Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Ericsson Technology Review: Cognitive technologies in network and business automation

594 views

Published on

Creating the highly automated environment that network operators and digital service providers will need in the near future requires the support of intelligent agents that are able to work collaboratively. At Ericsson, we believe that the most effective way to create such intelligent agents is by combining machine reasoning and machine learning techniques.

This Ericsson Technology Review article explains the role that these two cognitive technologies play in the creation of intelligent agents that have a detailed semantic understanding of the world and their own individual contexts. It also includes two proofs of concept that help demonstrate how the combination of machine reasoning and machine learning techniques makes it possible to create intelligent agents that are able to learn from diverse inputs, and share or transfer experience between contexts.

Published in: Technology
  • Be the first to comment

  • Be the first to like this

Ericsson Technology Review: Cognitive technologies in network and business automation

  1. 1. ERICSSON TECHNOLOGY COGNITIVE TECHNOLOGIES ANDAUTOMATION C H A R T I N G T H E F U T U R E O F I N N O V A T I O N | # 6 ∙ 2 0 1 8 Induced models Inferred knowledge Knowledge transfer Knowledge extraction Training examples Expert knowledge Predictions Features Actions Reasoning Planning Actions Machine learning (Numeric) Machine reasoning (Symbolic) Induced models Inferred knowledge Knowledge transfer Knowledge extraction Training examples Expert knowledge Predictions Features Actions Reasoning Planning Actions Machine learning (Numeric) Machine reasoning (Symbolic)
  2. 2. ✱ COGNITIVE TECHNOLOGIES 2 ERICSSON TECHNOLOGY REVIEW ✱ JUNE 28, 2018 JÖRG NIEMÖLLER, LEONID MOKRUSHIN The need to support emerging technologies will soon lead to radical changes in the operations of both network operators and digital service providers, as their businesses tend to be based on a complex system of interdependent, manually-executed processes. These processes span across technical functions such as network operation and product development, support functions such as customer care, and business-level functions such as marketing, product strategy planning and billing. Manually-executed processes represent a major challenge because they do not scale sufficiently at a competitive cost. ■Automationisanessentialpartofthesolution. AtEricsson,weenvisionanewinfrastructurefor networkoperatorsanddigitalserviceprovidersin whichintelligentagentsoperateautonomouslywith minimalhumaninvolvement,collaboratingtoreach theiroverallgoals.Theseagentsbasetheirdecisions onevidenceindataandtheknowledgeofdomain experts,andtheyareabletoutilizeknowledgefrom variousdomainsanddynamicallyadapttochanged contexts. Cognitivetechnologies Softwarethatisabletooperateautonomouslyand makesmartdecisionsinacomplexenvironmentis referredtoasanintelligentagent(apractical Forward-looking network operators and digital service providers require an automated network and business environment that can support them in the transition to a new market reality characterized by 5G, the Internet of Things, virtual network functions and software-defined networks. The combination of machine learning and machine reasoning techniques makes it possible to build cognitive applications with the ability to utilize insights across domain borders and dynamically adapt to changing goals and contexts. Cognitive IN NETWORK AND BUSINESS AUTOMATION technologies
