WIXLIB

Fujitsu World Tour2017shaping tomorrow with youFujitsu NorthAmerica TechnologyForum 2017Intel Inside .Powerful Productivity Outside.0Copyright 2017 FUJITSU

Connecting AI Technologies toReal-World NeedsDr. Fumihiro MaruyamaSenior ExpertArtificial Intelligence Research CenterFUJITSU LABORATORIES LTD.Intel Inside .Powerful Productivity Outside.1Copyright 2017 FUJITSU

Agenda1. FUJITSU AI Technology Brand “Zinrai”2. Cutting-edge AI technologies being developedby Fujitsu Laboratories3. Connecting AI Technologies to Real-World Needs2Copyright 2017 FUJITSU LABORATORIES LTD.

Fujitsu’s AI goalsTo create AI that collaborates withpeople and is human centricTo create AI that continuously evolvesTo create AI that can be incorporatedin products and services then deployed3Copyright 2017 FUJITSU LABORATORIES LTD.

FUJITSU AI Technology Brand MeaningBeing agile and intense Concept behind the brand nameDynamically realizing innovation in business and societyby rapidly supporting decision making and actions4Copyright 2017 FUJITSU LABORATORIES LTD.

A comprehensive framework for AIPeople / Businesses / SocietySensingActuationHuman Centric AI ZinraiSensing and n/staterecognitionrecognitionKnowledge Processing- Natural-language- tterndiscoveryPatterndiscoveryDecision and support& &-- InferenceInferencePlanningPlanning& &-- PredictionPredictionoptimizationoptimization& &- tionLearningee p l elearninga rn i n gDeepMachinelearningMachinelearningReinforcement on5Copyright 2017 FUJITSU LABORATORIES LTD.

Cutting-edge AI technologiesdeveloped by Fujitsu Laboratories6Copyright 2017 FUJITSU LABORATORIES LTD.

We extend the scope of Deep Learning Deep Learning Breakthrough technology of machine learning that extracts knowledge from data Automates feature extraction, a fundamental issue of machine learningFujitsu Laboratories’ scopeConventional scopeImageVehicleRecognitionVoiceTime-series dataTextPractical applicationCompany A: WhitePersonRecognitionHandwritingRecognitionGraph dataOriginal (2016/02)Original (2016/10)Leveraged bytopological data analysisBased on tensorexpressionmaleAge: 30Cloth: GrayBackpack: Pink7Copyright 2017 FUJITSU LABORATORIES LTD.

Classification of Time-Series Data It is difficult to accurately classify time-series data with extreme oscillations, such as that receivedfrom IoT devices. Fujitsu has developed deep learning technology that uses advanced chaos theory and topology toautomatically and accurately classify erratic time-series data.1Classification0.5Gyrodata0Accuracy: 85% (25% betterthan conventional �Gyro sensor data204060Time[s]80100*UC Irvine Machine Learning120Repository benchmark testMovement A within elevatorMovement B within elevatorTreadmill ATreadmill BStair-stepper AStair-stepper i数4030201000100200300400500600半径Step 1: Graphically representstime-series data using chaos theory(Attractor)Step 2: Quantifies diagrams usingtopology(Topological Data Analysis)Step 3: Learning and classificationusing a convolutional neural network To be incorporated into an IoT product in FY2016. Other time-series data includes log data from servers and sensor data obtained during sleep.8Copyright 2017 FUJITSU LABORATORIES LTD.

Classification of Time-Series DataDemo(デモビデオ)9Copyright 2017 FUJITSU LABORATORIES LTD.

Deep Learning with graph-structured dataPurposeTechnologyTo elicit new insights from graph-structured data that expressesconnections between people and thingsTechnology that allows existing deep learning to be applied tograph-structured data New tensor factorization technology converts graph-structured data to auniform expression Optimizing uniform expressions and neural network learningsimultaneouslyTensor-based uniform expression including whole �表現（テンソル表現）に変換Class ss BWider variety of features than handmade ackpropagation10Copyright 2017 FUJITSU LABORATORIES LTD.

Graph data appear in various business domainsFinanceBiologySocial NetworkSupply ChainGraph data that expressesrelationships betweenpeople and thingsSecurityTelecom/MobileHealthcare/Drug DiscoveryInternet11Copyright 2017 FUJITSU LABORATORIES LTD.

IT drug discovery: virtual screeningDemo Predicting a bond between a protein and chemical compounds12Copyright 2017 FUJITSU LABORATORIES LTD.

Outcomes in various domainsOutcomeDiscovered new insights and greatly improved learningaccuracyFintechMedical/DrugIoTAccount transactionCommunication logIT drug discoveryCredit risk predictionIntrusion detectionVirtual screening Detect high-riskloans Detect intrusions viaunexpected paths False positive rate:About 20% reduction False positive rate:About 30% reduction【Benchmark】 Open Data InstituteUK P2P lending market dataset【Benchmark】 DARPA IntrusionDetection Evaluation Data Set13 Discovered about 200new protein-compoundbond features Activity predictionaccuracy:About 10% increase【Benchmark】PubChem BioAssay data setCopyright 2017 FUJITSU LABORATORIES LTD.

Linked Open Data (LOD） Large-scale data storing and searching technology for LOD Developed and published LOD4ALL service based on fast graph databaseengine in collaboration with INSIGHTStore and search LODs on the webNew applications utilizing disparate dataOpen to the publicEvaCvasince January 2014http://evacva.netVisualizing characteristicsof the regionOpen to the publicsince December 2014http://lod4all.netPlatform for thecontestCollaborationwith INSIGHTConvert and integrate data into linked dataStoreLarge-scaleGraph DBAttach links14RDF StoreConversionLODTextCopyright 2017 FUJITSU LABORATORIES LTD.

