Introduction to Artificial Intelligence

Introduction to Artificial Intelligence

Dfda3ce33093a2ce23246410c5087a92?s=128

Ali Akbar S.

October 01, 2019
Tweet

Transcript

  1. Ali Akbar Septiandri Introduction to Artificial Intelligence

  2. TABLE OF CONTENTS WHAT IS AI? Why do people talk

    about it? IN DEPTH What is deep learning? Why is it different from machine learning? 01 03 02 04 05 06 AI in INDONESIA Where are we now? ETHICAL AI Is AI innately unbiased? FUTURE of AI What’s next for AI? AI for EVERYONE How to prepare individuals and companies for AI era?
  3. WHAT IS AI? 01 Why do people talk about it?

  4. Alan Turing, Computing Machinery and Intelligence I propose to consider

    the question, “Can machines think?” This should begin with definitions of the meaning of the terms “machine” and “think.”
  5. • Can machines be more intelligent than humans? • If

    so, what will happen to humanity? • How can we measure human intelligence? SOME PHILOSOPHICAL QUESTIONS
  6. Source: Digital Wellbeing

  7. None
  8. Full Self-Driving Tesla

  9. AI Adoption by McKinsey Global Institute (2017)

  10. AI DESIGNS • Rational actions • Mathematical and empirical evaluation

    • Psychology, neuroscience • Optimization theory, statistics, game theory, etc.
  11. Source: NVIDIA

  12. Jake VanderPlas “Fundamentally, machine learning involves building mathematical models to

    help understand data.”
  13. None
  14. WHY IS IT HARD?

  15. WHY IS IT HARD?

  16. WHAT IS IN THIS PICTURE?

  17. WHAT IS IN THIS PICTURE?

  18. ENTER: DEEP LEARNING This is the one responsible for the

    AI hype nowadays
  19. A DEEP NEURAL NETWORK

  20. None
  21. IN DEPTH 02 What is deep learning? Why is it

    different from machine learning?
  22. LEVELS OF INTELLIGENCE Simplest form of thought process Search problems,

    Markov decision processes, adversarial games Constraint satisfaction problems, Bayesian networks First-order logic, knowledge base
  23. MACHINE LEARNING TASKS UNSUPERVISED LEARNING SUPERVISED LEARNING REINFORCEMENT LEARNING

  24. SUPERVISED LEARNING Even better, can we tell that there is

    a cat and a dog in the image? How to identify cats or dogs in an image?
  25. BEFORE (CLASSICAL ML) IMAGE PROCESSING Edge detection, texture analyser, color

    histogram FEATURE EXTRACTION Eye position, eye colour, nose colour, fur type, leg counts MODEL Logit model, SVM, k-NN
  26. y = σ(β 0 + β 1 x 1 +

    β 2 x 2 + β 3 x 3 ) Trying to find the optimum weights for each feature
  27. AFTER (DEEP LEARNING) INPUT You only need raw images! FEATURE

    EXTRACTION + MODEL TRAINING Your model will do the feature extraction step for you OUTPUT Just define how many classes
  28. ...but we need millions of images!

  29. images from 20k categories AlexNet achieved a top-5 error of

    15.3% 14 mio
  30. Convolutional Neural Networks

  31. Generative Adversarial Networks (GANs)

  32. Recurrent Neural Networks

  33. UNSUPERVISED LEARNING Customer segmentation (clustering) Topic modelling Anomaly or outlier

    detection
  34. UNSUPERVISED LEARNING Word Representations

  35. REINFORCEMENT LEARNING Monte Carlo Tree Search for AlphaGo

  36. Source: NVIDIA

  37. AI in INDONESIA 03 Where are we now?

  38. None
  39. None
  40. None
  41. None
  42. None
  43. ETHICAL AI 04 Is AI innately unbiased?

  44. None
  45. None
  46. None
  47. None
  48. None
  49. None
  50. BIAS and FAIRNESS in AI

  51. “...a project to look for skin cancer in photographs. It

    turns out that dermatologists often put rulers in photos of skin cancer, for scale, but that the example photos of healthy skin do not contain rulers. To the system, the rulers (or rather, the pixels that we see as a ruler) were just differences between the example sets, and sometimes more prominent than the small blotches on the skin. So, the system that was built to detect skin cancer was, sometimes, detecting rulers instead.” (Evans, 2019) BIAS in AI
  52. UNDERSTANDING CAUSALITY Confounders Graphical presentation of confounding in directed acyclic

    graphs (Suttorp et al., 2014) Age Chronic kidney disease Mortality
  53. Dermatoscopic images See (Finlayson et al., 2019) ADVERSARIAL ATTACKS

  54. FUTURE of AI 05 What’s next for AI?

  55. Is it even possible? ARTIFICIAL GENERAL INTELLIGENCE

  56. AlphaStar by DeepMind

  57. Generating Images from Brain Signals

  58. None
  59. AI-Aided Prediction Deep Neural Networks Improve Radiologists’ Performance in Breast

    Cancer Screening (Wu et al., 2019) Human+AI AUC, Malignant Prediction
  60. AI for EVERYONE 06 How to prepare individuals and companies

    for AI era?
  61. None
  62. AI Transformation Execute pilot projects to gain momentum Customer segmentation,

    review categorization, photo classification 1st Build an in-house AI team Find people with specific skills related to ML 2nd Provide broad AI training Everyone should learn what AI is! Be it managers, team leaders, etc. 3rd Develop an AI strategy What are some vital things to enhance? 4th Develop internal & external communications Talk with board of directors, investors, your precious stakeholders 5th
  63. WHAT ML CAN DO Anything you can do with 1

    second of thought, we can probably now or soon automate
  64. WHAT ML CANNOT DO e.g. Market research and run an

    extended market report, give empathetic responses
  65. WORKING with an AI TEAM 1 Define acceptance criteria Sometimes,

    you don’t really need 100% accuracy 3 Ensure clean data Big data without correct labels can mislead you! 2 Find sufficient data The bigger, the better
  66. AI PITFALLS to AVOID Don’t think you need superstar AI

    engineers Don’t expect traditional planning w/o changes Don’t expect it to work the first time Don’t just hire 2/3 super ML engineers Don’t expect AI to solve everything
  67. by Ainun Najib AI/ML for KIDS

  68. None
  69. CREDITS: This presentation template was created by Slidesgo, including icons

    by Flaticon, and infographics & images by Freepik. Please keep this slide for attribution. Does anyone have any questions? pm@aliakbars.id @aliakbars uai.aliakbars.id THANKS