBrings"QualitativeChange"toExpressDeliveryIndustryYundaDataCentersYundaLogisticsResourcesYundaDistributionCentersBack-endSupportResourceAllocationFront-endSortingFig.
1.
ThreeImportantLinksinYundaExpressDeliveryLogisticsAsshowninFigure1,thethreemostimportantlinksinYunda'sexpressdeliverylogisticsarefront-endsorting,resourceallocationandback-endsupport.
Duringtheearlystages,IntelandYundaworkedtogethertobuildAIapplicationsfor"sizemeasurement","shipmentquantityprediction"and"datacenteranomalydetection".
TheresultsshowedthatAIapplicationscouldnotonlyhelpYundasignificantlyimprovetheoperationalefficiencyofitsdeliverylogisticssystem,butalsogreatlyreducetheintensityandcostofmanuallabor.
SizeMeasurementMeasuringthevolumeofshipmentsisoneofthecoretasksofYunda'sdistributioncenters.
Bypre-judgingthevolumeofshipments,thedistributioncenterstaffcanbetterplanforsorting,loadinganddistributingtoreducetheaveragetimetakenperdeliveryandthecostperkilogram.
Theconventionalwaytodothismainlyreliesonmanualmethod,whichisnotonlytime-consumingandlaborious,butalsopronetosignificanterror.
Yundaplanstousemachinevisiontechnologytoimprovetheefficiencyandqualityofmeasurement.
Throughhigh-speedphotographicapparatusesequippedonthedistributioncenterconveyorsystem,anAIapplicationcollectstheimageinformationonshipmentsandtransmitstheimageinformationtotheback-endserverforsizemeasurement.
Oncethesizemeasurementiscompleted,theback-endserversendsthedatabacktotheconveyorsystemwhereshipmentsaredeliveredtotheappropriatesortingandloadinglocationsaccordingtothemeasurementresults.
TheimageclassificationtechnologyprovidedbytheAnalyticsZooplatformplaysakeyroleinthiscase.
Usingthebuilt-inimagerecognitionmodeloftheplatform,theAIapplicationfirstextractsthecontouroftheshipmenttobemeasured,andthencarriesouttheentireAIprocessingflowfrommodeltraining,modelredefinitiontomodelinferencetoeventuallyobtainaccuratesizeoftheshipment.
ThewholeprocesstakesadvantageofthedeepleaningframeworkslikeTensorFlowprovidedbyAnalyticsZooandthepotentcomputingpowerofIntelXeonScalableprocessors.
2Following-onfromthis,Yundaisfurtherutilizingdeeplearningalgorithmsforoperationaloptimization.
Forexample,Yundaabstractedtheloadingprocessintoaclassicissueof"boxcutting",thatis,whenshipmentsofdifferentsizesarrive,itisessentialtoplantheorder,orientationandplacementoftheshipmentstomaximizetheshipmentquantityloadedandreducetransportcosts.
WiththehelpofIntel,Yunda'sAIteamadopteddeeplearningtoexpressthevariousstatesofthisissue,addingreinforcementlearningtolearnthisheuristicstrategy.
TheresultsofapplyingAIinthiswayarenotablybetterthanthetraditionaloperationoptimizationmethod.
Tobuildontheinitialresults,theteamsfrombothsideswillworktogethertounleashfurtherpotentialoftheAnalyticsZooplatformandplantoupgradethesizemeasurementofshipmentsto"accuratevolumetricmeasurement"withafurtheroptimizedAIapplication,aimingtofurtherimproveefficiencyandreducecost.
ShipmentQuantityPredictionThevolumeofbusinesswithintheexpressdeliveryindustryinChinaissignificantlyaffectedbye-commercepromotionssuchas"Double11"and"618"(twohugelypopularlarge-scaleonlineshoppingcarnivalsinChina).
Tomeetthechallengeofsurgesinbusinessvolumebroughtaboutbythesepromotionsandtoreduceriskssuchasoverflowingwarehouses,Yundaneedstoassemblelogisticsresourcesinadvance,includingfreightvehicles,sortersandpackagingmaterials,etc.
Forecastsofvolumebasedonpreviousexperienceare,however,notusuallyreliableenough.
Takethe"618"promotionin2018asanexample.
ThesoccerWorldCupinRussiawasinfullswingatthetimemeaningmanyfootballfanswereabsorbedinthegame,payinglittleattentiontoe-commercepromotions.
Asaresult,alargevolumeoflogisticsresourcewaswasted.
