ITP-NYU :: 3/24/2016

Introduction, neural networks

Class notes

Introduction ("The whole class in 1 hour")

• Supervised and unsupervised learning
• Linear classifiers
• What are neural networks
  • Forward pass demo
  • MNIST forward pass
  • visualizing weights
  • Limitations of ordinary neural nets
• Convolutional neural networks (quick run-through)
  • Convolution + pooling layers
  • ofxCcv demo
  • Applications of activations
    • Probing neurons, Deepdream, Style transfer
    • t-SNE visualization

Demos

• Simple forward pass
• MNIST forward pass
• Confusion matrix demos

Further reading

• wekinator: [intro] + [quick walkthrough] + [showcase]
• neural networks for classifying digits (Michael Nielsen)
  • very thorough review of how basic NNs work
• quick step-by-step demo of training a neural networks
• (optional) how neural networks are trained (backpropagation)
• (optional but recommended to skim) brief history by andrey kurenkov