Upgrade to Pro — share decks privately, control downloads, hide ads and more …

AI & Enterprise

AI & Enterprise

AI Congress London keynote covering how to understand and implement applied AI for business.

Melanie Warrick

January 30, 2018
Tweet

More Decks by Melanie Warrick

Other Decks in Technology

Transcript

  1. AI & Enterprise Melanie Warrick

  2. @nyghtowl

  3. @nyghtowl Multi years & Multi millions

  4. 90% of Data Last 2 Years

  5. 1.5GB / day avg person 1PB / day smart factory

    By 2020
  6. @nyghtowl Are you innovating with data?

  7. @nyghtowl Machine Learning

  8. @nyghtowl

  9. @nyghtowl “AI is the new electricity” - Andrew Ng ‘17

  10. @nyghtowl New tech is like magic Change => fear |

    excitement
  11. @nyghtowl “Science and engineering of making intelligent machines...” - John

    McCarthy ‘55
  12. @nyghtowl

  13. @nyghtowl

  14. @nyghtowl A.I. is the science of making technology smart

  15. @nyghtowl Artificial Intelligence Machine Learning Deep Learning

  16. @nyghtowl AlphaGo

  17. @nyghtowl Translate

  18. @nyghtowl Computer Vision

  19. @nyghtowl Sentiment Analysis

  20. @nyghtowl AI is not the destination How to apply it?

  21. @nyghtowl

  22. @nyghtowl Speech Recognition

  23. @nyghtowl Auction Site & Real-time Car Image Classifier 30K dealers

    | 5M cars Classification 20 - a couple minutes AUCNET
  24. @nyghtowl Compology Waste Removal Automation IoT Cameras & Vision Modeling

    Reduced pickups by ~50%
  25. @nyghtowl Largest South African pay-TV company Siloed data BigQuery Tailored

    subscriptions & real time offers real-time response 8.5M customers 27+ data sources 960M points consolidated MultiChoice
  26. @nyghtowl Goal Experts Data Solution

  27. @nyghtowl Goal Experts Data Solution

  28. @nyghtowl Define the Problem Revenue Growth Customer Satisfaction Responsiveness Sales

    Expenses Operations Processes Compliance Latency Product & Process Innovation
  29. @nyghtowl Structured Data Classification/ Regression • Customer Churn Analysis •

    Product Diagnostics • Forecasting Recommendation • Content Personalization • Product X-Sells/Up-sells Anomaly Detection • Fraud Detection • Asset Sensor Diagnostics • Log Metric Anomalies Unstructured Data Image Analytics • Identify damaged shipments • Explicit Content Classification • Identify “styles” in images Text Analytics • Call Center log analysis • Language Identification • Topic Classification • Sentiment Analysis Machine Learning - Applications
  30. @nyghtowl Breakdown and Evolve

  31. @nyghtowl Growing Use of ML at Google Number of directories

    containing model description files 2012 2013 2014 2015 1500 1000 500 0 Used across products:
  32. @nyghtowl 40% savings

  33. @nyghtowl Goal Experts Data Solution

  34. @nyghtowl Capture Process & Train Store Analyze & Deploy Build

    the Solution
  35. @nyghtowl Capture Process & Train Store Analyze & Deploy Build

    the Solution
  36. 163 Zetabytes (1 trillion gigabytes) of data by 2025

  37. Proprietary + Confidential Datacenter as a Computer

  38. @nyghtowl Capture Process Store Analyze & Deploy Capture Process &

    Train Store Build the Solution
  39. @nyghtowl Machine Learning Modeling Capture & Store Train Model Process

    Deploy & Serve Analyze & Improve
  40. @nyghtowl Make Data Coherent | Modeling

  41. @nyghtowl Neural Network = tons of multiply and add

  42. @nyghtowl Neural Network can extract hidden features from data

  43. @nyghtowl Neural Network is a function that can learn

  44. @nyghtowl 10 yrs => 1 day Simplify & Speed up

    Comprehension
  45. @nyghtowl Machine Learning Modeling Train Model Process Deploy & Serve

    Analyze & Improve Capture & Store
  46. @nyghtowl ML Data Lifecycle Solution Capture & Store Train Model

    Process Deploy & Serve Analyze & Improve Pub/ Sub Storage Cloud SQL Bigtable Dataflow Dataflow BigQuery TensorFlow Compute Engine ML Engine Datalab ML Engine Kubernetes Engine
  47. @nyghtowl End to End ML Pipeline Inputs Train model Pre

    processing Asset Creation Distributed training, Hyper-parameter tuning Deploy: Including Model Versioning REST API Prediction: load balanced, auto-scaled, scale-to-zero Model Remote Clients REST API call with input variables Asset Creation Model + Assets
  48. @nyghtowl Goal Experts Data Solution

  49. @nyghtowl Collaboration & Communication Business Engineers Researchers

  50. @nyghtowl Finding Solutions | Partners | Tools ➔ Proven experience?

