




版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
OpportunitiesandChallengesfromArtificialIntelligenceandMachineLearningfortheAdvancementofScience,
Technology,andtheOfficeofScienceMissions
AreportfortheAdvancedScientificComputingAdvisoryCommitteefromtheSubcommitteeonArtificialIntelligence,MachineLearning,Data-
intensiveScienceandHigh-PerformanceComputing
Chair:TonyHeySeptember2020
Caption:
ThecoverimageshowsthecrystalstructureofDy2Ti2O7inwhichthemagneticionDy3+oftherareearthelementDysprosium(shownincyan)occupiessitesonageometricallyfrustratedcorner-sharingtetrahedralnetwork.AI/MLmethodswereusedtosolveforthecouplingsinthematerialandtoidentifytheoriginofitsfreezingintoaglassystate.Neuralnetworkswereusedtoextractthephaseandcouplingsinthesystemfromdiffuseneutronscatteringdatabysolvingtheinversescatteringproblem.Thestrongsingle-ionicanisotropyofDy3+ionsdefinedbytheirmolecularenvironmentrestrictstheeffectivemagneticmomenttoaligneitherinwardoroutward.Themagneticmomentsimposeanice-rulewherelow-energyspinstatesarerestrictedtoatwo-inandtwo-outconfigurationforeachtetrahedronandthebreakingofthisice-rulecreatestwofractionalizedmagneticmonopoleswithoppositepolarity.
Acknowledgement:
AnjanaSamarakoonandAlanTennant,OakRidgeNationalLaboratory.
TableofContents
ExecutiveSummary 6
Introduction 6
Context 7
KeyFindings 8
RecommendationsforDOE’sOfficeofScience 12
Report 16
IntroductionandBackground 16
ChargeLettertoASCR 17
SubcommitteeInformationGatheringActivities 17
DOEastheleadagencyforAI/MLappliedtoFacilitiesScience 19
OpportunitiesandchallengesfromArtificialIntelligenceandMachineLearningforthe
advancementofscience,technology,andOfficeofSciencemissions 20
StrategiesfortheDOEOfficeofSciencetoaddressthechallengesanddeliveronthe
opportunities 21
Introduction 21
AIApplications 23
AIAlgorithmsandFoundations 31
AISoftwareInfrastructure 37
NewHardwareTechnologiesforAI 40
InstrumenttoEdgeComputing 41
AI/MLWorkforce:Training,Focusing,andRetention 42
UniversityPartnerships 44
CollaborationwithIndustry 45
Inter-AgencyCollaboration 46
InternationalCollaboration 47
ImportanceofASCR’slong-termAppliedMathematicsandComputerScience
ResearchPrograms 49
SummaryofConclusions 49
Figures 51
Figure1:AI,MachineLearning,DeepLearninginaNutshell 51
Figure2:WhatisaDataScientist? 52
Figure3:StructureofSCAIforScience10-yearInitiative 53
ReferencesandURLs 54
AppendixA:ChargeLetter 56
AppendixB:SubcommitteeMembers 58
AppendixD:ListofAcronyms 64
Acknowledgements 65
“AIwon’treplacethescientist,butscientistswhouseAIwillreplace
thosewhodon’t.”
AdaptedfromaMicrosoftreport,“TheFutureComputed”
ExecutiveSummary
Introduction
InFebruary2019,thePresidentsignedExecutiveOrder13859,MaintainingAmericanLeadershipinArtificialIntelligence[1].ThisorderlaunchedtheAmericanArtificialIntelligenceInitiative,aconcertedefforttopromoteandprotectAItechnologyandinnovationintheUnitedStates.TheInitiativeimplementsagovernment-widestrategyincollaborationandengagementwiththeprivatesector,academia,thepublic,andlike-mindedinternationalpartners.
Amongotheractions,keydirectivesintheInitiativecalledforFederalagenciesto:
PrioritizeAIresearchanddevelopmentinvestments,
Enhanceaccesstohigh-qualitycyberinfrastructureanddata,
EnsurethattheUSmaintainsaninternationalleadershiproleinthedevelopmentoftechnicalstandardsforAI,and
ProvideeducationandtrainingopportunitiestopreparetheAmericanworkforcefortheneweraofAI.