  3. 3. COGNITIVE TECHNOLOGIES ✱ JUNE 28, 2018 ✱ ERICSSON TECHNOLOGY REVIEW 3 Figure 1: The model of mind Sensing Thinking Acting Knowing Known facts Previous experience implementationofartificialintelligenceand machinelearning).Itperceivesitsenvironmentand takesactionstomaximizeitssuccessinachievingits goals.Thetermcognitivetechnologiesreferstoa diversesetoftechniques,toolsandplatformsthat enabletheimplementationofintelligentagents. ThemodelofmindshowninFigure1illustrates themaintasksofanintelligentagent,andthusthe mainconcernsofcognitivetechnologies.Themodel describestheprocessofderivinganactionor decisionfrominputandknowledge. Anintelligentagentneedsamodelofthe environmentinwhichitoperates.Technologiesused tocaptureinformationabouttheenvironmentare diverseanduse-casedependent.Forexample, naturallanguageprocessingenablesinteraction withhumanusers;networkprobesandsensors delivermeasuredtechnicalfacts;andananalytics systemprocessesdatatoproviderelevantinsights. Thepurposeofintelligentagentsistoperform Terms and abbreviations CPI – Customer Product Information | eTOM – Enhanced Telecom Operations Map | SID – Shared Information/Data | SLI – Service Level Index | TOVE – Toronto Virtual Enterprise
  4. 4. ✱ COGNITIVE TECHNOLOGIES 4 ERICSSON TECHNOLOGY REVIEW ✱ JUNE 28, 2018 actionsandcommunicatesolutions.Acting complementssensingininteractionwiththe environment.Thechoiceoftechniquesandtools isequallydiverseanduse-casedependent. Forexample,speechsynthesisenablesconvenient communicationwithusers,roboticsinvolves mechanicalactuation,andanintelligentnetwork managercanactbyexecutingcommandsonthe equipmentorchangingconfigurationparameters. Thethinkingphaseinthemodelofmindisthe sourceoftheintelligenceinanintelligentagent. Thinkingcanbeimplemented,forexample,asa logicprograminProlog,inanartificialneural network,orinanyothertypeofinferenceengine, includingmachine-learnedmodels. Thethinkingphasederivesitsdecisionsfrom factsandpreviousexperiencesstoredina knowledgebase.Thekeyisamachine-readable knowledgerepresentationintheformofamodel. Graphdatabasesandtriplestoresarefrequently usedforefficientstorage.Formalknowledge definitioncanbeachievedusingconceptsofRDF (theResourceDescriptionFramework)or descriptionlanguages,suchasUML(theUniversal MarkupLanguage)orOWL(theWebOntology Language). Machinelearningandmachinereasoning Therearetwotechnologicalpillarsonwhichan intelligentagentcanbebased:machinelearningand Figure 2: Machine reasoning and machine learning [1] Induced models Inferred knowledge Knowledge transfer Knowledge extraction Training examples Expert knowledge Predictions Features Actions Reasoning Planning Actions Machine learning (Numeric) Machine reasoning (Symbolic)
  5. 5. COGNITIVE TECHNOLOGIES ✱ JUNE 28, 2018 ✱ ERICSSON TECHNOLOGY REVIEW 5 machinereasoning(illustratedinFigure2).Both involvemakingpredictionsandplanningactions towardagoal.Eachhasitsownstrengthsand weaknesses. Machinelearningreliesonstatisticalmethods tonumericallycalculateanoptimizedmodelbased onthetrainingdataprovided.Thisisdrivenby wantedcharacteristicsofthemodel,suchaslow averageerrorortherateoffalsepositiveornegative predictions.Applyingthelearnednumericalmodel tonewdataleadstopredictionsoraction recommendationsthatarestatisticallyclosest tothetrainingexamples. AnexampleofalearnedmodelistheService LevelIndex(SLI)[2][3]implementedinEricsson ExpertAnalytics,whichpredictsauser’slevelof satisfaction.Thetraininginputismeasurements fromnetworkprobesthatshowtheQoSdeliveredto theusercombinedwithsurveysinwhichusersstate theirlevelofsatisfaction.Thelearnedmodelpredicts