Inviewofthis,YundahopestoimplementamoreaccurateshipmentquantitypredictionsolutionusingtheLongShort-TermMemory(LSTM)deeplearningalgorithmprovidedbytheAnalyticsZooplatform.
LSTMisatimerecursiveneuralnetwork.
AsLSTMcaneffectivelydividehistoricalinformationintolong-termmemoryandshort-termmemory,itisidealforthedevelopmentanddeploymentofpredictiveAIapplications.
UsingtheAnalyticsZooplatform,Yunda'sAIteamseamlesslyextendedtheAIapplicationsbasedontheLSTMalgorithmtoitsApacheHadoop*cluster,usingthevastamountofhistoricallogisticsdataformodeltrainingandinference.
IntelXeonScalableprocessorsprovidestrongcomputingpowerforthisprocess.
BasedontheactualneedsofYunda'sAIapplications,IntelandYundaoptimizedthesolutionforbetterperformancewiththeleadingtechnologicalfeaturesofIntelXeonScalableprocessors.
AdvancedtechnologieslikeIntelAdvancedVectorExtensions512(IntelAVX-512)integratedinIntelXeonScalableprocessorsdeliveroutstandingperformanceinparallelcomputingandthesetechnologiescangreatlyacceleratethetrainingandinferenceofapplications.
Currently,AI-basedshipmentquantitypredictionsystemshavebeendeployedinsomeofYunda'sdistributioncenters.
Feedbackfromthefrontlineshowsthatthisdeeplearning-basedsolutionisapproachingtheexpectedgoalinforecasting,evensurpassingtraditionalheuristicforecastingsolutionshelpingYundaachievebetterbusinessgainsandcostreductions.
DataCenterAnomalyDetectionAsthecoreofYunda'sdeliverylogisticssystem,Yunda'sdatacentersshouldertheresponsibilityofanalyzing,storingandtransmittingdatathroughoutthecompany'sentirebusinesschain.
Thedatacentersalsotakeonjobsrelatedtovariousapplicationssuchasdatamodelbuilding,dataextraction,transformationandloading,automatedreportingaswellasalgorithmdevelopment.
Currently,almostallYunda'sbusinessactivitiesarereliantonitsdatacenters.
Withtherapiddevelopmentofitsbusiness,Yunda'sdatacentershavefacedmanychallenges.
Thelongtechnologydevelopmentcycleandhighcostmakethedatacenterslessefficientatperformingadvancedanalysisofbigdatasetsovertime.
ImprovementindatacenterefficiencycanhardlykeepupwiththespeedofbusinessgrowthmeaningYundacanpotentiallystruggleduringpeakbusinessperiodslike"Double11"andSpringFestivalpromotion.
Furtherchallengesthatthecompanyfacesincludehackinganddatacongestion.
YundaisabletoleveragetheLSTMdeeplearningalgorithmprovidedbyAnalyticsZooplatformtoaddressthesechallenges.
TheLSTMalgorithmcanenhancetheperformanceoftheneuralnetworkthroughdifferentiatedmemoryinformation,andcanmoreaccuratelyanalyzeanddiscriminateinformation.
Inthisway,theLSTMalgorithmoffersuniqueadvantagesindataanalysisandprediction.
AnalyticsZoo:an"Analytics+AI"platformbasedonApacheSparkTohelpusersquicklyandefficientlybuildavarietyofAIapplicationsonApacheSparkandsimplifyend-to-enddevelopmentanddeploymentofsolutions,IntelandanumberofpartnerslaunchedAnalyticsZoo,aunifiedAnalyticsandAIplatform(https://github.
com/intel-analytics/analytics-zoo),whichseamlesslyintegratesAIframeworkssuchasTensorFlow*,Keras*,andBigDL*intothesameprocess,andeasilyscalehorizontallyintolargeApacheHadoop*/Sparkclusterenvironmentsforuserstoimplementdistributedtrainingandinference.
3CaseStudy|IntelligentTransformationBrings"QualitativeChange"toExpressDeliveryIndustryCaseStudy|IntelligentTransformationBrings"QualitativeChange"toExpressDeliveryIndustryForexample,indatacongestionforecasting,Yunda'sAIteamdeploysserverswithLSTMalgorithminthestoragesystemoftheirdatacenters.
Usingthealgorithm,thesystemcancontinuouslyreinforcetrainingbyusingkeyinformationintheexistinglogssuchastimeandhardwareaddress,whilefilteringoutalargeamountofirrelevantinformation.
Throughextensivetrainingandinferenceusingthislogdata,thesystemcanaccuratelypredictthepotentialrisksandweaknessesofthedatacenters.
