Slide 1

Slide 1 text

Quick overview of AWS AI/ML services Shoko Utsunomiya, Ph.D AWS Machine Learning Solutions Architect

Slide 2

Slide 2 text

© 2018, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark About me Shoko Utsunomiya, Ph.D • Machine Learning Solutions Architect • Providing solutions of ML on AWS • My favorite AWS service • Amazon SageMaker • Worked for Automotive OEM as engineer of Autonomous driving system • Associate Professor of National Institute of Informatics (2013- 2017)

Slide 3

Slide 3 text

© 2018, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Machine Learning at Amazon: A long heritage Personalized Recommendations Fulfillment automation & Inventory Management Drones Voice driven Interactions Inventing entirely new Customer experiences

Slide 4

Slide 4 text

© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark "*.-VTF DBTFJOWBSJPVTJOEVTUSJFT • $POUFOUTHFOFSBUJPO • 1SPNPUJPONBSLFUJOH • $PQZSJHIUJOGSJOHFNFOU JOTQFDUJPO • "VUPUBHHJOHPG DPOUFOUT • "VUPNBUJDDBQUJPO DSFBUJPO • 'BJMVSFEFUFDUJPOPG QSPEVDUJPOMJOF • 7JTVBMJOTQFDUJPOPG GJOJTIFEQSPEVDUT • *NQSPWFDVTUPNFS TFSWJDFUISPVHIWPJDF TFSWJDFBOEDIBUCPU • $BMMDFOUFS PQUJNJ[BUJPO • 4FMGESJWJOHUFDIOJRVF • &DPNNFSDF • 1SPEVDU SFDPNNFOEBUJPO • 1FSTPOBMJ[F • $SFEJUFWBMVBUJPO • "EWFSUJTJOH NBSLFUJOH • 6OBUUFOEFETUPSF • 1BUJFOUTIFBMUI DPOEJUJPOCBTFE PODMJOJDBMEBUB • 1SFEJDUIPTQJUBM TUBZBOESF IPTQJUBMJ[BUJPO • %SVHEJTDPWFSZ • .FEJDBMJNBHJOH EJBHOPTJT • &BSMZEJBHOPTJT • &YBNJOBUJPO • 'SBVEEFUFDUJPO • .BSLFUJOH'VUVSF 'PSFDBTU • *OWFTUNFOUQPSUGPMJP NBOBHFNFOU • /FXT"OBMZTJT • (FPTQBUJBMJNBHFBOBMZTJT • $IBUCPUQFSTPOBMBEWJTPS .FEJB FOUFSUBJONFOU %JTUSJCVUJPO 3FUBJM )FBMUIDBSF -JGFTDJFODF 'JOBODF 4FSWJDF USBOTBDUJPO .BOVGBDUVSF "VUPNPUJWF

Slide 5

Slide 5 text

© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Tens of thousands of customers running Machine Learning on AWS

Slide 6

Slide 6 text

© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark 5 merits for Machine Learning on AWS Use as much CPU / GPU as needed when needed System extensibility Improved agility for model building and service deployment Pay-as-you-go Cost optimization A lot of AWS Service / function Deploy worldwide instantly

Slide 7

Slide 7 text

© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Continuous ML development %BUB BDRVJTJUJPO Pre- processing Model development Model evaluation Model translation Deployment Monitoring, evaluation

Slide 8

Slide 8 text

© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Continuous ML development %BUB BDRVJTJUJPO Pre- processing Model development Model evaluation Model translation Deployment Monitoring, evaluation Very important to run this cycle at high speed to improve your AI...

Slide 9

Slide 9 text

© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark AWS Machine Learning Stack App developers with little knowledge of ML ML developers and data scientists ML researchers and academics Amazon SageMaker Ground Truth Algorithms Notebooks Marketplace Unsupervised Learning Supervised Learning Reinforcement Learning Optimization (Neo) Training Hosting Deployment Frameworks Interfaces Infrastructure Amazon Rekognition Image Amazon Polly Amazon Transcribe Amazon Translate Amazon Comprehend Comprehend medical Amazon Lex Amazon Rekognition Video Vision Speech Language Chatbots Amazon Forecast Forecasting Amazon Textract Amazon Personalize Recomme ndations Amazon EC2 P3 & P3DN Amazon EC2 C5 FPGAs AWS Greengrass Amazon Elastic Inference Amazon Inferentia Labeling Model development Training Hosting ML SERVICES ML FRAMEWORKS & INFRASTRUCTURE AI SERVICES New New New New New New New New New New New

