1_intro2AI.pdf
- 1. So, what is AI?
- 5. Machine Learning (Statistical Learning) 機器學習 vs. 程式設計 Francois Chollet, “Deep Learning with Python,” Manning, 2017 一般程式設計 機器學習
- 7. Five Tribes of Machine Learning 機器學習的五大門派 • Evolutionaries (演化法) • Connectionists (類神經網路) • Symbolists (邏輯歸納法) • Bayesians (貝氏機率) • Analogizers (類比近似) Pedro Domingos
- 8. The Master Algorithm – Pedro Domingos
- 9. 3 Basic Operations of Algorithms 演算法的三大元素 • All Algorithms can be Reduced to 3 Operations 1 0 0 0 1 0 1 1
- 10. XOR 1 1 0
- 12. Neuron 神經元
- 13. Frank Rosenblatt’s Perceptron (1957)
- 15. Number of Connections in the Brain 每個人腦中都有個小宇宙!? Neurons (for adults): 10^11, or 100 billion, 100000000000 Synapses (based on 1000 per neuron): 10^14, or 100 trillion, 100000000000000 (每個神經元約有1000突觸)
- 18. Deep Learning
- 19. 19
- 20. Learning XOR (1986) Geoffrey Hinton
- 23. Chain Rule
- 24. Example: Recognizing Handwritten Digits •MNIST dataset
- 25. 3Blue1Brown
- 27. Gradient Descent
- 29. Convolutional Neural Network (LeNet-5) • https://medium.com/@sh.tsang/paper-brief-review-of-lenet-1-lenet-4-lenet-5- boosted-lenet-4-image-classification-1f5f809dbf17
- 31. ImageNet Large Scale Visual Object Recognition Challenge (ILSVRC) ImageNet 影像分類 競賽 • 1000 categories • For ILSVRC 2017 − Training images for each category ranges from 732 to 1300 − 50,000 validation images and 100,000 test images. • Total number of images in ILSVRC 2017 is around 1,150,000 31
- 32. Error Rate on ImageNet Challenge
- 33. Convolutional Neural Network (AlexNet) • Alex Krizhevsky, Geoffery Hinton et al., 2012
- 34. Error Rate on ImageNet Challenge
- 39. Recurrent Neural Networks (RNN)
- 40. Unroll the RNN
- 41. Long Short-term Memory (LSTM)
- 45. DRL in Atari Mnih et al., “Human Level Control through Deep Reinforcement Learning,” Nature, 2015
- 46. Learning to Play Atari Games
- 47. AlphaGo
- 48. Dr. Aja Huang (黃士杰)
- 49. The Complexity of Go vs Chess
- 50. AlphaGo Zero
- 51. 51
- 52. 52
- 54. Human’s Hope
- 56. Virtual-to-real Learning • Inspired by DeepMind (Mnih et al., Nature, 2015) − “Human Level Control through Deep Reinforcement Learning” • Applied to computer vision applications − Image segmentation: Armeni et al. (2016), Qiu et al., (2017) − Indoor navigation: Brodeur et al. (2017), Gupta et al. (2017), Savva et al. (2017), Wu et al. (2018) − Autonomous vehicles: Marinez et al. (2017), Muller et al. (2018), Pan et al. (2017), Shah et al. (2018) 56 UnrealCV CAD2Real
- 57. 57 Semantic Segmentation Depth Prediction VIVID Autonomous Navigation Action Recognition
- 59. Searching for the Shooter 59
- 60. Limits of Deep Learning 深度學習無所不能?
- 61. No Idea of Real World Alcorn et al., “Strike (with) a pose: Neural networks are easily fooled by strange poses of familiar objects,”CVPR 2019.
- 65. Generative Adversarial Networks (GAN) • Ian Goodfellow
- 66. Painting like Van Gogh
- 68. DeepFake: Is this you? 68
- 70. Google’s AutoML • Learning neural network cells automatically 70 https://ai.googleblog.com/2017/11/automl-for-large-scale-image.html
- 73. 73
- 74. Transformer & GPT3 https://www.youtube.com/watch?v=SZorAJ4I-sA Attention is All You Need!
- 75. huggingface.co
- 76. huggingface.co
- 77. 10 Jobs of the Future Fortune India, 2015
- 78. 1. The Data Managers (Data Scientists) The sexiest job of the 21st century (Harvard Business Review)
- 79. 2. The Space Invaders Demand for satellite engineers is going up, but there’s little training available.
- 80. 3. The Electronic (Digital) Warriors Defense electronics engineers will be in demand, but supply will be very short.
- 81. 4. The Drone Operators (Fleet Commanders) The market is set to open up to commercial operators, but a dearth of qualified engineers could ground it.
- 82. 5. The New-age Printers •3D printing could change manufacturing forever, and all it needs are hardware engineers.
- 83. 6. The Robotic Engineers • The robots are coming. Technology research and advisory firm Gartner predicts that more than 30% of all jobs will be replaced by robots and smart machines by 2025, and Ray Kurzweil, director of engineering at Google, anticipates that human thinking will become hybrid (biological and artificial intelligence) by the 2030s.
- 84. 7. The Cyberspace Ninjas (Cybersecurity) Prevention of data theft is just one of their jobs.
- 86. 9. The Green Warriors • India’s big push towards clean energy is changing the job market.
- 87. 10. The Medical Mechanics AI CT Scan https://tectales.com/ai/ai-used-to-identify-different-types-of-brain-injuries.html
- 88. References 1. https://www.buzzfeed.com/kasiagalazka/science-fiction-things-that-actually-exist-now 2. https://www.geek.com/movies/10-movies-that-helped-create-real-technology-1740036/ 3. https://www.gadgetsnow.com/slideshows/8-sci-fi-movie-technologies-that-are-real-now/Video- calling/photolist/52869590.cms 4. What is backpropagation really doing? https://www.youtube.com/watch?v=Ilg3gGewQ5U 5. http://www.andreykurenkov.com/writing/ai/a-brief-history-of-neural-nets-and-deep-learning/ 6. https://pmirla.github.io/2016/08/16/AI-Winter.html 7. https://tw.saowen.com/a/6cdc2f1279016e566832bb1234e06d321992dd1fabcdf4a2e0a3e16fc0dc09dc 8. https://tectales.com/ai/ai-used-to-identify-different-types-of-brain-injuries.html 9. Transformers, explained: Understand the model behind GPT, BERT, and T5 10. https://transformer.huggingface.co/