thissatisfactionlevelfromnewQoSmeasurements. Machinereasoninggeneratesconclusionsfrom symbolicknowledgerepresentation.Widelyused techniquesarelogicalinductionanddeduction. Itreliesonaformaldescriptionofconceptsina model,oftenorganizedasanontology.Knowledge abouttheenvironmentisassertedwithinthemodel byconnectingabstractconceptsandterminologyto objectsrepresentingtheentitiestobeusedand managed.Forexample,“customersatisfaction,” “user”and“quantifies”areabstractconcepts.Based onthese,wecanassertthat“Adam”isauserand“4” istheSLIvaluerepresentinghissatisfaction.Wecan furtherassertinferencerules:“SLIquantifies satisfaction,”“SLIbelow5islow,”“lowsatisfaction causeschurn”.Basedonthisknowledge,amachine- reasoningprocesswouldlogicallyconcludethat Adamisabouttochurn.Itwouldtracethereasonto thelowSLIvalue. Hybridapproachestosymbolicneuralnetworks alsoexist.Thesearedeepneuralnetworkswitha numericandstatistics-basedcoreandanimplicit mappingofthemodel’snumericvariablestoa symbolicrepresentation. Designingintelligentagents Autonomousintelligentagentssupporthuman domainexpertsbyfullytakingovertheexecutionof operationaltasks.Doingthisconvincinglyrequires themtoreactandexecutefasterthanhumansandbe abletoovercomeunexpectedsituations,while makingfewererrorsandscalingtoahighnumberof managedassetsandtasks. Intelligentagentsaredevelopedanddeployedina softwarelifecycle.Assuch,theyprofitfromthe encapsulationprovidedbyamicroservice architecture,comprehensiveandperformantdata routingandmanagement,andadynamically scalableexecutionenvironment.Theabilitytocreate anoptimalthinkingcoreforanintelligentagent requiresagoodunderstandingofthefundamental characteristicsofmachinelearningandmachine reasoning. Theroleofabstraction Apersonusesabstractiontodistillessential informationfromtheinputpresented.Abstraction providesfocusandeasier-to-graspconceptsasa baseforreasoninganddecisions.Italsofacilitates communication. AUTONOMOUSINTELLIGENT AGENTSSUPPORTHUMAN DOMAINEXPERTSBYFULLY TAKINGOVERTHEEXECUTION OFOPERATIONALTASKS
  6. 6. ✱ COGNITIVE TECHNOLOGIES 6 ERICSSON TECHNOLOGY REVIEW ✱ JUNE 28, 2018 Interactingwithapersonorwithanother intelligentagentrequiresanintelligentagenttohave theabilitytooperateonthesamelevelofabstraction withasharedunderstandingofconceptsand terminology.Thisincludes,forexample,howgoals areformulatedandhowtheintelligentagents presentinsightsanddecisions. Machine-learnedmodelsarenumerical.They manageabstractionbymappingmeaningto numericalvalues.Thisconstitutesanimplicit translationlayerbetweenthenumerical representationandtheabstractsemantics. Ontology-basedmodelsaresymbolic.Withinan ontology,objectsareestablishedandlinkedtoeach otherusingpredicates.Machinereasoningdraws inferencefromthisrepresentationbylogical inductionanddeduction. Thesymbolicrepresentationassignedtoobjects, predicatesandnumericvaluesisconvention.Itis chosentousethesameabstractionandthesame terminologyasthedomainitreflects.Thisfacilitates anintuitiveexperiencewhenuserscreateand maintaintheknowledgebase. Businessstrategyplanningisagoodexampleofa highlyabstractdomain.Itdealswithconceptssuch asgrowth,churn,customers,satisfactionandpolicy. Numericaldataneedstobeinterpretedtodelivera meaningfulcontributionatthislevel.Anintelligent agentperformingthisinterpretationofdataisa valuableassistantinbusiness-levelprocesses. Theintroductionofintelligentagentswillnot makedomainexpertsunnecessary.Instead,thetask oftheexpertshiftsfromdirectinvolvementin operationalprocessestomaintenanceofthemodels thatdictatetheoperationofautonomousagents. Theabstractionofthemodelscontributestothe efficiencyofthedomainexpert.Apracticalexample