ThemissionoftheDepartmentofEnergy(DOE)istoensureAmerica’ssecurityandprosperitybyaddressingitsenergy,environmental,andnuclearchallengesthroughtransformativescienceandtechnologysolutions.IntermsofScienceandInnovation,theDOE’smissionistomaintainavibrantUSeffortinscienceandengineeringasacornerstoneofoureconomicprosperitywithclearleadershipinstrategicareas.
FromJulytoOctoberin2019,theArgonne,OakRidge,andBerkeleyNationalLaboratorieshostedaseriesoffourAIforScienceTownHallmeetingsinChicago,OakRidge,Berkeley,andWashingtonDC.Thefourmeetingswereattendedbyover1300scientistsfromthe17DOELabs,39companies,andover90universities.ThegoaloftheTownHallserieswas‘toexaminescientificopportunitiesintheareasofartificialintelligence,BigData,andhigh-performancecomputing(HPC)inthenextdecade,andtocapturethebigideas,grandchallenges,andnextstepstorealizingthese.’ThediscussionsatthemeetingswerecapturedinthefinalreportoftheAIforScienceTownHallmeetings[2].
InresponsetoachargeletterfromtheDOE’sOfficeofScience(SC),theAdvancedScientificComputingResearch(ASCR)programaskeditsAdvisoryCommittee(ASCAC)toestablishasubcommitteetoexplorethescientificopportunitiesandchallengesarisingfromtheintersectionofArtificialIntelligence(AI)andMachineLearning(ML)withdata-intensivescienceandhighperformancecomputing.Specifically,thisAIforSciencesubcommitteewasaskedto:
AssesstheopportunitiesandchallengesfromArtificialIntelligenceandMachineLearningfortheadvancementofscience,technology,andtheOfficeofSciencemissions.
IdentifystrategiesthatASCRcanuse,incoordinationwiththeotherSCprograms,toaddressthechallengesanddeliverontheopportunities.
ThisreportistheresultoftheSubcommittee’sinvestigationofthesechargequestions.TosetthecontextasummaryofAI,MLandDeepLearningisincludedherealongwithacharacterizationofdifferentrolesfordatascientists.Thisexecutivesummaryreportsthesubcommittee’skeyfindingsandrecommendations.
Context
ThetermArtificialIntelligencewascoinedbyJohnMcCarthyforaworkshopatDartmouthCollegeinNewHampshirein1956.Attheworkshop,McCarthyintroducedthephrase‘ArtificialIntelligence’whichhelaterdefinedas[3]:
‘Thescienceandengineeringofmakingintelligentmachines,especiallyintelligentcomputerprograms.’
Bycontrast,thefieldofMachineLearningislessambitiousandcanberegardedasasub-domainofartificialintelligence[4]:
‘Machinelearningaddressesthequestionofhowtobuildcomputersthatimproveautomaticallythroughexperience.Itisoneoftoday'smostrapidlygrowingtechnicalfields,lyingattheintersectionofcomputerscienceandstatistics,andatthecoreofartificialintelligenceanddatascience.Recentprogressinmachinelearninghasbeendrivenbothbythedevelopmentofnewlearningalgorithmsandtheoryandbytheongoingexplosionintheavailabilityofonlinedataandlow-costcomputation.’
Finally,DeepLearningneuralnetworksareasubsetof
MachineLearning
methodsthatarebasedon
artificialneuralnetworks
(ANNs)[5]:
‘AnANNisbasedonacollectionofconnectedunitsornodescalled
artificial
neurons
,whichlooselymodelthe
neurons
inabiologicalbrain.Eachconnection,likethe
synapses
inabiologicalbrain,cantransmitasignaltootherneurons.Anartificialneuronthatreceivesasignalthenprocessesitandcansignalneuronsconnectedtoit.The"signal"ataconnectionisa
realnumber
,andtheoutputofeachneuroniscomputedbysomenon-linearfunctionofthesumofitsinputs.Theconnectionsarecallededges.Neuronsandedgestypicallyhavea
weight
thatadjustsaslearningproceeds.’
