Diese Präsentation wurde erfolgreich gemeldet.
Die SlideShare-Präsentation wird heruntergeladen. ×

Cassie Kozyrkov. Journey to AI

Ad

Journey to AI
Cassie Kozyrkov,
Chief Decision Scientist, Google Cloud
@quaesita

Ad

Confidential & Proprietary 2
An unprecedented
opportunity
@quaesita

Ad

Confidential & Proprietary 3
AI in action
At 10 minutes?
It’s random luck
AI in action
At 10 minutes?
It’s random luck
@qu...

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Your AI journey awaits!
@quaesita
Wird geladen in …3
×

Hier ansehen

90 von 90 Anzeige
90 von 90 Anzeige

Cassie Kozyrkov. Journey to AI

Herunterladen, um offline zu lesen

Cassie Kozyrkov, Chief Decision Scientist at Google Cloud Journey to AI

“Machine learning and artificial intelligence are no longer science fiction, but what does it take to harness their potential? Let’s strip away the jargon in machine learning and AI to take a look at what’s easy, what’s hard, and how to spot opportunities.”​

As Chief Decision Scientist at Google Cloud, Cassie Kozyrkov advises leadership teams on decision process, AI strategy, and building data-driven organizations. She works to democratize statistical thinking and machine learning so that everyone – Google, its customers, the world! – can harness the beauty and power of data. She is the force behind bringing the practice of Decision Intelligence to Google and she has personally trained over 15,000 Googlers in machine learning, statistics, and data-driven decision-making. Before her current role, she served in Google’s Office of the CTO as Chief Data Scientist. Prior to joining Google, Cassie worked as a data scientist and consultant. She holds degrees in mathematical statistics, economics, psychology, and cognitive neuroscience. When she’s not working, you’re most likely to find Cassie at the theater, in an art museum, exploring the world, or curled up with a good novel.

Cassie Kozyrkov, Chief Decision Scientist at Google Cloud Journey to AI

“Machine learning and artificial intelligence are no longer science fiction, but what does it take to harness their potential? Let’s strip away the jargon in machine learning and AI to take a look at what’s easy, what’s hard, and how to spot opportunities.”​

As Chief Decision Scientist at Google Cloud, Cassie Kozyrkov advises leadership teams on decision process, AI strategy, and building data-driven organizations. She works to democratize statistical thinking and machine learning so that everyone – Google, its customers, the world! – can harness the beauty and power of data. She is the force behind bringing the practice of Decision Intelligence to Google and she has personally trained over 15,000 Googlers in machine learning, statistics, and data-driven decision-making. Before her current role, she served in Google’s Office of the CTO as Chief Data Scientist. Prior to joining Google, Cassie worked as a data scientist and consultant. She holds degrees in mathematical statistics, economics, psychology, and cognitive neuroscience. When she’s not working, you’re most likely to find Cassie at the theater, in an art museum, exploring the world, or curled up with a good novel.

Anzeige
Anzeige

Weitere Verwandte Inhalte

Ähnlich wie Cassie Kozyrkov. Journey to AI (20)

Weitere von IT Arena (20)

