What is deep learning?

  • linear algebra + optimization
  • Can demostrate with Excel
  • backprop is just chain rule
  • optimize a and b has different learning rate
  • sum of squared of derivitives
  • learning rate goes from 1/10,000 to 2
  • when get to the bottom it’s easy to overs step
  • learning rate schedule
  • average derivitive goes large means it’s out of scope
  • momentum 20:37
  • Tips 23:29
    • keep it simple
    • can visulize
    • small changes can have big effect

What is deep learning for? 25:44

  • layers sucssively discover progresively more complex features from input data
  • learn things with unstructured data
    • images
    • natual language
    • 3D
    • audio
    • signal and time seriers

deep learning vs internet company,

Think deep learning for what.

Unsupervised learning 39:10 Siamese network word2vec