    ➔ Temporary or full production? ➔ Scale & flexibility? ➔ Turnaround time? ➔ Success measure? ➔ Fit into existing systems? ➔ End to end?
  51. @nyghtowl End to End ML Pipeline Inputs Train model Pre

    processing Asset Creation Distributed training, Hyper-parameter tuning Deploy: Including Model Versioning REST API Prediction: load balanced, auto-scaled, scale-to-zero Model Remote Clients REST API call with input variables Asset Creation Model + Assets
  52. @nyghtowl Language API Vision API Translate API Speech API Remove

    Entry Barriers
  53. @nyghtowl Remote Input Predictions served, Load balanced, Auto-scaled, Scale-to-zero Pretrained

    ML Pipeline REST API Capture & Send Deploy & Serve
  54. @nyghtowl Translate Perfect translation Human Neural (GNMT) Phrase-based (PBMT) English

    > Spanish English > French English > Chinese Spanish > English French > Spanish Chinese > Spanish Translation model Translation quality old: PBMT new: GNMT
  55. @nyghtowl AutoML Vision Input Train Deploy Serve REST API Remote

    Clients AutoML Vision
  56. @nyghtowl Custom ML Modeling Collect & Store Train Model Process

    Deploy & Serve Analyze & Improve
  57. @nyghtowl Goal Experts Data Solution

  58. *Source: IDC Applied AI Focus & Simplify the Problem Utilize

    External Expertise
  59. @nyghtowl Thank you... Lindsay Cade Erin O’Connell Isabel Markl Eli

    Bixby Adam Shin Amy Unruh Sara Robinson June Andrews Tyler Benjamin Benjamin Chehebar
  60. @nyghtowl Resources • Garnter Road to Enterprise AI: https://www.gartner.com/imagesrv/media-products/pdf/rage_frameworks/rage-frameworks-1- 34JHQ0K.pdf

    • Google Cloud Platform: https://cloud.google.com/ • Launchpad Studio (AI startup incubator): https://developers.google.com/programs/launchpad/studio/ • Applied AI Examples: https://www.forbes.com/sites/robertadams/2017/01/10/10-powerful-examples-of-artificial-inte lligence-in-use-today/2/#1efe823e3c8b • TensorFlow: https://www.tensorflow.org/ • TensorFlow without a PhD: https://cloud.google.com/blog/big-data/2017/01/learn-tensorflow-and-deep-learning-without-a- phd • Deep Learning Training: www.fast.ai • End-to-end ML with TensorFlow on GCP: https://codelabs.developers.google.com/codelabs/end-to-end-ml/index.html?index=..%2F..%2Fi ndex#0
  61. @nyghtowl Image References: • iStock/diego_cervo • iStock/phive2015 • iStock/alexialex •

    iSTock/Rinelle • iStock/Noctiluxx • iStock/VladimirFloyd • istock/Besjunior • https://commons.wikimedia.org/wiki/File:Stones_go.jpg • http://www.ebss.co.jp/ebs/worldwide/service/implementation.htm • https://www.sciencedaily.com/releases/2013/05/130522085217.htm • https://images.yourstory.com/2016/09/Innovation-is-a-state-of-mind.png?auto=compress • Rosenfeld Media | https://www.flickr.com/photos/rosenfeldmedia/6949089460 | Copyright and disclaimer notice: https://creativecommons.org/licenses/by/2.0/ | License notice: https://creativecommons.org/licenses/by/2.0/legalcode • https://pixabay.com/en/backgammon-board-game-cube-strategy-1903940/ • http://alphastockimages.com/ • https://cloud.google.com/speech/ • https://dillonspace.blogspot.com/2014/08/hitting-bullseye.html • https://cdn.technologynetworks.com/tn/images/thumbnails/rectangle/cloud-based-research-informatics-improving-collabora tion-increasing-agility-and-reducing-operating-289722.png • https://www.techemergence.com/machine-learning-in-genomics-applications/ • https://www.biggerpockets.com/renewsblog/wp-content/uploads/2016/03/transcontinental-railroad.jpg • http://goo.gl/fymPMI
  62. @nyghtowl Questions? @nyghtowl gcppodcast.com