Anzeige

Cassie Kozyrkov. Journey to AI

  1. 1. Journey to AI Cassie Kozyrkov, Chief Decision Scientist, Google Cloud @quaesita
  2. 2. Confidential & Proprietary 2 An unprecedented opportunity @quaesita
  3. 3. Confidential & Proprietary 3 AI in action At 10 minutes? It’s random luck AI in action At 10 minutes? It’s random luck @quaesita
  4. 4. Confidential & Proprietary 4 AI in action At 120 minutes? It’s flawless AI in action At 120 minutes? It’s flawless @quaesita
  5. 5. Confidential & Proprietary 5 AI in action At 240 minutes? Strategy emerges AI in action At 240 minutes? Strategy emerges @quaesita
  6. 6. Artificial intelligence Machine learningNeural networks Models Confidential & Proprietary@quaesita
  7. 7. Confidential & Proprietary@quaesita Machine learning is an approach to making lots of small decisions
  8. 8. Confidential & Proprietary Machine learning vs traditional programming Information AnswerRECIPE @quaesita
  9. 9. Confidential & Proprietary Traditional programming: Handcrafted recipe ? (recipe) @quaesita
  10. 10. Confidential & Proprietary Traditional programming: Handcrafted recipe 0 1 0 1 1 0 1 0 1 0 0 1 (information) (answer) CODE (recipe) @quaesita
  11. 11. Confidential & Proprietary Machine learning: Recipe learned from data (information) (answer) 0 1 0 1 1 0 1 0 1 0 0 1 (information) (answer) 0 1 1 1 1 0 1 0 1 0 0 0 0 1 1 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 0 1 0 1 1 1 1 0 0 1 0 1 1 1 0 1 1 1 @quaesita
  12. 12. Confidential & Proprietary (information) (answer) 0 1 0 1 1 0 1 0 1 0 0 1 (information) (answer) 0 1 1 1 1 0 1 0 1 0 0 0 0 1 1 0 1 0 0 0 1 0 0 0 ? (recipe) 1 0 0 0 0 1 0 1 0 1 1 1 1 0 0 1 0 1 1 1 0 1 1 1 ML MODEL (recipe) Machine learning: Recipe learned from data @quaesita
  13. 13. Confidential & Proprietary Step-by-step machine learning @quaesita
  14. 14. Confidential & Proprietary Step 1 Define your objective.
  15. 15. Confidential & Proprietary Define your objective @quaesita
  16. 16. Confidential & Proprietary Step 1: Define your objective ? (information) ? (answer) ? (recipe) Think about desired outputs @quaesita
  17. 17. Confidential & Proprietary Step 1: Define your objective ? (information) Y / N (answer) ? (recipe) @quaesita
  18. 18. Confidential & Proprietary I’d like: Y/N with at least 80% accuracy. Step 1: Define your objective ? (information) Y / N (answer) ? (recipe) @quaesita
  19. 19. Confidential & Proprietary@quaesita ? (information) Y / N (answer) ? (recipe) I’d like: Y/N with at least 80% accuracy. Step 1: Define your objective @quaesita
  20. 20. Confidential & Proprietary Step 2 Wrangle your data.
  21. 21. Confidential & Proprietary Wrangle your data @quaesita
  22. 22. Confidential & Proprietary Get some data to learn from ? (information) Y / N (answer) ? (recipe) Identify some useful inputs @quaesita
  23. 23. Confidential & Proprietary Y / N (answer) ? (recipe) Age Review Score (information) Get some data to learn from @quaesita
  24. 24. Confidential & Proprietary 100 50 30 20 Split your data @quaesita
  25. 25. Confidential & Proprietary Data inputs Wine classification Age in years Reviewscore 4 2 0 -2 2 4 6 8 @quaesita
  26. 26. Confidential & Proprietary Use a line? Wine classification Age in years Reviewscore 4 2 0 -2 2 4 6 8 Use a line? @quaesita
  27. 27. Confidential & Proprietary Step 3 Train your solution
  28. 28. Confidential & Proprietary Train your solution @quaesita
  29. 29. Confidential & Proprietary Train your solution ? (information) Y / N (answer) Pick an algorithm! ? (recipe) Age Review Score (information) @quaesita
  30. 30. Confidential & Proprietary Train your solution Wine classification Age in years Reviewscore 4 2 0 -2 2 4 6 8 Support Vector Classifier Decision Tree Neural Network Algorithm selection: @quaesita
  31. 31. Confidential & Proprietary Wine classification Age in years Reviewscore 4 2 0 -2 2 4 6 8 Support Vector Classifier Decision Tree Neural Network Algorithm selection: Train your solution @quaesita