Theartificialneuronsinthesenetworksarearrangedinlayersgoingfromaninputlayertoanoutputlayerwithconnectionsbetweentheneuronsinthedifferentlayers.DeeplearningneuralnetworksaremerelyasubsetofsuchANNswithverylargenumbersofhiddenlayers.OntheImageNetImageRecognitionChallenge,the2015competitionwaswonbyateamfromMicrosoftResearchusingaverydeepneuralnetworkofover100layersandachievedanerrorrateforobjectrecognitioncomparabletohumanerrorrates[6].Figure1triestocapturetheessenceofthisAI,MachineLearning,andDeepLearninghierarchy[7]
Figure2attemptstodefinethreedifferentrolesforadatascientist[8].Thefirstroleisthatofadataengineerwhoisexpertatoperatingclosetothecomputers,instruments,andsensorsthatgeneratethedata.ThesecondroleisthatofadataanalystwhousesadvancedstatisticsandAI/MLmethodstoexploretheexperimentaldatasetsandassisttheresearchertoextractnewscience.Finally,inthisclassification,thereisathirdroleofdatacuratorwhoisexpertinmanaginglargedatasets,curatingthedatawithsuitablemetadataforre-use,andlaterarchiving.AllthreeoftheseaspectsofdatasciencearerelevantfortheproposedAIforScienceinitiative.
KeyFindings
FindingA
ThegrowingconvergenceofAI,Data,andHPCprovidesaonceinagenerationopportunitytoprofoundlyacceleratescientificdiscovery,createsynergiesacrossscientificareas,andimproveinternationalcompetitiveness.
Scienceandcomputingarenowinaneraofpost-Moore’sLawsilicontechnologiesandthereisanurgentneedforasea-changeintheprogrammabilityandproductiveuseofincreasinglycomplex/heterogeneoussystemsandtheseamlessintegrationofdata,algorithms,andcomputingresources.DoingsowillhelpmanagethechallengesofBigData,carryingoutscienceatscaleusingDOE’smostadvancedfacilities,leveragetheworkforceattheLabs,andsetthestagefortheemergenceanddevelopmentofrobustandreliableAIsystemswiththeabilitytolearnforthemselvesindomain-sciencespecificareas.
FindingB
SciencecangreatlybenefitfromAImethodsandtools.However,commercialsolutionsandexistingalgorithmsarenotsufficienttoaddresstheneedsofscienceautomationandscienceknowledgeextractionfromcurrentandfutureDOEfacilitiesanddata.
CurrentAIsolutionscanbesuccessfullyappliedtoconductavarietyofdataanalyses.However,newalgorithms,foundations,andtoolsareessentialtoaddressinguniquescienceconcernsinabroadspectrumofscienceapplications.AIalgorithmsneedtobeabletodealwithsparse,heterogeneous,andun-labeleddatasetsthatareoftenexpensivetocollectandarchiveandbeabletogeneratemodelsthatincorporatedomainknowledgeandphysicalconstraints.AI-enabledexperimentaldesignandcontrolarenecessaryforoptimaluseofDOEfacilities.Inthesciencecontext,AImethodsneedtohaveprovablecorrectnessandperformance,beabletoexposebiases,andtoquantifyuncertainties,errors,andprecision.
FindingC
AdoptingAIforSciencetechnologiesthroughouttheOfficeofSciencewillenableUSscientiststotakeadvantageofthetremendousnewadvancesintheDOE’sscientificuserfacilities.
TheDOE’sOfficeofScienceprovidesUSresearcherswithaccesstothelargestandmostdiversesuiteofscientificexperimentalfacilitiesintheworld–fromX-raysynchrotronsandneutronsourcestointegrativegenomicsandatmosphericradiationfacilities–aswellastotheworld’smostcapablehighperformancecomputingfacilities.UpgradestotheseuserfacilitiesandnewnuclearphysicsfacilitiescomingonlinenowandoverthenextdecadewilldramaticallyincreasetheamountofnewdataproducedacrossallofthescientificdomainssupportedbytheOfficeofScience,posingnewchallengesandnewopportunities.Science-awareAItechnologieswillallowustoextractinformationandscientificunderstandingfromthesetremendousnewdatasources.
