Track 6 Session 3_如何藉由 AWS AI 和機器學習平台搭建多功能的 AI 解決方案.pptx
•Télécharger en tant que PPTX, PDF•
0 j'aime•825 vues
Recommandé
Recommandé
Contenu connexe
Tendances
Tendances (20)
Similaire à Track 6 Session 3_如何藉由 AWS AI 和機器學習平台搭建多功能的 AI 解決方案.pptx
Similaire à Track 6 Session 3_如何藉由 AWS AI 和機器學習平台搭建多功能的 AI 解決方案.pptx (20)
Plus de Amazon Web Services
Plus de Amazon Web Services (20)
Track 6 Session 3_如何藉由 AWS AI 和機器學習平台搭建多功能的 AI 解決方案.pptx
- 1. © 2020, Amazon Web Services, Inc. or its affiliates. All rights reserved. 如何藉由 AWSAI 和機器學習平台搭建 多功能的AI 解決方案 T r a c k 6 | S e s s i o n 3 Ivan Cheng (鄭志帆) Manager, Solutions Architecture Amazon Web Services
- 2. $50 B By 2021, global spending on AI and cognitive technologies will exceed $50 billion Source: IDC, 2018 InnovationDecision making Customer experience Business operations Competitive advantage Centerpiecefor digitaltransformation
- 3. Our mission at AWS Put machine learning (ML) in the hands of every developer
- 4. 200+ new features and services launched in the last year Broadest and deepest set of AI and ML services Up to 70% cost reduction in data labeling Up to 10x faster performance Up to 75% lower inference cost Reduced deployment time Accelerate your adoption of ML with Amazon SageMaker Built on the most comprehensive cloud platform optimized for ML AWS named as a leader in Gartner’s Infrastructure as a Service (IaaS) Magic Quadrant for the 9th consecutive year WhyAWS for ML?
- 5. 10,000+ customers | More ML happens on AWS than anywhere else MLishappeningincompaniesofeverysizeandineveryindustry
- 6. Technology Bringing AI into your digital transformation requires a new stack that makes it easier to put ML to work
- 7. Vision Speech Text Search Chatbots Personalization Forecasting Fraud Development Contact centers Ground Truth Augmented AI ML Marketplace Neo Built-in algorithms Notebooks Experiments Model training & tuning Debugger Autopilot Model hosting Model Monitor Deep learning AMIs & containers GPUs & CPUs Amazon Elastic Inference AWS Inferentia FPGA Amazon Rekognition Amazon Polly Amazon Transcribe +Medical Amazon Comprehend +Medical Amazon Translate Amazon Lex Amazon Personalize Amazon Forecast Amazon Fraud Detector Amazon CodeGuru AI services ML services ML frameworks & infrastructure Amazon Textract Amazon Kendra Contact Lens For Amazon Connect Amazon SageMaker Studio IDE NEW NEW! NEW! NEW! NEW! NEW! NEW! NEW! NEW! NEW! NEW! Amazon SageMaker TheAWS MLstack Broadest and most complete set of ML capabilities
- 8. Vision Speech Text Search Chatbots Personalization Forecasting Fraud Development Contact centers Ground Truth Augmented AI ML Marketplace Neo Built-in algorithms Notebooks Experiments Model training & tuning Debugger Autopilot Model hosting Model Monitor Deep learning AMIs & containers GPUs & CPUs Amazon Elastic Inference AWS Inferentia FPGA Amazon Rekognition Amazon Polly Amazon Transcribe +Medical Amazon Comprehend +Medical Amazon Translate Amazon Lex Amazon Personalize Amazon Forecast Amazon Fraud Detector Amazon CodeGuru AI services ML services ML frameworks & infrastructure Amazon Textract Amazon Kendra Contact Lens For Amazon Connect Amazon SageMaker Studio IDE NEW NEW! NEW! NEW! NEW! NEW! NEW! NEW! NEW! NEW! NEW! Amazon SageMaker TheAWS MLstack Broadest and most complete set of ML capabilities
- 9. TheMLworkflowis iterativeand complex Bringing ML to all developers Prepare Build Train & tune Deploy & manage 101011010 010101010 000011110 Collect and prepare training data Choose or build an ML algorithm Set up and manage environments for training Train, debug, and tune models Deploy model in production Manage training runs Monitor models
- 10. Prepare Build Train & tune Deploy & manage 101011010 010101010 000011110 AmazonSageMakerhelps you build,train,and deploy models Collect and prepare training data Choose or build an ML algorithm Set up and manage environments for training Train, debug, and tune models Deploy model in production Manage training runs Monitor models
- 11. Prepare Build Train & tune Deploy & manage 101011010 010101010 000011110 Fully managed data processing jobs and data labeling workflows Web-based IDE for ML Automatically build and train models Collect and prepare training data Choose or build an ML algorithm Set up and manage environments for training Train, debug, and tune models Deploy model in production Manage training runs Monitor models AmazonSageMakerhelps you build,train,and deploy models
- 12. Collect and prepare training data Choose or build an ML algorithm Set up and manage environments for training Train, debug, and tune models Deploy model in production Manage training runs Monitor models Prepare Build Train & Tune Deploy & Manage 101011010 010101010 000011110 Fully managed data processing jobs and data labeling workflows Classification • Linear Learner • XGBoost • KNN Working with text • BlazingText • Supervised • Unsupervised Computer vision • Image classification • Object detection • Semantic segmentation Recommendation • Factorization machines Anomaly detection • Random Cut Forests • IP insights Regression • Linear Learner • XGBoost • KNN Topic modeling • LDA • NTM Forecasting • DeepAR forecasting Feature reduction • PCA • Object2Vec AmazonSageMakerhelps you build,train,and deploy models One-click notebooks, built-in algorithms and models
- 13. Prepare Build Train & tune Deploy & manage 101011010 010101010 000011110 One-click training AmazonSageMakerhelps you build,train,and deploy models Collect and prepare training data Choose or build an ML algorithm Set up and manage environments for training Train, debug, and tune models Deploy model in production Manage training runs Monitor models Web-based IDE for ML Automatically build and train models Fully managed data processing jobs and data labeling workflows One-click notebooks, built-in algorithms and models
- 14. Prepare Build Train & tune Deploy & manage 101011010 010101010 000011110 One-click training Debugging and optimization AmazonSageMakerhelps you build,train,and deploy models Collect and prepare training data Choose or build an ML algorithm Set up and manage environments for training Train, debug, and tune models Deploy model in production Manage training runs Monitor models Web-based IDE for ML Automatically build and train models Fully managed data processing jobs and data labeling workflows One-click notebooks, built-in algorithms and models
- 15. Prepare Build Train & tune Deploy & manage 101011010 010101010 000011110 One-click training Debugging and optimization Visually track and compare experiments Web-based IDE for ML Automatically build and train models Collect and prepare training data Choose or build an ML algorithm Set up and manage environments for training Train, debug, and tune models Deploy model in production Manage training runs Monitor models AmazonSageMakerhelps you build,train,and deploy models Fully managed data processing jobs and data labeling workflows One-click notebooks, built-in algorithms and models
- 16. Prepare Build Train & tune Deploy & manage 101011010 010101010 000011110 One-click training Debugging and optimization Visually track and compare experiments One-click deployment and automatic scaling AmazonSageMakerhelps you build,train,and deploy models Web-based IDE for ML Automatically build and train models Collect and prepare training data Choose or build an ML algorithm Set up and manage environments for training Train, debug, and tune models Deploy model in production Manage training runs Monitor models Fully managed data processing jobs and data labeling workflows One-click notebooks, built-in algorithms and models
- 17. Prepare Build Train & tune Deploy & manage 101011010 010101010 000011110 One-click training Debugging and optimization Visually track and compare experiments Automatically spot concept drift Web-based IDE for ML Automatically build and train models Collect and prepare training data Choose or build an ML algorithm Set up and manage environments for training Train, debug, and tune models Deploy model in production Manage training runs Monitor models AmazonSageMakerhelps you build,train,and deploy models One-click deployment and automatic scaling Fully managed data processing jobs and data labeling workflows One-click notebooks, built-in algorithms and models
- 18. Prepare Build Train & tune Deploy & manage 101011010 010101010 000011110 One-click training Debugging and optimization Visually track and compare experiments Add human review of predictions Web-based IDE for ML Automatically build and train models Collect and prepare training data Choose or build an ML algorithm Set up and manage environments for training Train, debug, and tune models Deploy model in production Manage training runs Monitor models AmazonSageMakerhelps you build,train,and deploy models One-click deployment and automatic scaling Fully managed data processing jobs and data labeling workflows One-click notebooks, built-in algorithms and models Automatically spot concept drift
- 19. Prepare Build Train & tune Deploy & manage 101011010 010101010 000011110 One-click training Debugging and optimization Visually track and compare experiments Add human review of predictions Fully managed with automatic scaling for 75% less Web-based IDE for ML Automatically build and train models Collect and prepare training data Choose or build an ML algorithm Set up and manage environments for training Train, debug, and tune models Deploy model in production Manage training runs Monitor models AmazonSageMakerhelps you build,train,and deploy models One-click deployment and automatic scaling Fully managed data processing jobs and data labeling workflows One-click notebooks, built-in algorithms and models Automatically spot concept drift
- 20. Vision Speech Text Search Chatbots Personalization Forecasting Fraud Development Contact centers Ground Truth Augmented AI ML Marketplace Neo Built-in algorithms Notebooks Experiments Model training & tuning Debugger Autopilot Model hosting Model Monitor Deep learning AMIs & containers GPUs & CPUs Amazon Elastic Inference AWS Inferentia FPGA Amazon Rekognition Amazon Polly Amazon Transcribe +Medical Amazon Comprehend +Medical Amazon Translate Amazon Lex Amazon Personalize Amazon Forecast Amazon Fraud Detector Amazon CodeGuru AI services ML services ML frameworks & infrastructure Amazon Textract Amazon Kendra Contact Lens For Amazon Connect Amazon SageMaker Studio IDE NEW NEW! NEW! NEW! NEW! NEW! NEW! NEW! NEW! NEW! NEW! Amazon SageMaker TheAWS MLstack Broadest and most complete set of ML capabilities
- 21. One interface to review past evaluations and detection logic Pre-built fraud detection model templates Models learn from past attempts to defraud Amazon Automatic creation of custom fraud detection models Amazon SageMaker integration New: Automate online fraud detection with Amazon Fraud Detector
- 22. Amazon Personalize • Fully managed service for generating personalized recommendations • Uses the same ML technology that is being used at Amazon.com • Generates highly relevant recommendations using deep learning techniques • Does not require ML expertise to use • Builds custom and private ML models using your own data Based on the technology that powers personalization at Amazon.com
- 23. Customized personalization API Amazon Personalize User events and interactions User metadata (demographics – optional) Item metadata (catalog information – optional) Fully managed by Amazon Personalize Amazon Personalize Behind the scenes Inspect data Select hyper- parameters Train models Optimize models Host models Real-time feature store Identify features
- 24. Key features Context-aware recommendations Automated ML Bring existing algorithms from Amazon SageMaker Continuous learning to improve performance Amazon Personalize Improve customer experiences with personalization and recommendations
- 25. Vision Speech Text Search Chatbots Personalization Forecasting Fraud Development Contact centers Ground Truth Augmented AI ML Marketplace Neo Built-in algorithms Notebooks Experiments Model training & tuning Debugger Autopilot Model hosting Model Monitor Deep learning AMIs & containers GPUs & CPUs Amazon Elastic Inference AWS Inferentia FPGA Amazon Rekognition Amazon Polly Amazon Transcribe +Medical Amazon Comprehend +Medical Amazon Translate Amazon Lex Amazon Personalize Amazon Forecast Amazon Fraud Detector Amazon CodeGuru AI services ML services ML frameworks & infrastructure Amazon Textract Amazon Kendra Contact Lens For Amazon Connect Amazon SageMaker Studio IDE NEW NEW! NEW! NEW! NEW! NEW! NEW! NEW! NEW! NEW! NEW! Amazon SageMaker TheAWS MLstack Broadest and most complete set of ML capabilities
- 26. Culture Setting your organization up for success
- 27. Enabling thenextML developers Partnerships with MOOCs ML educational devices AWS DeepLens Deep learning Training and certification AWS ML training and certification AWS DeepComposer Generative AI AWS DeepRacer Reinforcement learning
- 28. The world’s first machine learning-enabled musical keyboard for developers
- 29. Choose from jazz, rock, pop, or classical, or build your own custom genre model in Amazon SageMaker Input a melody by connecting the AWS DeepComposer keyboard Publish your tracks to SoundCloud from the console; export MIDI files to your favorite DAW Creativemeets generative
- 31. powered by
- 32. LearnmachinelearningwithAWSTrainingandCertification Learn at your convenience with 65+ free digital courses, or register for a live instructor-led class featuring hands-on labs and opportunities for practical application Explore tailored machine learning (ML) paths for business decision makers, data platform engineers, data scientists, and developers Take the AWS Certified Machine Learning – Specialty exam to validate expertise in building, training, tuning, and deploying ML models Visit the ML learning paths at https://aws.training/ML Resources created by the experts at AWS to help you build and validate machine learning skills
- 33. Thank you! © 2020, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Notes de l'éditeur
- First call deck for CXO level introduction to Amazon AI.
- 最近很多大規模的宣傳在 AI/ML 正對 digitial transformation 50年前已經有 ML technology 最近最大的改變是因為 Cloud Computing - ML technology 更容易使用和取得 IDC: by 2021, global spending on AI and cognitive technologies will exceed $50 billion. 在幾個 use case Customer experience conversational interface （對話方式, chatbot) personalization and recommendation Business operation and Decision Making 透過 Forecast - improve demand planning 的正確度 Innovation & Competitive Advantage Supply chain management 的 automation Healthcare - reactive to predictive care
- Simplify Machine Learning Easy for developer to build smart application
- Broadest & Deepest set of AI/ML services 在去年 AWS launched 200+ new features & services Developer 透過這些技術去開發簡單/複雜的 AI applications 透過 Amazon Sagemaker 加速 ML 的 adoption Sagemaker: 讓你可以更簡單的 build, train 和 deploy … Most comprehensive (全方位）的 cloud platform Not only ML, but also security, analytics, compute services, etc. should be put into consideration AWS 有全方位的 technology Gartner ..
- 10,000+ 客戶在 AWS 上面跑 ML - More than anywhere else From startup to enterprise Mention: Nascar, Intuit, NFL, etc.
- Technology 正對 enterprise 利用 AI 這件事情 - 我們可以用2個面向: Technology — 解決的問題不一樣，需要的工具耶不一樣
- AWS ML Stack AWS provide 提供完整的 ML 技術 不同的客戶角色 - 利用不同的 layer ML fw & infrastructure For expert ML practitioner 開發 algorithm ，maintain 自己的 infrastructure，等等 提供常用的: tensor flow, MXnet, pytorch ML practitioner 通常會用多個 framework 解決問題 根據 Research 85% tensor flow / 83% pytorch running on AWS Provide deep learning AMI and container Quickly launch Amazon EC2 instances pre-installed with popular deep learning frameworks to train AI models Amazon Elastic Inference Attach low cost GPU 加速功能附加到 Amazon EC2 和 Sagemaker instances 或 Amazon ECS 任務，最多可節省 75% cost.
- ML Services Not many ML practitioner Want to make it easy for data scientist build, train, and deploy Amazon SageMaker
- ML Workflow is complex ML development complex no integrated tool for end to end flow Prepare data —> Build / need to write your own algorithm / choosing the right algorithm Train & Tune Prepare and manage training environment Model tuning Depoy & Manage Host inference How to scale production environment How to monitor models Etc.
- Amazon Sagemaker help you on this Let’s see how
- Prepare data collection and prep is important in ML Build and Manage training data is often complicated and time-consuming Amazon SageMaker Ground Truth helps you build and manage training datasets. your own labelers / Amazon Mechanical Turk / Vendor from Marketplace continuously learns from labels done by humans to make automatic annotations
- Build Amazon SageMaker Notebooks Managed Jupyter Notebook compute resources are fully elastic – scale up / down one-click sharing of notebooks. All code dependencies are automatically captured, so you can easily collaborate with others. Algorithm pre-build algorithms Choose from pre-built: supervised algorithms such as XGBoost and linear/logistic regression or classification, unsupervised learning such as with k-means clustering and principal component analysis (PCA) You can also shop for algorithms, models, and data in the AWS Marketplace. Bring your own algorithm - docker container
- Train & Tune One Click training It’s easy to train model in SageMaker Specify data location in Amazon S3 One click - SageMaker help you to setup and manage environment compute cluster perform training output to s3 Tear down cluster optimized data streams from Amazon S3 to SageMaker. can specify data distribution based on your algorithm: if you’d like all of the data to be sent to each node in the cluster, or Amazon SageMaker to manage the distribution of data
- Debugging and Optimization 在開發過程中遇到 error 需要 debug / tune Amazon SageMaker Debugger Full visibility of ML training process automatically detects and alerts developers to occurring errors make it easy for developer to debug complete insights 資訊 into the debugging process.
- Compare Experiments What if you have multiple model to experiment that you want to compare Amazon SageMaker Experiments – track and compare models Evaluate training experiement record training artifacts including datasets, algorithms, parameters system automatically organizes, ranks experiments based on chosen metric, and produces data visualizations quickly compare and identify the best performing models
- Deploy & Manage Amazon SageMaker makes it easy to deploy your trained model into production with a single click auto-scaling Amazon ML instances across multiple availability zones for high redundancy. Amazon SageMaker will launch the instances, deploy your model, and set up the secure HTTPS endpoint for your application Your application simply needs to include an API call for inference endpoint.
- Automatically Spot concept drift Have error in production? Amazon SageMaker Model Monitor monitors models in production and detects errors so you can take remedial actions. Analyzes the data collected at regular frequency based on built-in rules or customer-provided rules Integrated with CW - metrics emitted in CloudWatch so you can set up alarms 做 notification
- Add human review Many machine learning applications require humans to review low confidence predictions to ensure the results are correct. Amazon Augmented AI (Amazon A2I) build the workflows required for human review of ML predictions.
- Scalable prod environment Auto scale ML instances - specify min and max End to end — Web-based IDE for ML development: Amazon SageMaker Studio Automatically build and train models based on your data - SageMaker Autopilot allowing to maintain full control and visibility. Process is completely transparent. dive into the details of how it was created refine based on your needs
- Now, at the top of the stack For developers who want to add solution-oriented intelligence to their applications through an API call rather than developing and training their own models Vision (Rek Image and video) “tell me what’s in this image,” “which celebrities are in this image,” and, “is this image safe for work?” Speech – (Polly, Transcribe) – If you want to turn text into speech we have Polly, and if you want to transcribe audio we have Transcribe Text (translate, comprehend, textract) – If you want to translate text to different languages we have translate, and if you want to understand what’s being said in those translated corpuses of text you have Comprehend, if you want to extract text from documents using OCR++, you have Textract. Then you have the things which Amazon is relatively famous for. Alexa and chatbots with Lex, personalization, and forecasting Customers have asked us to continue to consider adding services around areas where we have a lot of experience and data from our consumer business… Amazon Fraud Detector, a new fraud management service for detecting online identify and payment fraud in real-time. No machine learning experience required. Amazon CodeGuru, a new machine learning service to automate code reviews and identify your most expensive line of code. Kendra, a new service which reinvents enterprise search with machine learning. Contact Lens for Amazon Connect provides machine learning powered contact center analytics for Amazon Connect
- Amazon Fraud Detector Fraud teams often lack in-house ML experts Amazon FraudDetector - 全受管詐騙偵測器服務 根據Amazon 超過 20 年的詐騙偵測專業知識 Use case: - new account registration - EC order checkout - online payment How? - Upload your historical fraud datasets to Amazon S3 - Select from pre-built fraud detection model templates - automatically train, test, and deploy a customized fraud detection model your appliation can call the API to do inference
