the magic of unsupervised learning agustinus nalwan
play

THE MAGIC OF UNSUPERVISED LEARNING Agustinus Nalwan Head of AI - PowerPoint PPT Presentation

THE MAGIC OF UNSUPERVISED LEARNING Agustinus Nalwan Head of AI Carsales.com.au A LITTLE BIT ABOUT MYSELF AI 4 Years SAVING OUR FOREST BUSHFIRE DETECTION BACK TO THE MAIN TOPIC THE MAGIC OF UNSUPERVISED LEARNING SUPERVISED LEARNING


  1. THE MAGIC OF UNSUPERVISED LEARNING Agustinus Nalwan Head of AI Carsales.com.au

  2. A LITTLE BIT ABOUT MYSELF AI 4 Years

  3. SAVING OUR FOREST BUSHFIRE DETECTION

  4. BACK TO THE MAIN TOPIC

  5. THE MAGIC OF UNSUPERVISED LEARNING

  6. SUPERVISED LEARNING

  7. SUPERVISED LEARNING IMAGE RECOGNITION

  8. CAR RECOGNITION

  9. Ford Kuga Titanium

  10. SUPERVISED LEARNING I am awesome

  11. HOW DO YOU TRAIN THE AI?

  12. TRAINING BMW X5 Ford Ecosport Hyundai i30

  13. LOTS OF THEM

  14. SERIOUSLY LOTS OF THEM

  15. 10,000,000 CARS

  16. 10,000,000 CARS LABELLED!

  17. FROM MULTIPLE ANGLES

  18. PROBLEMS • Huge effort • Not practical

  19. MOST WORLD DATAS ARE UNLABELED • Facebook photos • Youtube videos • Twitter feeds

  20. UNSUPERVISED LEARNING

  21. UNSUPERVISED DEEP LEARNING GENERATING FACES

  22. IMAGE GENERATION • Variational Auto Encoder (VAE) • Generative Adversarial Network (GAN)

  23. VAE • Understand the subject (face) • Generate new subject

  24. HOW DOES IT WORK?

  25. AUTOENCODER 20x20 40x40 80x80 80x80 40x40 20x20 50x1 Latent Vector Decoder Encoder

  26. MSE, FPL - Error Images Minimize

  27. LATENT VECTORS Reconstructed Original Image Image 8.5 5.2 8.7 -3.2 1.5 -2.4 4.3 4.5 5.4 4.2 -2.0 0.9 4.3 -3.5 1.4

  28. LATENT VECTORS Reconstructed Original Image Image 8.5 5.2 8.7 -3.2 1.5 -2.4 4.3 4.5 5.4 4.2 -2.0 0.9 4.3 -3.5 1.4 Hair-length

  29. LATENT VECTORS Reconstructed Original Image Image 8.5 5.2 8.7 -3.2 1.5 -2.4 4.3 4.5 5.4 4.2 -2.0 0.9 4.3 -3.5 1.4 Hair-length Skin-color

  30. LATENT VECTORS Reconstructed Original Image Image 8.5 5.2 8.7 -3.2 1.5 -2.4 4.3 4.5 5.4 4.2 -2.0 0.9 4.3 -3.5 1.4 Hair-length Skin-color Eye-size Gender Age

  31. HOW COULD IT BE POSSIBLE? SMALL LATENT DIMENSION REMEMBER THE SIGNIFICANT DIFFERENCES

  32. PACKING A TRAVEL BAG

  33. BIG LUGGAGE

  34. SMALL BAG Latent Vector Important Features Learned

  35. MEMORY GAMES Info 1 Info 2 Info 3 Info 4 Info 1 Info 2 Info 3 Info 4 Gender Hair Color Skin Color Age

  36. HOW DO WE USE IT TO GENERATE NEW FACE?

  37. AUTOENCODER 20x20 40x40 80x80 80x80 40x40 20x20 50x1 Latent Vector Decoder Encoder

  38. AUTOENCODER 50x1 20.1, 10.5, -5.2, 6.2, … Latent Vector Decoder

  39. AUTOENCODER 50x1 20.1, 10.5, -5.2, 6.2, … Latent Vector Decoder

  40. LATENT SPACE DISTRIBUTION Land of no face

  41. LATENT SPACE

  42. MORPHING

  43. FACE ARITHMETIC

  44. LATENT VECTORS Reconstructed Original Image Image 8.5 5.2 8.7 -3.2 1.5 -2.4 4.3 4.5 5.4 4.2 -2.0 0.9 4.3 -3.5 1.4 Hair-length Skin-color Eye-size Gender Age

  45. LATENT VECTORS Reconstructed Original Image Image 8.5 5.2 8.7 -3.2 1.5 -2.4 4.3 4.5 5.4 4.2 -2.0 0.9 4.3 -3.5 1.4 Hair-length

  46. LATENT VECTORS Reconstructed Original Image Image 8.5 5.2 8.7 -3.2 1.5 -2.4 4.3 4.5 5.4 4.2 -2.0 0.9 4.3 -3.5 1.4 0.7 0.1 0.2 Hair-length

  47. 20.1, 10.5, -5.2 5.0, 6.3, -5.6 Bang vector 15.1, 4.2, 0.4

  48. BANG PREVIEW Bang vector 1.0, 5.3, 3.2 15.1, 4.2, 0.4 16.1, 9.5, 3.6

  49. ADDING GLASSES

  50. REMOVING GLASSES

  51. ADDING YOUTH

  52. GAN 
 GENERATIVE ADVERSARIAL NETWORK

  53. Real Face Discriminator Prediction Images Network Real or Fake Punished on Discriminator Network’s Generated failure Face Image Punished on Discriminator Network’s Generative Random Vector success Network

  54. MORPHING USING GAN

  55. OTHER APPLICATION OF GENERATIVE NETWORK

  56. SUPER RESOLUTION

  57. SUPER RESOLUTION

  58. SUPER RESOLUTION

  59. SUPER RESOLUTION

  60. SUPER RESOLUTION

  61. SUPER RESOLUTION

  62. SUPER RESOLUTION 256x512 64x128 4x

  63. SUPER RESOLUTION

  64. UNSUPERVISED LEARNING WHY IS IT IMPORTANT ?

  65. AGI

  66. WHAT’S NEEDED FOR AN AGI

  67. SELF LEARNING Supervised Unsupervised

  68. KNOWLEDGE EXTRACTION

  69. REASONING Reasoning Knowledge Extraction Information

  70. WITHOUT REASONING

  71. WITHOUT REASONING

  72. TAKE AWAY NOTHING

  73. DANGER !!!

  74. THANK YOU

  75. QUESTION?

Recommend


More recommend