1、Managing Machine LearningProductsWhere Grab is at now in Southeast AsiaOver 6 millionrides daily217 Cities inVIETNAM8 CountriesMYANMARTHAILANDPHILIPPINESOver 100 millionCAMBODIAmobile downloadsMALAYSIASINGAPOREOver 6.6 millionmicro-entrepreneursenabledINDONESIAFIVE YEARS IN SINGAPOREMobility service
2、sto suit differentneedsDeveloping ourtech workforceGoing cashlesswith taxis andmerchantsImprovinglivelihoodsthrough one appWe hear the FUTURE that Singaporeans & Southeast Asianconsumers wantSource: Word cloud of Entries from 500+consumers across Southeast Asia whenasked the question:“What would you
3、 like to see in the Future ofGrab”Services at your fingertips in oneplaceAccess online services via appsOnly app that offers key dailyservices with one tapOnly app that you can use allacross Southeast AsiaThe one EVERYDAY APP for your daily essential needsGrabNowGrabCoachYour EverydayJustGrabApp.Gra
4、bShuttleGrabCycle6 Years Ago4 Years Ago3 Years Ago2 Years Ago1 Year AgoThis YearDABCThe one app to travel & pay Plan your trip the way you like it (budget,speed and comfort) Book your whole journey in one click Pay just onceIntroductionOur transportation solution will become increasingly intelligent
5、DigitalIntelligentPersonalConnectedIntroductionProvide intelligent servicesAllocationRecommendationPositionCommunicationMachine LearningIntroducing ML productsRecognitionClassificationRecommendationPersonalizationOptimizationDetectionPredictionAutomationMACHINE LEARNING | RECOGNITIONMachine learning
6、 vs human learningHuman learningMachine learningWhat makes us human?MACHINE LEARNING | RECOGNITIONYour algorithm is as good as your dataExample: Computer visionVariationIlluminationPoseHow much noise can the system take before it falls apart?MACHINE LEARNING | RECOGNITION90% of ML is about dataData
7、cleaning1234Speed of learningLabelling costLong tail data and edge casesMachine learning is still largely taught by humans as teachersMACHINE LEARNING | RECOGNITIONCreate systems that operate in the real worldOrganized databasesData brokersData sharingSimulationEmbedded processingCrowdsourcingthis i
8、s why product managers have a jobATTORNEY-CLIENT PRIVILEGED|Why ML?Example: Support ticket categorization|Trade offs in machine learningExample: Job Recommendations|What is personalizationGrab Car Pick UpSocial?Tall, Half-Caff,Soy Latte At120 DegreesCustomized travelneedsML based,personal stylingWor
9、kplaceergonomic settings|Why personalizeRiders are.Our ObjectivesCommuters Personal and relevant: helpuser find what they are lookingfor Consistent experience:Remove friction points Value adding: Create morevalue to the marketplace Delight and excite: Build longterm loyaltyCasual ridersSingle car HH
10、Global Trotters|How to personalize|Optimize user experience at personal level|Machine reflects biases of peopleExample: Feeds|The importance of objective functionExample: traffic optimizationMACHINE LEARNING | OPTIMIZATIONDefining a good objective function is difficultHuman strategyObjectiveTimeRank
11、ing# of turbo pointsMachine StrategyMACHINE LEARNING | DETECTION + PREDICTIONCommunication between machine and humansInputsOutputMACHINE LEARNING | AUTOMATIONSometimes human-machine collaboration is even harder than human-humancollaboration.Objective function is how you build alignment with machineMACHINE LEARNING | AUTOMATION.resulting in serious consequencesberlingen mid-air collisionTCAS vs Traffic Controller decisionMACHINE LEARNING | AUTOMATIONPhilosophy and ethicsClassic ethical dilemma:the trolley problemThe AV challenge:Ethics and legal questionsQ&A