FindingD
RealizingthepotentialforagenerationalshiftinscientificexperimentationattheDOELaboratoriesduetoscience-drivenAI/MLtechnologiesrequiresfarmorethansimplycomputepowerandencompassesthefullspectrumofcomputinginfrastructures,rangingfromubiquitoussensorsandinterconnectivityacrossdevicestoreal-timemonitoringanddataanalytics,andwillrequireaconcertedandcoordinatedR&DeffortonAI/MLalgorithms,tools,andsoftwareinfrastructure.
AcrosstheSCprograms,scientificapplicationsofArtificialIntelligence(AI)andMachineLearning(ML)canbuildonthepowerofsensornetworks,edgecomputing,andhighperformancecomputerstotransformscienceandenergyresearchinthefuture.GiventhehighlyspecializednatureofmanyDOEfacilitiesandscientificresearchdomains,itisnotpossibletorelysolelyonthird-partyAI/MLresearchanddevelopment(R&D)forthistransformation.TheDOEwillneedtobuilditsownR&Dprogramsthatfocusonthemostchallengingscience-drivenapplications.SoftwareinfrastructurewillberequiredthatcombinesleadershipinAI/MLtoolsandalgorithmswiththeDOE’straditionalstrengthsinsimulationandmodelingtechnologiesandthatcanexecuteonnewcomputingplatformscapableofhighperformanceonbothtypesofapplications.TheanticipatedreturnswillhelpensurethattheUScontinuestomaintainandenhanceleadershipinbothdata-intensivescienceandhighperformancecomputing.
FindingE
TheDOELabsareuniquelypositionedtointegrateAI/MLtechnologiesacrossahostofscientificchallengesthankstotheenviablecultureofco-designteamsconsistingofscientificusers,instrumentproviders,theoreticalscientists,mathematiciansandcomputerscientiststhathasprovensosuccessfulintheExascaleComputingProject.
Thesubcommittee,therefore,seesacompellingneedforAI/MLtechnologiestobeincorporatedintoalloftheDOE’sscientificresearchcapabilitiesinordertoeffectivelysupporttheOfficeofScience’smissionsinenergy,nationalsecurity,fundamentalsciences,andtheenvironment.DOE’sNationalLaboratories,togetherwithUSuniversityandindustrypartners,havethenecessaryassetstoinitiatealarge-scaleprogramtoacceleratethedevelopmentofsuchcapabilitiesandthenecessaryworkforcetonotonlymeettheirSCmissionneedsbutalsobenefitallofDOE’sactivities.
FindingF
TheimpactofaDOE-drivenAI/MLstrategyforsciencewillhavenationalimplicationsfarbeyondtheOfficeofScienceandwilldrivenewindustrialinvestments,includingacceleratingengineeringdesigns,synthesizingmaterials,andoptimizingenergydevices,aswellasadvancinghardwareandsoftwarecomputingcapabilities.
Thebenefitstothenationindevelopingpowerfulandbroad-basedAIforSciencecapabilitiesintheDOELaboratorieswillextendwellbeyondtheDOE’sprograms.ThedevelopmentofcomprehensiveAI/MLcapabilitieswillbenefitothergovernmentagenciesandabroadrangeofindustriesinthiscountry,includingenergy,pharmaceutical,aircraft,automobile,entertainment,andothers.MorepowerfulAIcapabilitieswillallowthesediverseindustriestomorequicklyengineernewproductsthatcanimprovethenation’scompetitiveness.Inaddition,therewillbeconsiderableflow-downbenefitsthatresultfrommeetingboththehardwareandsoftwareAIchallenges.InitiatingamajorprogramfocusedonapplyingAI/MLtechnologiestotheDOEscientificchallengeswouldbelikelytoleadtosignificantgainsinUScompetitivenessinseveralcriticalareasandtechnologies.
FindingG
AworkforcetrainedinadvancedAI/MLtechnologieswouldplayapivotalroleinenhancingUScompetitiveness.