Slide 10

Slide 10 text

© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark AWS Machine Learning Stack App developers with little knowledge of ML Amazon Rekognition Image Amazon Polly Amazon Transcribe Amazon Translate Amazon Comprehend Comprehend medical Amazon Lex Amazon Rekognition Video Vision Speech Language Chatbots Amazon Forecast Forecasting Amazon Textract Amazon Personalize Recomme ndations AI SERVICES New New New New Time series analysis Forecast Recommendation Personalize

Slide 11

Slide 11 text

© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark AWS Machine Learning Stack ML developers and data scientists ML researchers and academics Amazon SageMaker Ground Truth Algorithms Notebooks Marketplace Unsupervised Learning Supervised Learning Reinforcement Learning Optimization (Neo) Training Hosting Deployment Frameworks Interfaces Infrastructure Amazon EC2 P3 & P3DN Amazon EC2 C5 FPGAs AWS Greengrass Amazon Elastic Inference Amazon Inferentia Labeling Model development Training Hosting ML SERVICES ML FRAMEWORKS & INFRASTRUCTURE New New New New New New New

Slide 12

Slide 12 text

© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark “Undifferentiated Heavy Lifting” in Machine Learning &OWJSPONFOU • &TUJNBUFESFTPVSDFTOFFEFEBOEEFDJTJPOUPCVZ • *UUBLFTUJNFBOEFGGPSUUPQSPDVSFJOUIFGJSTUQMBDF • %FWFMPQNFOUUFBNOFFETBVOJGPSNEFWFMPQNFOUFOWJSPONFOUNBOZUJNFT • 'SBNFXPSLJOTUBMMBUJPO WFSTJPODPOUSPM $POTUSVDUJPOBOEMFBSOJOHPGNBDIJOFMFBSOJOHNPEFM • 6OFWFOIBSEXBSFFOWJSPONFOUTVDIBT$16(16 • 4DBMBCMF3FTPVSDFTBOE%JTUSJCVUFE$PNNVOJDBUJPO&OWJSPONFOUGPS%JTUSJCVUFE -FBSOJOH 0QFSBUJPO • *OGFSFODFFOWJSPONFOUQSFQBSBUJPOBOENPEFMIPTUJOH • "EJGGFSFOUTLJMMTFUUIBONBDIJOFMFBSOJOHJTSFRVJSFE Photo by Victor Freitas on Unsplash

Slide 13

Slide 13 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Amazon SageMaker 1 2 3

Slide 14

Slide 14 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark 1 2 3 Amazon SageMaker: Build, Train, and Deploy ML Models at Scale

Slide 15

Slide 15 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark © 2018, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Amazon SageMaker Ground Truth Build highly accurate training datasets and reduce data labeling costs by up to 70% using machine learning

Slide 16

Slide 16 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Successful models require high-quality data

Slide 17

Slide 17 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Amazon SageMaker Ground Truth demo

Slide 18

Slide 18 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Amazon SageMaker: Build, Train, and Deploy ML Models at Scale

Slide 19

Slide 19 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark - - - - . / . . - - .- . / . - . / - / / - . .. / / . / / . / . - - - ) - - ( -

Slide 20

Slide 20 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Model name Supervised? Algorithm description Descriptions Linear Learner Supervised Linear regression regression, classification XGBoost Supervised XGBoost (eXtreme Gradient Boosting) regression, classification PCA Unsupervised Principal Component Analysis Dimensions reduction k-means Unsupervised K-means Clustering k-NN Supervised K-nearest neighbour Clustering Factorization Machines Supervised Matrix factorization Recommend, regression, classification Random Cut Forest Unsupervised robust random cut tree Anomaly detection in time series LDA (Latent Dirichlet Allocation) Supervised Generative statistical model Topic modeling _.BDIJOFMFBSOJOHBMHPSJUINT_ SageMaker built-in algorithms ˟ -%"ךؔٔآشٕכ侄䌌ז׃ https://docs.aws.amazon.com/sagemaker/latest/dg/algos.html

Slide 21

Slide 21 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Field Model Supervised algorithms descriptions Image Processing Image classification supervised ResNet Image classification Object Detection supervised SSD (Single Shot multibox Detector) Detect multiple images with labeled bounding box Semantic Segmentation supervised FCN, PSP, DeepLabV3 (ResNet50, ResNet101) Detection of object region in image in pixel unit NLP seq2seq supervised Deep LSTM Text summaryvoice recognition Neural Topic Model unsupervised NTM, LDA Structuring text data Blazing text unsupervised Word2Vec Sentiment analysis supervised Text Classification Text mining Object2Vec supervised Genelized Word2Vec Classification, recommendation 5JNFTFSJFT DeepAR Forecasting supervised Autoregressive RNN Stochastic time series forecasting "OPNBMZ EFUFDUJPO IP Insights unsupervised NN Detection of malicious IP address _%FFQMFBSOJOHNPEFMT_ SageMaker build-in algorithms https://docs.aws.amazon.com/sagemaker/latest/dg/algos.html