Corporate analysis (HighStreet)PurposeCross-sectional analysis of various aspects of corporate activitiesfrom financial statements, stock prices, products and reputationsTechnology Integrating various data including financial and social network information Dynamic graph network analysis to detect similarities and anomaliesDiscover discrepanciesby integrating multiple reportsOutcomeVisualize relations between enterprisesvia people by integrating executive listsDiscovery of new relationships beyond performance comparisonand business transactions15Copyright 2017 FUJITSU LABORATORIES LTD.

AMED: Integrated Database for Clinical Genome Fujitsu and Kyoto University Start R&D for AI that Infers Clinical Interpretations of Genome and Gene Polymorphism Building a knowledge base that aggregates academic paper and public databases in the medical field, using LOD utilization platform Inferring clinical interpretations of Gene Variant using the unique machine learning technologyAggregate several databasesAI Curation@11 hospitalIntegrationAcademic PaperLOD Search ateTreatmentsInference Enginebased on Machine LearningVariant DBDiseaseClinVarGenePathogenicDrug DBDrugProteindrugBankOriginal DBFunctionLOD KnowledgeBaseClinical Info.CancerGenome Info.Refractory diseaseDementia Infection16IntegratedDatabasefor ClinicalGenomeetc.To achieveGenomeMedicineCopyright 2017 FUJITSU LABORATORIES LTD.

Collaborative Research with San Carlos HospitalNew healthcare solution designed to improve clinical decision-makingTechnology HIKARI Advanced Clinical Research Information SystemPurpose Knowledge discovery by combining medical records and public open data Anonymization of integrated data gathered from both clinical and non-clinical data sources for ensuring patient privacyLess than 5 seconds36,000patientsNon-clinicaldataHIKARICompare results30 patientsClinical historyOutcome5 senior clinicians(18-25 years of experience) 20 minutes per patientHIKARI gives clinicians more time and better information to help the patient17Copyright 2017 FUJITSU LABORATORIES LTD.

Interactive Q&AExtract necessary information through natural interaction with the userPurposeTechnologyAccurate information extraction by analyzing relations between wordsRealizing natural interaction by utilizing LODSpeech Understanding HokkaidoI want to go toHawaii.I will go to Hokkaido,where my sister lives,with a friend.Hokkaido.It is still cold there.Semantic ionGoIntention DetectionHawaiiWishGoing to Hawaii is a wish.The real destination is Hokkaido.The companion is not a sisterbut a friendMachineLearningNot destinationKnowledge-basedDialog GenerationGenerating answer candidateswith relevant informationSelecting an he customer highly appreciated the merit of the technology based on a trialLooking at deployment in customer-facing business and sales18Copyright 2017 FUJITSU LABORATORIES LTD.

Interactive Q&ADemo19Copyright 2017 FUJITSU LABORATORIES LTD.

Connecting AI Technologies toReal-World Needs20Copyright 2017 FUJITSU LABORATORIES LTD.

AI is relevant to every onAIFinanceContaminationTestProduct LineAutomationSymptomAI alogueRecommendation ailAISocialInfrastructureDrug ltDiagnosisAdvancedIndustrial AISceneSupportDrone edding SDN OptimizationAIAutomatedOperationFault RecoveryNetwork21AutomatedOperationFault RecoveryAIAutomatedOperationFault RecoveryServer/Data centerCopyright 2017 FUJITSU LABORATORIES LTD.

Issues and Approaches Issues Bridging gaps between (general-purpose) AI technologies and real-world needs Streamlining and accelerating of application processes Ongoing approaches PoCs/PoBs with customers Sharing knowledge about customers inside Fujitsu Providing application-oriented APIs in addition to function-oriented ones22Copyright 2017 FUJITSU LABORATORIES LTD.

Zinrai Platform Service 30 APIs will be available (some are from April 2017) Function-oriented APIs for basic use and application-oriented APIs for rapid development Proven through more than 300 AI-related business projects and field trialsSensing & RecognitionEmotionImageSpeechstateprocessing processing tedAPIsKnowledge singand discoveryTextanalysisKnowledgestructuring Classification InferencePatterndiscoveryImageSpeechrecognition to textEmotionrecognitionHandwritten- Speechtext recognitionsynthesisKnowledgeLine-of-sight Speechcomprehensionsearching ecognitionStaterecognitionTraffic imagerecognitionCompanyDomain specific informationsemantic search searchingDecision & SupportInferencePrediction & Conversationsrecommenand planning optimization &-dationsDialogPrediction generationQuestion-Optimization liveryplanningProduction and FAQ searchlogistics networkCommuni-cations botCopyright 2017 FUJITSU LABORATORIES LTD.

Fujitsu’s Advanced Technologies Applied to APIs・Image processing with theworld's fastest class ofdeep learning processingSensing & chprocessing・The world's largest scale LOD search・Data linking technology with highaccuracy・High-precision technology for contextcomprehension・Automatic, natural dialog generationKnowledge ProcessingEmotionand processingand discoveryDecision & Speechto andwrittentext ionInferenceand gnitionrecognitionPrediction &optimizationConversations& ationorientedAPIsTraffic icsemantic iveryplanningProduction andlogistics networkFAQSearchsearchFAQCommuni-cations botCopyright 2017 FUJITSU LABORATORIES LTD.

Intel Inside .Powerful Productivity Outside.25Copyright 2017 FUJITSU

from financial statements, stock prices, products and reputations Discovery of new relationships beyond performance comparison and business transactions Integrating various data including financial and social network information Dynamic graph network analysis to detect similarities and anomalies Discover discrepancies