Thetraining,focusing,andretentionofacadreofyoungpeople,expertsinbothinventinganddeliveringthetechniquesandtechnologiesofAI/MLforscienceandengineeringapplications,iscriticaltothesuccessoftheAIforScienceagenda.TheOfficeofScienceDOELaboratoriescanplayakeyroleincooperationwiththeNationalScienceFoundation(NSF).Overthepast20years,theInformationTechnology(IT)industryhasexpandeddramatically,drivenbye-commerce,socialmedia,cloudservices,andsmartphones.Inrecentyears,theemergenceoftheInternetofThings(IoT),thewidespreaddeploymentofhealthcaresensors,increasingindustrialautomation,andthedevelopmentofautonomousvehicleshavefurtherexpandedthedomainofAI/MLdataanalyticsandservices.Inresponsetothesegrowingworkforcedemands,moststudentsarenowtrainedinsoftwaretoolsandtechniquesthattargetcommercialopportunities.Atpresent,commercialtoolsarerathergenericandnotwell-targetedtoscientificapplications.AnAIforScienceinitiativewoulddeliverscientificAI/MLtoolsandenvironmentsappropriatefortraininganewgenerationofscientistsandengineers.
FindingH
PartneringwithotherAgenciesandwithinternationaleffortswillbeimportanttodeliverontheambitiousgoalsofanAIforScienceinitiative.
TheNSFandNIH,thetwoothermajorscience-focusedfundingagenciesintheUS,alsohaveorareplanning,majorinvestmentsinAI/MLprogramsfortheirscientificdomains.InseveralareasthereareclearsynergiesofresearchinterestandtheDOEshouldexplorepossiblemechanisms
forcollaborativeprojectswithotheragenciessuchasNISTandDODinanyDOEAIforScience
initiative.
OthercountrieshavealsorecognizedthepotentialbenefitsofapplyingAL/MLtechnologiestoscience.ThesubcommitteebelievesthattherewouldbeabenefitintheDOEcollaboratingwith‘like-mindedinternationalpartners’onaspectsofanAIforScienceresearchagendathatarelikelytobeofmutualbenefit.
RecommendationsforDOE’sOfficeofScience
Creationofa10-yearAIforScienceInitiative
Inordertocreatetheworld-leadingAIsystemsandapplicationsneededtodrivescientificproductivityanddiscoveryinscienceandtechnologydramaticallybeyondthatachievablewithtraditionalscientificsupercomputing,werecommendthattheDOEOfficeofSciencestartaten-yearprogramtodevelopanambitiousAIforScienceinitiative,asrecommendedintherecentPCASTreport[9].Thisprogramshouldencompassfoundationalresearchintonew,science-awareAImethodologies,specificallydesignedforDOEmission-criticalchallenges,andAIsolutionsthatcanbedeployedinoperationalsettingsatleadingDOEresearchfacilities.Theinitiativeshouldprovideaclear,guidedroadmapfromresearchtodeployment.TheDOElaboratoriescanplayakeyrolehere,offeringleading-edgeexascalesupercomputersandlargeexperimentalfacilitiesgeneratingincreasinglylargescientificdatasets,aswellasprovidingcriticalexpertiseinmathematics,computerscience,andexperiencewithDOEmission-specificapplications.Nootheragencyhasthebreadth,criticalmass,orrecentlargeprojectmanagementexperiencetoundertakethiscross-disciplinaryAIforSciencechallenge.However,thereisaclearcaseforthebenefitsofcollaborationwithotheragenciesandothercountries,toleverageexistingexpertisetomaximumadvantage.Partnershipswithotherfundingagenciesandothercountriesarethereforestronglyencouraged.
StructureofanSCAIforScienceInitiative
ItisrecommendedthatthisAIforScienceinitiativebestructuredaroundfourmajorAIR&Dthemes:
AI-enabledapplications
AIalgorithmsandfoundationalresearch
AIsoftwareinfrastructure
NewhardwaretechnologiesforAI
Thesubcommitteebelievesthatthisten-yearAIforScienceinitiativeshouldbefundedatthesamescaleasthesuccessfulExascaleComputingInitiative(ECI)andExascaleComputingProject(ECP).Essentialforthesuccessofsuchaninitiativeisthattheworkofthesefourthemesmustbeclosely-coupledinamannersimilartothatusedintheECP,astheadvancesandimprovementsinoneareacaninformadvancesandimprovementsinotherareas.