  32. 32. Confidential & Proprietary Support Vector Classifier Algorithm selected: Wine classification Age in years Reviewscore 4 2 0 -2 2 4 6 8 Run ML algorithm Train your solution @quaesita
  33. 33. Confidential & Proprietary Support Vector Classifier Algorithm selected: Run ML algorithm Wine classification Age in years Reviewscore 4 2 0 -2 2 4 6 8 Train your solution @quaesita
  34. 34. Confidential & Proprietary Get model (recipe) Wine classification Age in years Reviewscore 4 2 0 -2 2 4 6 8 Train your solution @quaesita
  35. 35. Confidential & Proprietary Get model (recipe) Train your solution @quaesita
  36. 36. Confidential & Proprietary New data? Use recipe! Train your solution @quaesita
  37. 37. Confidential & Proprietary New data? Use recipe! Train your solution @quaesita
  38. 38. Confidential & Proprietary Support Vector Classifier Decision Tree Neural Network Algorithm selection: Wine classification Age in years Reviewscore 4 2 0 -2 2 4 6 8 Train your solution @quaesita
  39. 39. Confidential & Proprietary Support Vector Classifier Decision Tree Neural Network Algorithm selection: Wine classification Age in years Reviewscore 4 2 0 -2 2 4 6 8 Train your solution @quaesita
  40. 40. Confidential & Proprietary Run ML algorithm Wine classification Age in years Reviewscore 4 2 0 -2 2 4 6 8 Decision Tree Algorithm selected: Train your solution @quaesita
  41. 41. Confidential & Proprietary Decision Tree Algorithm selected: Run ML algorithm Wine classification Age in years Reviewscore 4 2 0 -2 2 4 6 8 Train your solution @quaesita
  42. 42. Confidential & Proprietary Decision Tree Algorithm selected: Run ML algorithm Wine classification Age in years Reviewscore 4 2 0 -2 2 4 6 8 Train your solution @quaesita
  43. 43. Confidential & Proprietary Get model (recipe) Wine classification Age in years Reviewscore 4 2 0 -2 2 4 6 8 Train your solution @quaesita
  44. 44. Confidential & Proprietary Get model (recipe) Train your solution @quaesita
  45. 45. Confidential & Proprietary New data? Use recipe! Train your solution @quaesita
  46. 46. Confidential & Proprietary New data? Use recipe! Train your solution @quaesita
  47. 47. Confidential & Proprietary Decision tree Which model do we trust? Support vector classifier @quaesita
  48. 48. Confidential & Proprietary Step 4 Test your model.
  49. 49. Confidential & Proprietary Test your model @quaesita
  50. 50. Confidential & Proprietary@quaesita 100 50 30 20 Test with new data Training set Validation set Test set Fitting the model Soft interim evals Rigorous final eval
  51. 51. Confidential & Proprietary@quaesita Take some fresh data Testing data Age in years Reviewscore 4 2 0 -2 2 4 6 8
  52. 52. Confidential & Proprietary@quaesita Apply each model to fresh data Support vector classifier Decision tree
  53. 53. Confidential & Proprietary@quaesita Get classifications Support vector classifier Decision tree
  54. 54. Confidential & Proprietary@quaesita Compare with the correct answers Truth Age in years Reviewscore 4 2 0 -2 2 4 6 8
  55. 55. Confidential & Proprietary@quaesita Get model performance Support vector classifier Decision tree Test Accuracy: 100% Test Accuracy: 75%
  56. 56. Confidential & Proprietary@quaesita And the winner is... Statistical Output (CI): [80%, 100%] Statistical Output (CI): [51%, 90%] Support vector classifier Decision tree
  57. 57. Confidential & Proprietary Testing is crucial! It’s the key to responsible ML and AI. Make sure your solution actually works on relevant, new data. @quaesita
  58. 58. Confidential & Proprietary Applied ML is easier than most people think Tinker and have fun, try out anything you like, but assess performance carefully. @quaesita
  59. 59. Confidential & Proprietary Success takes data You need separate datasets for training, validation, and testing. @quaesita
  60. 60. Confidential & Proprietary Machine learning is an approach to making repeated decisions that involves algorithmically finding patterns in data and using these to make recipes that deal correctly with brand new data. @quaesita