DuringthedeploymentofthisAIapplication,Intelprovidedanumberofbuilt-inlearningmodelsbasedontheLSTMalgorithmthroughtheAnalyticsZooplatformandofferedawealthofreferenceusecasestoYunda'sAIteam.
Intelalsoprovidedateamofexperiencedexpertsthroughremoteassistance,on-siteguidanceandtelephonecommunicationtoassistYundatobuildbusinessmodelsasefficientlyaspossible,achievingtwicetheresultwithhalftheeffort.
OutlookThroughcollaborationonbuildingAIapplicationsfor"sizemeasurement","shipmentquantityprediction"and"datacenteranomalydetection",YundaandIntelhaveestablishedanefficientcommunicationmechanismandaccumulatedexperienceinbuildingbusinessmodels.
ThiscollaborationhasproducedresultsthathavereceivedunanimouscommendationfromYunda'sstafffromfrontlineworkerstomanagement.
Yundaplanstopushlarge-scaledeploymentoftheseAIapplicationsincitiesincludingBeijing,Shanghai,GuangzhouandShenzhenwithinthenexttwoyears.
YundaalsoplanstodevelopanddeploymoreAIapplicationsbasedontheAnalyticsZooplatformandotheradvancedIntelproductsandtechnologies.
Onesuchplanistointroducenaturallanguageprocessingtechnologytobuildabrandnewintelligentcustomerservicesystem.
ThiswillnotonlyhelpYundatoalleviatethecurrentpressureonitscustomerserviceteamandimproveservicequality,butalsoenablecustomerserviceinformationtobedigitizedandbecomeanothervaluabledataasset.
Additionally,YundaplanstointroduceVideoProcessingUnits(VPUs)likeIntelMovidiusTMMyriadTMX.
TheaimhereistoimprovethequalityandefficiencyofitsOpticalCharacterRecognition(OCR)byequippingtheVPUsinhigh-speedphotographicapparatusesandtakingadvantagesofVPUs'powerfuledgeAIprocessingcapability.
Inthefuture,theapplicationofdiversenewtechnologiessuchasrobots,smartwarehousetechnology,unmanneddrivingandintelligentexpressdeliverycabinetswillfurtherdriveYunda'sintelligenttransformation.
ThiswillenableYundatoprovidebetterandmoreconvenientservicestohundredsofmillionsofuserstobecomeamodelenterpriseleadingtechnologicalandbusinessinnovationintheexpresslogisticssector.
Toimprovetheefficiencyofdevelopmentanddeployment,AnalyticsZooprovidesuserswithrichend-to-endprocessingflowanalysisandAIsupport,including:Easy-to-useabstractmodelssuchasprocessingflowsupportforSparkDataFrameandML,conveyancelearningsupport,andPOJOstyleserviceAPIs;Commonfeatureoperationsforimages,texts,and3Dimages;Built-indeeplearningmodelssuchastextcategorization,recommendationandobjectdetection;Referenceusecases,suchastimeseriesanomalydetection,frauddetection,imagesimilaritysearch,etc.
1Datacitedfrom"ChinaExpressDeliveryDevelopmentIndexReportfortheFourthQuarterof2017"fromtheStatePostBureauofthePeople'sRepublicofChina,http://www.
spb.
gov.
cn/sj/zgkdfzzs/201801/t20180112_1467247.
html2,4DatacitedfrominternalstatisticalmaterialsprovidedbyYunda.
3GlobalBrandStrategy—ViewpointofMr.
StanShih.
CITICPublishingHouse.
Inteltechnologies'featuresandbenefitsdependonsystemconfigurationandmayrequireenabledhardware,softwareorserviceactivation.
Performancevariesdependingonsystemconfiguration.
Noproductorcomponentcanbeabsolutelysecure.
Checkwithyoursystemmanufacturerorretailerorlearnmoreatintel.
com.
CostreductionscenariosdescribedareintendedasexamplesofhowagivenIntel-basedproduct,inthespecifiedcircumstancesandconfigurations,mayaffectfuturecostsandprovidecostsavings.
Circumstanceswillvary.
Inteldoesnotguaranteeanycostsorcostreduction.
IntelandXeonaretrademarksofIntelCorporationintheU.
S.
and/orothercountries.
AfulllistofInteltrademarksortrademarkandbrandnamedatabasescanbefoundunderthetrademarksectionatintel.
com.
*Othernamesandbrandsmaybeclaimedasthepropertyofothers.