I2oT Advanced Direction of the Internet of Things文档格式.docx
《I2oT Advanced Direction of the Internet of Things文档格式.docx》由会员分享,可在线阅读,更多相关《I2oT Advanced Direction of the Internet of Things文档格式.docx(9页珍藏版)》请在冰豆网上搜索。
knowledgeproducing;
strategyformulation;
intelligentinternetofthings
T1Introduction
herearetwomainmotivationsforexpandingtheInternettotheInternetofThings(IoT).Thefirstmotivationistoexpandtheamountofinformationsharedbydatabasesandobjectsintherealworld.Thesecondmotivationistoenableusersnotonlytoshareinformationbutalsocontrolobjectsintherealworld.ThesemakeIoTmuchmoreattractiveinsociety.Inotherwords,IoTisagoodadvancementoftheconventionalInternet.
Intermsoftechnologicaldevelopment,however,IoTisstillinitsinfancyandcanbegreatlyimprovedbyendowingIoTfunctionswithmuchmoreintelligence[1].Significantprogresshasbeenmadeinartificialintelligence(AI)overthepastdecade.AllAItechnologiesneededtomakeIoTmoreintelligentandevolveintoI2oTarenowfeasible.ThemainconcernatthemomentishowtounderstandandeffectivelyapplyAItechnologiestocurrentIoTsystems.
2ABriefDescriptionofIoT
ThepurposeofIoTistoexpandthefunctionsofexistingInternetandmakeitmoreuseful.WithIoT,userscansharenotonlyinformationprovidedbyhumansandcontainedindatabasesbutalsoinformationprovidedbythingsinphysicalworld.ThesimplifiedfunctionalmodelofIoTisshowninFig.1.
AsinFig.1,IoThassensors,foracquiringinformationaboutthestateofthings;
anembeddedprocessor,forproducingordersthatregulatethestateofthings;
wirelesstechnology,fortransferringinformationfromsensorstoInternetandInternettocontroller;
andcontrolunit,forexecutinghumanordersregulatingthestateofthings.
TakeIoTformaintainingroomtemperatureforexample.Astandardroomtemperatureisdesignatedinadvance,andtheactualroomtemperatureisacquiredbythesensor(s)andtransferredviawirelesstoInternet.Afterreceivingtheactualroomtemperature,theembeddedprocessorcomparesitwiththedesignatedvalueandgeneratesanordertoregulatetheroomtemperatureandkeepitwithinacertainrange.ThisorderisimmediatelysenttothecontrolunitviatheInternetandwirelessunitandisexecutedbytheactuatorofthecontrolunit. Ifinformationaboutthestateofthethingconcernedcanbeacquiredbysensorsandcontrolledbyactuators,andifthefunctionperformedbytheembeddedprocessorisnottoocomplicated,theIoTtechnologyisfeasible.
IfphysicalthingsandtheirenvironmentinIoTbecomecomplex,thefunctionsoftherequiredembeddedprocessorsalsobecomecomplex,andconventionaltechnologiesofthecurrentIoTwillnolongerbesatisfactory.
Unfortunately,problemswithcomplexfactorsareveryoftenimportanttoeconomicandsocialdevelopment.Atypicalexampleisairpollutionoveralargearea.Anothertypicalexampleisglobalwarming.Peoplewanttoknowinformationabouttheairqualityandweatherconditionsandcontrolthemincertainways.Therefore,efficientlydealingwithcomplexproblemsisanunavoidableresponsibilityofscientists.
Themostpromisingapproachtohandlingsuchcomplexproblemsisartificialintelligent.Thereasonforthisproposalisthefactthatcentralneedforsolvingcomplexproblemsisthelearningability.
3FundamentalConceptsandPrinciplesof
ArtificialIntelligence
Inanarrowsense,AIhastraditionallyimpliedthesimulationoflogicalhumanthinkingusingcomputertechnology.Withinthisframework,thefieldsofartificialneuralnetworks(ANNs)[2]-[4]andsensor?
motorsystems(SMSs)[5]-[7]wereconsideredextraneous,eventhoughbothfieldshavebeenconcernedwithsimulatingthefunctionsofthehumanbrain.ANNandSMShadtoformanewdisciplinecalledcomputationalintelligence(CI).ComputationalintelligencehasbecometheotherapproachtoAI.ItismorereasonableforthetermAItoencompassbothAIinnarrowsenseandCI.Inthecontemporarysense,AIisnowre?
termedunifiedAI[8]-[9].
Inthispaper,AImeansunifiedAI,ageneraltermrepresentingthetheoryandtechnologyrelatedtosimulatingintellectualabilitiesofhumanbeing,includingtheabilitytounderstandandsolveproblems.WhatfollowsisabriefexplanationofhowAIcanhandlecomplexproblems[10]-[12].