isthedesignofdecisionprocessesofexpertsystems proposingactions.Thesesystemsreachananswer bycheckingatreeofbranchingconditions.Even withasmallnumberofvariables,manuallydesigning theseconditionsisatime-consumingand unintuitivetask.Anintelligentagentcancompile thetreefromknowledgeaboutthereasonsfor proposinganaction.Managingtheabstractrules isaconsiderablymoreintuitivebecause the abstractionrisestotheleveltheexpert isusedtothinkingat. Obtainingandmanagingknowledge Theintelligentdigitalassistantexample(seeproof ofconcept#1onpage8)demonstratesanautomated processthatcontributesknowledge.Theassistantis generatedfromproductmanualswritteninnatural languagebyadocument-crawlerapplication.Based onexistingknowledge,itidentifiesandclassifiesthe informationprovidedinthedocuments.Itasserts thisinformationasadditionalknowledge. Furthermore,sitedatastoredincatalogsand inventoriesisautomaticallyandcontinuously assertedintheknowledgebase.Thiskeepsthe knowledgeup-to-date,andthereasoningresults adaptdynamicallytochangedfacts. Theintelligentdigitalassistantalsousesimage recognition.Itidentifiesphysicalelementsandthe currentsituationfromimagesandassertsitsfindings intheknowledgebase.Thisdemonstratesa transformationofnumericdataintosymbolic knowledge.Deep-learningbasedneuralnetworks areparticularlysuccessfulatthistaskofidentifying BUSINESSSTRATEGY PLANNINGISAGOOD EXAMPLEOFAHIGHLY ABSTRACTDOMAIN
  7. 7. COGNITIVE TECHNOLOGIES ✱ JUNE 28, 2018 ✱ ERICSSON TECHNOLOGY REVIEW 7 patternsindataandclassifyingthemsymbolically. Theintelligentdigitalassistant’suseofimage recognitionanditsabilitytoreadnaturallanguage documentsshowthatnotallknowledgeformachine reasoningneedstooriginatefromahumandomain expert.Machine-learning-basedprocessescanadd knowledgeandkeepitup-to-datebasedonwhatis learnedfromdata. Inthisrespect,itisimportanttodifferentiate betweendataandknowledge.Dataisvaluesas providedbytheenvironment.Knowledgeisthe interpretationofthesevalueswithrespecttothe semanticsthatareappliedtogivethedataits meaning.Dataandinformationmodelscategorize dataobjects.Analyticscreatesfurtherknowledge frommultipledataelementsandthedomaincontext. Aknowledgebasepreservesthisknowledgefor reasoning.Whenfacingcontinuouslychangingdata, aswarmofspecializedintelligentagentscankeep theknowledgeup-to-date. Inmachinelearning,thelearnedmodelisthe knowledge,andtrainingexamplesarethemain source.Domainexpertsareinvolvedinselecting variablesanddatasources,andinconfiguringthe learningprocessesaccordingtouse-casegoalsand constraints.Thesuccessoflearning–and consequently,theperformanceofalearning-based intelligentagent–mainlydependsontheavailability andqualityoftrainingdata. Reinforcementlearningisavariantofmachine learningthatlearnsfromasetofrulesanda simulationoftheenvironment.Therefore,itdoes notnecessarilydependonexampledata.However, thelearnedmodelisalsonotbasedonexperience. Themanualdesignofknowledgebydomain expertsremainsamajorsourceofknowledgefor machinereasoning.Thedomainexpertscreatea stablecoreframeworkofassertedterminologyand concepts.Basedonthis,theyexpresstheirdomain expertisebyassertingfurtherconceptsand inferencerules.Theyalsodesigntheapplications thatassessdatasourceandautomaticallyassert knowledge.Thisrequiresstafftobewelltrained inknowledgemanagement,withefficientprocesses andtoolsforknowledgelife-cyclemanagement. Awell-designedmetamodelestablishesa standardforconsistentknowledgerepresentation. Anyknowledgemanagementcompetencegapcan usuallybefilledbyknowledgeengineers,whocan listentothedomainexpertsandtransfertheir knowledgeintoamodel. Amajortaskinmodelingisassemblinga