  61. 61. Confidential & Proprietary What is AI? @quaesita
  62. 62. Vision @quaesita
  63. 63. Communication @quaesita
  64. 64. Confidential & Proprietary 64 Playing games
  65. 65. Confidential & Proprietary AI tasks Information Answer? @quaesita
  66. 66. Confidential & Proprietary AI tasks Cat!? @quaesita
  67. 67. Confidential & Proprietary AI tasks Cat!? @quaesita
  68. 68. Confidential & Proprietary AI tasks Too difficult to write explicit instructions for Complicated! ? Cat! @quaesita
  69. 69. Confidential & Proprietary AI succeeds at very complicated tasks that programmers can’t write instructions for by hand @quaesita
  70. 70. Confidential & Proprietary Data Compute Sophisticated algorithms @quaesita
  71. 71. Confidential & Proprietary 71 Science fiction becomes reality 71 @quaesita
  72. 72. Confidential & ProprietaryGoogle Cloud Platform 72 Deep learning Geoff Hinton: engineering of artificial neural networks. Sophisticated algorithms
  73. 73. Confidential & Proprietary 73 Deep Learning
  74. 74. Confidential & Proprietary 74Proprietary + Confidential GPUs Researchers began to notice that neural network mathematics closely resembled the algorithms to shade pixels in graphics cards (GPUs) @quaesita
  75. 75. Confidential & ProprietaryGoogle Cloud Platform 75 Massive labeled datasets Fei-Fei Li: enabling learning through data. Data
  76. 76. Confidential & ProprietaryGoogle Cloud Platform 76 Hardware and compute Jeff Dean: TensorFlow and success at scale. Compute
  77. 77. Confidential & Proprietary 77 The future is here
  78. 78. Vision @quaesita
  79. 79. Confidential & Proprietary@quaesita Google Photos
  80. 80. Image Classification AI fixes satellite image maps automatically. @quaesita
  81. 81. Communication @quaesita Communication Google Cloud Platform Text-to-Speech
  82. 82. Confidential & Proprietary@quaesita
  83. 83. Confidential & Proprietary@quaesita Smart reply in Inbox 12% of all responses sent on mobile
  84. 84. Confidential & Proprietary 84 Playing games
  85. 85. Self-driving cars @quaesita
  86. 86. Cooling data centers @quaesita
  87. 87. Confidential & Proprietary • Predictive inventory planning • Recommendation engines • Upsell and cross-channel marketing • Market segmentation and targeting • Customer ROI and lifetime value Retail • Alerts and diagnostics from real-time patient data • Disease identification and risk stratification • Patient triage optimization • Proactive health management • Healthcare provider sentiment analysis Healthcare and Life Sciences • Aircraft scheduling • Dynamic pricing • Social media – consumer feedback and interaction analysis • Customer complaint resolution • Traffic patterns and congestion management Travel and Hospitality • Risk analytics and regulation • Customer Segmentation • Cross-selling and up-selling • Sales and marketing campaign management • Credit worthiness evaluation Financial Services • Predictive maintenance or condition monitoring • Warranty reserve estimation • Propensity to buy • Demand forecasting • Process optimization • Telematics Manufacturing • Power usage analytics • Seismic data processing • Carbon emissions and trading • Customer-specific pricing • Smart grid management • Energy demand and supply optimization Energy, Feedstock and Utilities @quaesita
  88. 88. ML PlatformML Pre-Trained APIs Cloud Vision Cloud Natural Language Cloud Translation Cloud Speech Cloud Video Intelligence ML Accelerators Cloud GPU Cloud TPU Cloud ML Engine Cloud Dataproc Cloud Dataflow Google BigQuery Cloud Job Discovery ML Professional Services & PartnersML & Data Science Tools ML Libraries Cloud Datalab Cloud Dataprep ASL Google Cloud AI @quaesita
  89. 89. Confidential & Proprietary Decision making Domain expertise Analytics Reliability EngineeringSteps to success 1. Define success 2. Prepare your data 3. Try out different models 4. Test on lots of new data 5. Use your solution! @quaesita
  90. 90. Your AI journey awaits! @quaesita

×