WhatAIsimulatesandoffersisnotanythingelsebutthelearningabilityofhumanbeings,i.e.,learningtounderstandandsolvetheproblem.Therefore,learningisthecentralfeatureinAIandlearning?
technologyisthekeytohandlingproblems.
ThesimplestmodelforAIisroughlyabstractedinFig.2.
Ontologicalinformation(OI)inFig.2isinformationaboutthestateandpatternofthestatevariancethatarepresentedbytheobjectintheenvironmentoftheoutsideworldandthataretheresourcesandcluesforlearningtounderstandtheproblem.Ontheotherhand,thesubject’sactionorreactionappliedtotheobjectcanbelearntbasedonanunderstandingoftheproblem. AmorespecificfunctionalmodelofthetechnologiesinAIisshowninFig.3.InFig.3,AItechnologiesareinterconnectedandinteractwitheachother.
3.1CategoriesofAITechnology
3.1.1Perception
ThistechnologyisusedtoacquiretheOIabouttheobjectorprobleminitsenvironment.ItisalsothetechnologyforconvertingOItoepistemologicalinformation(EI).
Epistemologicalinformationisinformationperceivedbythesubjectaboutthetrinityoftheform(syntacticinformation),content/meaning(semanticinformation),andutility/value(pragmaticinformation)concerningOI.
UnlikethetraditionalconceptofinformationproposedbyClaudeShannon,EIcomprisesthetrinityoftheform,content/meaning,andutility/valueandisthebasisoflearning.ThisiswhyEIisalsooftencalledcomprehensiveinformation.
TheessentialfunctionofperceptionistoconvertOItoEI.ThisisthefirstclassofinformationconversioninAI.
3.1.2Cognition
ThemainfunctionofcognitiontechnologyistoconvertEI,whichisperceivedbythesubjectfromOI,intothecorrespondingknowledgeabouttheobject.ThisisthesecondclassofinformationconversionneededinAI.TheonlypossibleapproachtoconvertingEItoknowledgemustbelearning―thereisnootherway.
3.1.3Decision?
Making
Thetechnologyusedindecision?
makingconvertsEItointelligentstrategy(IS)basedonknowledgesupportandisdirectedbythegoalofproblemsolving.Thestrategyisjusttheproceduralguidanceforproblem?
solving.ThisisthethirdclassofinformationconversioninAI.
Theradicalfunctionofdecision?
makingtechnologyislearningtofindtheoptimalsolutionforagivenproblem.ThereareusuallyanumberofwaysofachievingthedesignatedgoalfromastartingpointexpressedbyEI.Adecisionshouldbemadethroughintelligentuse,vialearning,oftherelevantknowledgeprovided.
3.1.4Strategy?
Execution
ThistechnologyisusedtoconverttheISintointelligentaction(IA)thatwillsolvetheproblem.
3.1.5Strategy?
Optimization
Becauseofvariousnon?
idealfactorsinallsub?
processesinFig.3,thereareoftenerrorswhenintelligentactionisapplied.Theseerrorsareregardedasnewinformationandarefedbacktotheinputoftheperceptionofthemodel.Withthisnewinformation,theknowledgecanbeimprovedvialearning,andthestrategycanbeoptimized.Suchanoptimizationprocessmightcontinuemanytimesuntiltheerrorissufficientlysmall. Insum,alltheAItechnologiesheretomentionedarelearning?
based,andthisiswhyAIispowerful.
3.2ImplementationIssuesfortheThreeClassesof
InformationConversion
PerceptiontechnologycanbeimplementedusingthemodelinFig.4,whichconvertsOItoEI,thetrinityofX,YandZ,andisthefirstclassofinformationconversion.
Fig.4showsthattheontologicalinformation(denotedS)isappliedtotheinputoftheperceptionmodelandmappedtothecorrespondingsyntacticinformation(denotedX).Next,thepragmaticinformation(denotedZ)canberetrievedfromtheknowledgebase,inwhichmanyX?
Zpairs,{X(i),Z(i)},arestored.WhenXismatchedwithX(i0),thenZ(i0)isregardedasthepragmaticinformationcorrespondingtoX.Incasenomathcanbefound,theequationcanbeusedtofindZ;
Z=Cor(X,G)
(1)
whereXandGareexpressedasvectors;
andCoristhecorrelationoperation.BecauseXandZarenowavailable,thesemanticinformationYcanbeinferredfrom:
Y=λ(X,Z)εS
(2)
whereSisthespaceofsemanticinformation,andλisthelogicoperationmappingthepairof(X,Z)toYinS.ThismeansthatYisasubsetofSwhenbothXandZaresimultaneouslyvalid.Inotherwords,YisdeterminedbythejointconditionsofXandZ(Fig.5).
Asaresult,OIisconvertedintoEI,whichisthetrinityofX,YandZ,viathemodelinFig.4.Thistechnologyiscompletelyfeasibleinpractice.
CognitiontechnologycanbeimplementedusingthemodelinFig.6,with