knowledgebaseaccordingtouse-case requirements.Ontologiescanintegrateand interconnectanyformallydefinedmodelallowing extensivereuse.Forexample,dataandinformation modelsusedinapplicationprogramminginterface designconstituteafoundationforassertingdata objects.eTOM[4]andSID[5]areindustry-standard modelscontributingcommontelecommunication terminology.TOVE[6][7]orEnterpriseOntology[8] cancoverbusinessconcepts.Theywereusedinthe businessanalyticsorchestrationexample[9] (seepage9)forinterpretingbusiness-levelquestions. Animportantpartoftheknowledgeof autonomousintelligentagentsistheirgoals. Thedomainexpertusesgoalstotelltheintelligent agentwhatitissupposedtoaccomplish.Ideally, theyareformulatedasabstractbusiness-levelgoals deriveddirectlyfromthebusinessstrategyofthe organization.Thisrequiresbroadknowledgeand adaptabilitytobebuiltintotheintelligentagents, butitpromisesahighlevelofautonomy. ITISIMPORTANTTO DIFFERENTIATEBETWEEN DATAANDKNOWLEDGE
  8. 8. ✱ COGNITIVE TECHNOLOGIES 8 ERICSSON TECHNOLOGY REVIEW ✱ JUNE 28, 2018 Sensing Thinking Acting Knowing / Modeling Query & answer dialog Inference & planning Presentation Expert rules Document crawler Linked data adaptation Context data Object detection Knowledge base CPI store Site data Theintelligentdigitalassistant(seeFigure3)is designedtoassistfieldtechnicianswhoservice basestations[10].Thetechnicianinteractswiththe assistantthroughamobiledevice.Theassistantuses augmentedrealitytoderivethebasestationtype, configurationandstatethroughobjectdetection andvisiblelightcommunication.Forexample,itcan readthestatusLEDofthedevice.Theassistant providesinstructionsandvisualguidancetothe technicianduringmaintenanceoperations. Itdownloadscontextualdataaboutthesiteand requestsanyadditionalinformationthatcould notberetrievedautomaticallythroughaquery andanswerdialogue. Theintelligentdigitalassistantiscurrently aproofofconceptimplementedbyEricsson Research.Wehaveimplementedanddeployedthe machine-reasoningsystemonbackendservers. Thesystemcollectssensedinput,analyses symptomsandpresentscorresponding maintenanceproceduresasaproposedseriesof actions.Domainexpertshavemanuallydesigned theproceduralknowledgeforproblemresolution. Additionally,adocumentcrawlerautomatically readsoperationaldocumentation,whichallowsthe assistanttopresentdocumentsthatarerelevantfor thecurrenttaskstothetechnicianforreference. #1: INTELLIGENT DIGITAL ASSISTANT Proofofconcept Figure 3: Intelligent digital assistant
  9. 9. COGNITIVE TECHNOLOGIES ✱ JUNE 28, 2018 ✱ ERICSSON TECHNOLOGY REVIEW 9 SensingThinkingActing Sensing Knowing Thinking Acting Knowledge base Query interface Query analysis Analytics orchestration Explanation of results Results analysis Assertion of insights Processing Business domain concepts Analytics service descriptions Data interpretation rules Thebusinessanalyticsorchestrationusecase (see Figure4)wasimplementedatEricssonasa proofofconceptwithinamasterthesisproject[9]. Itdemonstrateshowtheabstractlevelofbusiness conceptscanbelinkedwiththetechnicallevelof data-drivenanalytics,sothatintelligentagentscan operateacrossthelevels.Theusecasestartswith abusinessquestionthatcanbesolvedthrough analytics.Anintelligentagentactsasabusiness consultant,providinganalytics-basedassistance toauser.Itanalyzesthequestion,planstheneeded analyticsandorchestrates theexecutionofsuitable analyticsapplications.Whentheresultsareavailable, theintelligentagentreasonsabouttheirmeaningin thecontextofthequestionandexplainstheanswer totheuser. Theinferenceisbasedonaknowledgebase thatcontainsacombinationofabusinessconcept ontologyandabstractservicedescriptionsof analyticsapplications.Itwasbuiltusingexistingand freelyavailablebusinessontologiescombinedwith manually-designedknowledge. #2: BUSINESS ANALYTICS ORCHESTRATION Proofofconcept Figure 4: Business consulting through analytics