Slide 22

Slide 22 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark AWS Marketplace for Machine Learning ML algorithms and models available instantly Subscribe in a single click Available in Amazon SageMaker K E Y F E A T U R E S Automatic labeling via machine learning IP protection Automated billing and metering Browse or search AWS Marketplace S E L L E R S Broad selection of paid, free, and open-source algorithms and models Data protection Discoverable on your AWS bill B U Y E R S

Slide 23

Slide 23 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark • More than 200 algorithms for retail, media, etc. are available on AWS Marketplace for ML Algorithm Purchaser: • Training job and inference endpoint (also batch inference job ok) with Amazon SageMaker Algorithm Exhibitor: • Upload the model by concealing the contents of the model AWS Marketplace for Machine Learning

Slide 24

Slide 24 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Amazon SageMaker: Build, Train, and Deploy ML Models at Scale

Slide 25

Slide 25 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Amazon SageMaker: Build, Train, and Deploy ML Models at Scale

Slide 26

Slide 26 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Amazon SageMaker: Build, Train, and Deploy ML Models at Scale

Slide 27

Slide 27 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark New machine learning capabilities in Amazon SageMaker to build, train and deploy with reinforcement learning

Slide 28

Slide 28 text

© 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Supervised Learning (ASR, computer vision) Unsupervised Learning (Anomaly detection, identifying text topics) Introducing Reinforcement Learning Reinforcement Learning (RL)

Slide 29

Slide 29 text

© 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Reinforcement learning Achieve outcomes, not decisions Robotics Industrial controls Natural language systems Games

Slide 30

Slide 30 text

© 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. How does RL work?

Slide 31

Slide 31 text

© 2018, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Amazon SageMaker Reinforcement Learning • 3-PO4BHF.BLFSJOUFHSBUFEXJUITJNVMBUJPOFOWJSPONFOUPGHBNJOHBOESPCPUJDT • 4VQQPSUT$PBDIBOE3-3BZBTSFJOGPSDFNFOUMFBSOJOHUPPMLJU • "843PCP.BLFS 3PCPTDIPPM FUDBWBJMBCMFWJB04"T0QFO"* (ZNJOUFSGBDF • %JTUSJCVUFEMFBSOJOHFOWJSPONFOU EJTUSJCVUJPOPGUSBJOJOHBOEFOWTJNVMBUJPO Container for Agent Container for Agent Container for environment Container for environment &OWJSPONFOU TJNVMBUPS RoboMaker, … OpenAI gym "HFOU RL tool kit Coach, RLLib Redis Action by policy Observation, reward Train policy

Slide 32

Slide 32 text

© 2018, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark SageMaker Reinforcement configuration ~ DeepRacer with RoboMaker ~ 0CKFDUJWFGVODUJPO4FMGESJWJOHBDDPSEJOH UPMBOFDFOUFS &OWJSPONFOU%ESJWJOHTJNVMBUPSIPTUFE BU"843PCP.BLFS 4UBUF107JNBHFTPGESJWFSTFOUUISPVHI DBNFSBPODBS "DUJPOTEJGGFSFOUTUFFSJOHBOETQFFE DPOGJHVSBCMF 3FXBSEQMVTXIFODPNFDMPTFSUP DFOUFSMBSHFQFOBMUZJNQPTFECZHFUUJOHPVUPGUIF USBDL1PTTJCMFUPTFUEFUBJMTTVDIBTBEEJOHB TUFFSJOHQFOBMUZ

Slide 33

Slide 33 text

© 2018, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark "84%FFQ3BDFS :PVDBOGJOEJO"NB[PODPN BU 4QFDJGJDBUJPO • TDBMFSBEJPDPOUSPMDBS • *OUFM"UPNQSPDFTTPS • .QJYFM QDBNFSB • 8J'J BD • #BUUFSZGPSDPNQVUF I BOE GPSNPUPS N • 6CVOUV-54 304 3PCPU 0QFSBUJOH4ZTUFN 0QFO7JOP https://aws.amazon.com/deepracer/

Slide 34

Slide 34 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark Amazon SageMaker: Build, Train, and Deploy ML Models at Scale

Slide 35

Slide 35 text

© 2019 Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark • • • • • • ,