[정보] GCP 신기술 동향 소개 구글코리아 김종대 엔지니어

GCP 신기술 동향 소개 구글코리아 김종대 엔지니어 관련 자료입니다.

내용 참고 하시기 바랍니다.

 

 

GCP 신기술 동향 소개 구글코리아 김종대 엔지니어 from rockplace

 

 1. How Google Cloud Platform Is Different Jongdae Lim Cloud Architect jongdaelim@google.com
 2.  Google Mission Statement Organize the world’s information and make it universally accessible and useful
 3.  Seven cloud products with one billion users each up to 40% of World’s Internet Traffic
 4.  About hardware, software, and operations
 5.  Use Google innovation to solve the problems that matter most
 6.  LANGUAGE API VISION API APP ENGINE COMPUTE ENGINE CONTAINER ENGINE BIG QUERY DATA FLOW MACHINE LEARNING CLOUD STORAGE NETWORKING COMPUTE ENGINE Teams, mobility, devices Connected business platforms App development & management Data analytics & machine learning Infrastructure, storage, network Security / Scale / Control
 7.  What Makes Different?
 8.  Confidential & Proprietary Networking & Infrastructure
 9.  Edge points of presence (>100) Google fiber # # Future region and number of zones Current region and number of zones 3 3 2 3 3 3 3 3 2 4 3 3 2 Frankfurt Singapore S Carolina N Virginia Belgium London Taiwan Mumbai Sydney Oregon Iowa São Paulo Finland Tokyo Montreal California Netherlands 3 3 33 Google global cache edge nodes (>1000) Google Cloud Platform Regions 15 current regions. 6 new regions coming, Thirty-nine new zones over 2.5 years
 10.  The Network Matters Typical Cloud Provider Google Cloud
 11.  Confidential & Proprietary Security & Reliability
 12.  Layered Defense in Depth Security DeploymentUsage Operations Application Network Storage OS+IPC Boot Hardware
 13.  MORE AUTOMATED MORE CONTROL Encryption by default First on GCP Cloud Key Management Service Customer-Supplied Encryption Keys First on GCP for VMs Keep keys in the cloud, for direct use by cloud services Keep keys on-premise, and use them to secure your cloud services World-class encryption activated by default Encryption by Default More choice and control
 14.  Site Reliability Engineering How Google Runs Production Systems
 15.  Confidential & Proprietary Economics
 16.  Original Cloud Promise Use only what you need. Pay only for what you use. Typical Cloud Reality Prepayments, forecasting, and cost optimization teams.
 17.  Monthly Usage Price -10% -20% -30% 100% 75% 50% 25% 0% 25 % 50 % 75 % 100 % 24% average savings Automatic Sustained Use Discounts Original Cloud Promise
 18.  Confidential & Proprietary How do I create intelligence from data? Solutions for connecting data across different systems and environments, generating targeted insights from your data, and enabling applications to automatically self-learn and improve through Machine Learning.
 19.  Google research in data technologies 2012 20132002 2004 2006 2008 2010 GFS MapReduce Bigtable (HBase) Colossus Dremel Flume Megastore Spanner Millwheel PubSub F1 Google Research Publications referenced are available here: http://research.google.com/pubs/papers.html The Datacenter as a Computer: An Introduction to the Design of Warehouse-Scale Machines, 2009 http://research.google.com/pubs/pub35290.html
 20.  Google research in data technologies 2012 20132002 2004 2006 2008 2010 Google Research Publications referenced are available here: http://research.google.com/pubs/papers.html The Datacenter as a Computer: An Introduction to the Design of Warehouse-Scale Machines, 2009 http://research.google.com/pubs/pub35290.html Cloud Storage Dataproc Bigtable Cloud Storage BigQuery Dataflow Datastore Dataflow PubSub Spanner Spanner
 21.  What AlphaGo's win meant to you? 2 Years back, 9–15 March 2016 “Most observers at the beginning of the 2016 matches expected Lee to beat AlphaGo” including ME! Now, AlphaGo Zero 1. Beat the previous version of AlphaGo (Final score: 100–0). 2. Learn to perform this task from scratch, without learning from previous human knowledge (i.e. recorded game play). 3. World champion level Go playing in just 3 days of training.
 22.  Machine Learning lets us solve problems without knowing explicitly how to create the solution
 23.  Proprietary & Confidential Machine learning is already improving many of our products Search Search ranking Speech recognition Android Keyboard & speech input Play App recommendations Game developer experience Gmail Smart reply Spam classification Drive Intelligence in Apps Chrome Search by image Assistant Smart connections across products YouTube Video recommendations Better thumbnails Maps Parsing local search Translate Text, graphic and speech translation Cardboard Smart stitching Photos Photos search
 24.  Deep learning boosts Google Translate My translation, 2016 Google Translate, will soon be using a new algorithm that is entirely based on deep learning, announced on 27 Sep. 2016.
 25.  Google DC Ops Applying ML lead to 40% reduction in cooling energy in Google datacenters.
 26.  Confidential & Proprietary
 27.  Google Cloud AI Mission Statement Democratize AI by making it accessible, fast and useful for enterprises and developers
 28.  Powered by Open Source
 29.  Confidential & Proprietary Hardware Accelerated AI
 30.  ● 15 - 30x faster ● 30 - 80x operations/watt TPU vs Conventional processors
 31.  Jeff Dean, Google Senior Fellow 2014 Training a Large Scale Machine Translation Model 24 hours on 32 GPUs 6 hours on ⅛ of a TPU Pod
 32.  Confidential & Proprietary Machine Learning Options Use Our Models Leverage Google’s domain expertise No tools or expertise required Train Your Own Build on your own specialized domain expertise Use Google tools for building and training models OR
 33.  Confidential & Proprietary AI Infrastructure Cloud GPUs Cloud TPUsCompute Engine NetworkingCloud Storage Platform, Libraries, Tools DataLabCloud ML Engine BigQuery Perception Services Vision Video Intelligence Speech Natural Language Translation Google Cloud AI Platform
 34.  Confidential & Proprietary Data Scientist Cloud ML Engine Build Custom Models ML researcher Use & extend OSS SDK App Developer Perception Services Use pre-trained models Matching the tool for the job
 35.  Confidential & Proprietary Custom task What type of ML problem are you solving? Specific to your dataset ML APIsTensorFlow “bob” Generic task Someone else has solved this before Trained on common classes “cat”
36.  Confidential & Proprietary AI building blocks for all developers Cloud Natural Language Cloud Vision Cloud Translation Cloud Speech Cloud Video Intelligence Stay Tuned Fully trained ML models from Google Cloud Platform
 37.  Confidential & Proprietary Vision API Demo http://vision-explorer.reactive.ai/
 38.  Confidential & Proprietary Cats vs. Dogs? looks easy..
 39.  Confidential & Proprietary Reality is.. Convolutional Neural Network http://playground.tensorflow.org/
 40.  Confidential & ProprietaryConfidential & Proprietary UPDATEDEPLOYEVALUATETUNE ML MODEL PARAMETERS ML MODEL DESIGN DATA PREPROCESSING State of the Industry: Complex & Time Intensive Large computational resource . Machine learning expertise . Manual data labeling
 41.  Confidential & Proprietary Source: Data scientists= Kaggle Data scientist community , Developers: Evans Data Corporation the figure in 2016 was 21m State of the Industry: Lack of Expertise Very few users today can create a custom ML model. To democratize AI, we need to make AI accessible to millions more 1000’s Deep Learning Researchers 21M Developers Confidential & Proprietary <1M Data Scientists
 42.  Confidential & Proprietary Custom task Can anything do it for me? Specific to your dataset ML APIsTensorFlow “happy” Generic task Someone else has solved this before Trained on common classes “cat” Cloud AutoML
 43.  Confidential & ProprietaryConfidential & Proprietary UPDATEDEPLOYEVALUATETUNE ML MODEL PARAMETERS ML MODEL DESIGN DATA PREPROCESSING Introducing Cloud AutoML A technology that can automatically create a Machine Learning Model UPDATEDEPLOYEVALUATETUNE ML MODEL PARAMETERS ML MODEL DESIGN DATA PREPROCESSING
 44.  Confidential & Proprietary Simple transfer learning with efficient model training for quick demo within minutes High accuracy model based upon learning to learn within a day Coming first - Cloud AutoML Vision DATA PREPROCESSING You ONLY do AutoML - Learning to Learn
 45.  AutoML - Simple Graphical User Interface IMPORT - Training Data TRAIN
 46.  Confidential & Proprietary AutoML Demo https://youtu.be/GbLQE2C181U
 47.  Confidential & Proprietary Why Cloud AutoML? Your own custom models Simple Limited ML expertise needed High quality Confidential & Proprietary
 48.  Confidential & ProprietaryGoogle Cloud Platform 50 AutoML “The AutoML networks are smaller, more efficient, and better than the best in the world today.”
49.  Confidential & Proprietary Just for Cats vs. Dogs? Custom Model designed & tuned in years AutoML trained in hours LCD/LED Panel Vision Test >= A Display Manufacturer in Korea did PoC with only real 60K data..
 50.  Confidential & Proprietary Media & entertainment content management Healthcare Retail search and extended shelf Insurance
 51.  Vision Speech Language Reasoning AutoML for All
 52.  Thank You! Jongdae Lim jongdaelim@google.com