  10. 10. ✱ COGNITIVE TECHNOLOGIES 10 ERICSSON TECHNOLOGY REVIEW ✱ JUNE 28, 2018 Machinelearningandmachinereasoning hybridsolutions Gooddecisionsandplansareoftenbasedon understandingmultipledomains.Forexample, expertsinnetworkoperationknowaboutnetwork incidentsandtheappropriateprocedurestosolve them.Theycananalyzetechnicalrootcausesand applycorrectiveandpreventiveactions.Thesame expertsusuallyalsoknowsomefactsaboutthe broaderbusinessenvironment.Knowingabout financialgoalsandServiceLevelAgreementshelps themtoprioritizetasks.Byunderstandingthe applicationdomainofadeviceortheconcernsofa user,theycancustomizethesolution.Theymight alsoknowaboutmarketingeffortsorproductsin developmentandproactivelyprovideconsulting. Allthisknowledgeallowsanexperttomaketheright decisions.Forintelligentagents,itisachallengeto operatewiththesameamountofdiverseknowledge andtoprovideanequallydiverserangeofactions. Theroleofmachinereasoning Theknowledgeusedinmachinereasoningispure datadecoupledfromtheimplementationofthe inferenceengine.Changesinbehaviorand extensionsofscopemustthereforebereached bychangingthemodeldataratherthanthe implementationoftheintelligentagent.Therefore, machine-reasoningmodelsarewellsuitedto integratingontologiesandinferencerulesfrom multipledomains,ifformalandsemantic consistencyispreserved. Ideally,alayerofcoreconceptsandterminology commontoalldomainsshouldbeusedtoanchor domain-specificmodels.Thisallowsinference enginestotraverseacrossdomainbordersanddraw conclusionsfromallconstituentdomainmodels. Ifthemodelsfromdifferentdomainsalreadyuse similarconcepts,butdefinethemdifferently,a“glue” modelcanrelatethembyintroducingknowledge aboutthedifferences. Thedrawbacksofthemulti-domainknowledge basedescribedherearethecomplexityof maintainingmodelconsistencyandtheperformance oftheinferencegenerationduetothenumberof knowledgeelementstoprocess. Theroleofmachinelearning Inmachinelearning,eachadditionaldomain contributesyetanothersetofvariablesadding furthernumericaldimensionstothemodel. Thisintroduceschallengessuchastheneedfor trainingexamplesthatcontainconsolidateddata samplesfromalldomains.Thereisalsoanincrease inthenumberofdatapointsrequiredtoreach acceptablestatisticalcharacteristics.Thecombination ofmoredimensionsandhigherdatavolumeincreases theprocessingcost.Furthermore,eachchangein scoperequiresafulllife-cycleloopincludingdata selection,implementation,deploymentandlearning untilanewmodelisavailableforproductiveuse. Consideringthesechallenges,machine-learned modelsarebestsuitedtobespecialistsinconfined tasks.Asecondarylayerofmodelscanthenbuildon thespecialistinsightsandevaluatetheminabroader context.Thesecondtieroperatesonhigher abstractionwithconceptsfrommultipledomains. However,sincetrainingexamplesatthislevelare broadinscope,theytendtobehardtoobtain. Domainexpertsarestillavailable,though,sousing machinereasoningisalwaysfeasible.Ingeneral, machinelearningexcelsatinferencethatresults fromprocessinglargeamountsofdata,while machinereasoningworksverywellindrawing conclusionsfrombroad,abstractknowledge. Hybridsolutions Theresultisanenvironmentcomprisedof orchestratedorchoreographedintelligentagents. Coordinationandcollaborationisdonethroughthe knowledge.Amachine-learnedmodelcan contributeitsfindingsthroughasynchronous assertion.Amappingapplicationisdesignedto monitorthenumericoutputofamachine-learned modeloranalyzethelearnednumericmodelitself. Whennewoutputisgenerated,oranewversionof