Figure3illustratesanoverviewofapossibleroadmapforsuchanAIforScienceinitiative.AsfortheECIandECP,theroadmapforthisproposedAIforScienceinitiativeenvisagesaninitial‘incubation’researchphaseofcoordinatedprojectswithco-designcentersconnectingthefourmajorthemes.PartnershipsacrossallOfficeofSciencedomains,withparticipationfromuniversitiesandprivateindustry,wouldbeinitiatedearlyintheprogram.ThegoalofthisresearchphaseistospecifytheapplicationgrandchallengesandAI/MLtoolsandservicesrequiredasdeliverablesinthemorefocusedprojectR&DandDeploymentphases,wherebroad
engagementoftheDOEresearchcommunitybecomescritical.SincetheseappliedR&DandDeploymentphaseswillinevitablygeneratenewquestionsandchallenges,havingtheresearchphasecontinuingandoverlappingwiththeR&DandDeploymentphaseswillsignificantlyincreasethechancesofsuccessfortheAIforScienceProject.
AnInstrument-to-EdgeInitiative
ThesubcommitteebelievesthatASCR,inclosecooperationwithBESandwiththeotherscienceprogramsintheOfficeofScience,shouldworkwithscientists,users,andthebroadacademiccommunitytodefinerequirements,conductresearch,competitiveprocurementanddesignahighlyintegratedend-to-endsystemandsoftwarestackthatconnectsinstrumentsattheedgetotheneededAIcomputingresources.Integratingnationalandglobaldatasources(largescaleexperimentalfacilities,observationalnetworksterrestrial&space-based,etc.)posesuniqueopportunitiesandchallengesthatrequireaddressingfoundationalresearchinthecontextofleading-edgescientificexperiments.Integratedsystemsforacquiring,analyzing,transforming,storing,andmaintainingscientificresults,capturingprovenance,andcontributingbroadlyaccessedanalyticalworkflowswithinDOEsupportedcomputationalinfrastructurecouldbetransformative.Thereare,however,severechallengesthatwillneedtobeconfrontedintermsofprivacy,security,commerciallicensingofdata,andintegrateddataservices.
BuildingonASCR’sco-designexperienceinECP,applicationusers,softwareinfrastructuredevelopers,AI/MLresearchers,andLabandindustryhardwarespecialistsshouldbeencouragedtodefine,develop,andcontributetoacommonsoftwarestackforAI/MLEdgecomputingresourcesacrossthedifferentfacilities.ThesoftwareinfrastructureshouldsupportsomegenericservicesatthefacilitiesbutalsoallowtheeasycreationofspecializedAI-basedsoftwarepipelinesspecifictothefacilityandcapableofsupportingcouplingtoparticularinstrumentsinsomecases.
Training,focusing,andretentionofAI/MLworkforce
Industry,nationallaboratories,government,andbroadareasofacademicresearcharemakingmoreusethaneverbeforeofAI,ML,andsimulation-baseddecision-making.Thistrendisapparentacrossmanydomainssuchasenergy,manufacturing,finance,andtransportation.TheseareallareasinwhichAIisplayinganincreasinglysignificantrole,withmanymoreexamplesacrossscience,engineering,business,andgovernment.Researchandinnovation,bothinacademiaandintheprivatesector,areincreasinglydrivenbylarge-scalecomputationalapproachesusingAIandMLtechnologies.WiththissignificantandincreasedusecomesademandforaworkforceversedintechnologiesnecessaryforeffectiveandefficientAI/ML-basedcomputationalmodelingandsimulationandbigdataanalytics,aswellasthefundamentalsofAI/MLalgorithms.Graduateswiththeinterdisciplinaryexpertiseneededtodevelopand/orutilizeAItechniquesandmethodsinordertoadvancetheunderstandingofphysicalphenomenainaparticularscientific,engineering,orbusinessfieldandalsotosupportbetterdecision-makingareinhighdemand.