- Amazon Personalize 就像您專屬的 Amazon.com 機器學習推薦團隊 Personalization project, typical>6mo, now you can get it done in weeks The technology behind is well tested for performance and scale Personalize uses deep learning sequence model which greatly improve result Personalize can be used by all developers and does not need ML expertise Models are private to you and only used for generating recommendations for your users
- Start with providing 3 kinds of datasets User interactions – what are the users clicking on, what are they purchasing, what are they watching. This has the strongest signal for personalization and is the only dataset compulsorily needed. The other two are optional Inventory (aka catalog) – details about your items. Items IDs + price, category, genre (if books), type etc. User details – details about the user – age, gender etc. Once you get the data in it is very simple to get a trained personalization model in Personalize Under the hood Personalize does feature engineering, section of hyperparameters, model training, hosting, deployment and management Models are private to you and only used for generating recommendations for your users
- With Amazon Personalize you are able to delight your customers. Deliver high quality recommendations in real time Train a recommendation model with a few clicks Generate recommendations for almost any product or content
- To summarize, the AWS AI and ML stack has three layers. Each layer addressing different audiences: ML Frameworks & Infrastructure: For expert machine learning practitioners who work at the framework level and are comfortable building, training, tuning, and deploying machine learning models. ML Services: For every day developers and data scientists we built and launched Amazon SageMaker, a managed ML service to build, train, and deploy machine learning models quickly. AI Services: Developers with no prior machine learning experience can easily build sophisticated AI driven applications, like an AI driven contact center, live media subtitling, understanding voice of the customer, content moderation, or identity safety and verification, often
- While AI technology is important to build successful AI initiatives, the organization culture is also important— skills in-house or can partner for expertise
- We have created a range of ways for developers of any skill levels to get started with machine learning. 2./ We’ve created a portfolio of educational devices to help put new machine learning techniques into the hands of developers in unique and fun ways, with AWS DeepLens, AWS DeepRacer, and the AWS DeepRacer League – and now AWS DeepComposer. For more traditional classroom-based learning, there are a few options. 4./ We have a range of online and training available – most of which you can get started for free! 5./ AWS ML training, the curriculum that is used to train Amazon developers is available to you through AWS, for free 6./ We also partner closely with leading online course providers such as Udacity, Coursera and edX to bring relevant courses that help developers build their machine learning skills.
- What is AWS DeepComposer? AWS DeepComposer gives developers a creative way to get started building generative AI models – hands on with a musical keyboard - create a melody that will transform into a completely original song in seconds, all powered by AI. Generative AI is one of the biggest recent advancements in artificial intelligence technology because of its ability to create something new.
- 1/ input a melody by connecting the AWS DeepComposer keyboard to your computer, or play the virtual keyboard in the AWS DeepComposer console. AWS DeepComposer accepts music in MIDI format. 2/ Choose from jazz, rock, pop, symphony or Jonathan Coulton pre-trained models. you can also build your own custom genre model in Amazon SageMaker. 3/ You can then publish your tracks to SoundCloud in one click, or export MIDI files to your favorite Digital Audio Workstation (like Garage Band) and get even more creative.
- Consultant Service The Amazon ML Solutions Lab pairs your team with Amazon machine learning experts to prepare data, build and train models, and put models into production. hands-on educational workshops with brainstorming sessions, ‘work backwards’ from business challenges, and then go step-by-step through the process of developing machine learning-based solutions. At the end of the program, you will be able to take what you have learned through the process and use it elsewhere in your organization to apply ML to business opportunities. We help customers adopt ML with more than 145 engagements so far including AstraZeneca, NFL, and the MLB.
- NextGen stats We helped bring together domain experts and technical experts in our ML Solutions Lab Capture and process data – use EC2, S3, and EMR Machine learning models built on Amazon SageMaker to output predictions, stats and more. Distribute result - leverages AWS Lambda, Amazon ElastiCache, Quicksight, RDS, Route S3, API gateway, and DynamoDB to put the insights, predictions, and stats into broadcast analysis, scouting, and coaching tools. 台灣 customer – Formosa Plastic / 台塑
- What is next? machine learning curriculum we used to train thousands of Amazon’s own developers and made it available digitally, for free. We also offer live classes with accredited AWS instructors who teach using a mix of presentations, discussion, and hands-on labs. Explore tailored learning paths by role for ML at aws.training/ML