  11. 11. COGNITIVE TECHNOLOGIES ✱ JUNE 28, 2018 ✱ ERICSSON TECHNOLOGY REVIEW 11 themodelisavailable,themappingapplication interpretsitinthedomaincontext,determinesits meaningandgeneratesarespectivesymbolic representation.Thisconstitutesnewknowledge thatisassertedintotheknowledgebase. Alternatively,anapplicationincorporatinga machine-learnedmodelcanbelinkeddirectlyinto theknowledgebaseactingasaproxyfora knowledgeobject.Areasoningprocesswouldcall thelinkedapplicationwhentherespective knowledgeisneeded.Theapplicationgeneratesa replybasedonallcurrentlyavailabledata. Bothmethodscreateahybridofmachinelearning andmachinereasoningthatenablesdynamic adaptationofthereasoningresultsbasedon learningandthelatestdata.Asynchronousassertion actslikeadomainexpertcontinuouslyupdating knowledge.Aknowledgeproxyapplication synchronouslygeneratesknowledgeondemand. However,thiscomesatthecostofdelayingthe reasoningprocess. Symbolicneuralnetworks Symbolicneuralnetworksspecializeinlearning abouttherelationshipsbetweenentities.They implicitlyabstractfromanunderlyingstatistical model,whichallowsthemtoanswerabstract questionsdirectly.Oneexampleisimageprocessing combiningmultiplemachine-learnedmodels. Onemodelidentifiestheobjectsseen.Another learnsabouttherelationshipbetweentheobjects. Athirdhaslearnedtointerpretquestionsasked. Duetotheimplicitabstractionanduseofsymbolic representation,theinsightsgeneratedbythese modelswouldintegrateseamlesslyintoaknowledge baseandfurtherreasoning.However,getting referencedataforlearningisachallengeinthis scenarioandwouldusuallybedependentonhuman expertscreatingsamples.Asthissetuphasmachine learningatitscore,italsodoesnotscalewelltoahigh numberofconcernsandvariables.Nevertheless, itcanfindandcontributeknowledgeaboutnew relationshipsthatwashithertounknowntoexperts. Tieredimplementation Thetieredimplementationapproachusesmachine learningonthelayerofspecialistmodelsand machinereasoningforconsolidationacross domains.Thisassignmentofrolesreflectsstrengths ofthetechnologyfamilies,althoughadifferent selectionispossibledependingontheusecaseand environment.Forexample,machinelearningmaybe successfullyappliedforcross-domainconsolidation iftrainingdataisavailable.Andmachinereasoning canimplementaspecialistintelligentagent,for example,ifitincorporatesthemanually-designed rulesofahumandomainexpert. Conclusion Intelligentagentswiththeabilitytowork collaborativelypresentthebestopportunityfor networkoperatorsanddigitalserviceprovidersto createtheextensivelyautomatedenvironmentthat theirbusinesseswillrequireinthenearfuture. Cognitivetechnologies–andinparticulara combineduseofmachinereasoningandmachine learning–providethetechnologicalfoundationfor developingthekindofintelligentagentsthatwill makethisflexible,autonomousenvironmenta reality.Theseagentswillhaveadetailedsemantic understandingoftheworldandtheirownindividual contexts,aswellasbeingabletolearnfromdiverse inputs,andshareortransferexperiencebetween contexts.Inshort,theyarecapableofdynamically adaptingtheiractionstoabroadrangeofdomains andgoals. COGNITIVETECHNOLOGIES WILLMAKETHISFLEXIBLE, AUTONOMOUSENVIRONMENT AREALITY
  12. 12. ✱ COGNITIVE TECHNOLOGIES 12 ERICSSON TECHNOLOGY REVIEW ✱ JUNE 28, 2018 References 1. AcadiaUniversity,OnCommonGround:Neural-SymbolicIntegrationandLifelongMachineLearning (researchpaper),DanielL.Silver,availableat: http://daselab.cs.wright.edu/nesy/NeSy13/silver.pdf 2. Ericsson Technology Review, Generating actionable insights from customer experience awareness, September 30, 2016, Niemöller, J; Sarmonikas, G; Washington N, available at: https://www.ericsson.com/ en/ericsson-technology-review/archive/2016/generating-actionable-insights-from-customer-experience-awareness 3. AnnalsofTelecommunications,Volume72,Issue7-8,pp.431-441,Subjectiveperceptionscoring: psychologicalinterpretationofnetworkusagemetricsinordertopredictusersatisfaction,2017,Niemöller,J; Washington,N,abstractavailableat:https://link.springer.com/article/10.1007%2Fs12243-017-0575-6 4. TMForum,GB921BusinessProcessFramework(eTOM),R17.0.1,availableat: https://www.tmforum.org/resources/suite/gb921-business-process-framework-etom-r17-0-1/ 5. TMForum,GB922InformationFramework(SID),Release17.05.1,availableat: https://www.tmforum.org/resources/suite/gb922-information-framework-sid-r17-0-1/ 6. Berlin:Springer-Verlag,pp.25-34,TheTOVEprojecttowardsacommon-sensemodeloftheenterprise, IndustrialandEngineeringApplicationsofArtificialIntelligenceandExpertSystems,1992,Fox,M.S., availableat:https://link.springer.com/chapter/10.1007/BFb0024952 7. UniversityofToronto,TOVEOntologies,availableat:http://www.eil.utoronto.ca/theory/enterprise-modelling/ tove/ 8. CambridgeUniversityPress,TheKnowledgeEngineeringReview,Vol.13,Issue1,pp.31-89,TheEnterprise Ontology,March1998,King,M;Moralee,S;Uschold,M;Zorgios,Y,abstractavailableat: https://www. cambridge.org/core/journals/knowledge-engineering-review/article/enterprise-ontology/17080176D5F06DEAEA8 DBB2BAA9F8398 9. TilburgUniversity,MediatingInsightsforBusinessNeeds,ASemanticApproachtoAnalyticsOrchestration (master’sthesis),June2016,Alhinnawi,B. 10. EricssonMobilityReport2018,Applyingmachineintelligencetonetworkmanagement,StephenCarlsson, availableat:https://www.ericsson.com/en/mobility-report/reports/june-2018
  13. 13. COGNITIVE TECHNOLOGIES ✱ JUNE 28, 2018 ✱ ERICSSON TECHNOLOGY REVIEW 13 Further reading 〉〉 CIO, Artificial intelligence is about machine reasoning – or when machine learning is just a fancy plugin, November 3, 2017, Rene Buest, available at: https://www.cio.com/article/3236030/machine-learning/ artificial-intelligence-is-about-machine-reasoning-or-when-machine-learning-is-just-a-fancy-plugin.html 〉〉 Microsoft Research, From machine learning to machine reasoning – An essay, February 13, 2013, Léon Bottou, available at: https://www.microsoft.com/en-us/research/wp-content/uploads/2014/01/mlj-2013.pdf Jörg Niemöller ◆ is an analytics and customer experience expert in solution area OSS. He joined Ericsson in 1998 and spent several years at Ericsson Research, where he gained experience of machine-reasoning technologies and developed an understanding of their business relevance. He is currently driving the introduction of these technologies into Ericsson’s portfolio of Operations Support Systems / Business Support Systems solutions. Niemöller holds a degree in electrical engineering from TU Dortmund University in Germany and a Ph.D. in computer science from Tilburg University in the Netherlands. Leonid Mokrushin ◆ is a senior specialist in cognitive technologies at Ericsson Research. His current focus is on investigating new opportunities within artificial intelligence in the context of industrial and telco use cases. He joined Ericsson Research in 2007 after postgraduate studies at Uppsala University, Sweden, with a background in real- time systems. He received an M.Sc. in software engineering from Peter the Great St. Petersburg Polytechnic University, Russia, in 2001. theauthors
  14. 14. ✱ COGNITIVE TECHNOLOGIES 14 ERICSSON TECHNOLOGY REVIEW ✱ JUNE 28, 2018 ISSN 0014-0171 284 23-3316 | Uen © Ericsson AB 2018 Ericsson SE-164 83 Stockholm, Sweden Phone: +46 10 719 0000

×