Astrongresearchprogramwillcruciallyrelyonacomplementaryeducationandskillscomponent,whichisasimportantasprovidingadequateinfrastructuresupport.AsemphasizedintheASCRECPTransitionreport[10],thisisalsoatimelyandimportantopportunitytofocusSCeffortstocreateamorediverseandinclusiveworkforce.Acontinuingsupplyofhigh-qualitycomputationalanddatascientistsavailableforworkatDOElaboratoriesisofvitalimportance.Inhighperformancemodelingandsimulation,forexample,theDOEComputationalScienceGraduateFellowship(CSGF)programhassuccessfullyprovidedsupportandguidancetosomeofthenation'sbestscientificgraduatestudents,andmanyofthesestudentsarenowemployedinDOElaboratories,privateindustry,andeducationalinstitutions.Weneedasimilarfellowshipprogramtomeettheincreasingrequirementforcomputationalanddatascientiststrainedtotackleexascaleanddata-intensivecomputingchallenges.Inaddition,theDOESCshouldexplorethepossibilitiesforcollaborationwiththeNSFabouttheprovisionofrelevanttrainingprogramsinAI/MLtechnologiesandtheirapplicationtoscience.
Inter-Agencycollaboration
AlthoughtheNSFhaslongbeenregardedastheleadagencyforfundamentalAIresearch,DOEisclearlytheleadagencyforresearchinvolvingtheintersectionof‘BigScience,BigData,andBigComputing.’DOEhasnotonlyestablishednationalandinternationalleadershipinHPCandsupercomputingbutisalsoaleaderintheapplicationofAI/MLtechnologiestotheverylargescientificdatasetsgeneratedbytheirlarge-scaleexperimentalfacilities.
WiththeNIH,theDOESChasasuccessfulcollaborationwiththeNationalCancerInstitute(NCI)intheCANDLEproject[11].DOEisnowdevelopinganMOUwithboththeNSFandNIHonaprogramofcollaborativeresearchinComputationalNeuroscience.Thesubcommittee,therefore,recommendsthattheSCexplorenewopportunitiestoworkwithbothNSFandNIHinareaswheretherewouldbeaclearbenefitforscientificprogressunderaDOE-ledAIforScienceinitiative.TheremayalsobeopportunitiestoworkwithotherUSfundingagencies,suchasNISTandDOD,inareasofmutualinterest.
Internationalcollaboration
Thereisaneedforbroad-based,coordinatedactionbylike-mindedinternationalpartnerstoharnesstheglobalscientificsoftwarecommunitytoaddressthetremendousopportunitiesindata-intensivesciencestemmingfromthehugeincreaseinscientificdatacollectionrates.ComputationalanddataanalyticalmethodsdrivenbyAI/MLarenowuniversallyacceptedasindispensableforfutureprogressinscienceandengineering.
InternationalleadershipinAIforScienceoverthecomingdecadewillhingeontherealizationofanintegratedsetofprogramsspanningthefourinterdependentareasnotedabove–AI-enabledapplications,AIalgorithmsandfoundationalresearch,AIsoftwareinfrastructure,andnewhardwa
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- ESG体系下的AI研究:多维投资增效防范伦理风险
- 冷链物流温控技术在冷链食品冷链配送中的质量保障体系优化与提升报告
- 2025年医药行业CRO模式下的供应链管理与物流优化报告
- 短视频平台内容版权纠纷处理与行业规范报告
- 绿色金融产品创新与绿色金融市场创新产品创新政策效应分析报告
- 民办教育机构2025年合规运营与品牌形象升级研究报告
- 文明校园广播稿(范本14篇)
- 快递行业Presentation:需求韧性持续、价格波动加剧
- 县级网格化监督管理制度
- 景区巡查安全管理制度
- 剪刀式登高车安全技术交底
- 职业生涯提升学习通超星期末考试答案章节答案2024年
- 规章制度之培训学校教学管理制度
- 部编人教版小学4四年级《道德与法治》下册全册教案
- 江苏省盐城市2023年七年级下册《数学》期末试卷与参考答案
- DB34T 4705-2024 职业健康检查工作规范
- 七年级数学下册 专题 不等式(组)中新定义运算&程序性问题(解析版)
- 《 大学生军事理论教程》全套教学课件
- 药物相互作用
- 无线电装接工考试:初级无线电装接工考试题库(题库版)
- 2024年高考真题和模拟题物理分类汇编专题08 电场(原卷版)
评论
0/150
提交评论