MEPX
Chapter 2 of 10All chapters
  1. 1The artificial neuron
  2. 2Layers and depth
  3. 3How it learns
  4. 4Activation functions
  5. 5Architectures
  6. 6Keeping it general
  7. 7Training in practice
  8. 8Transfer learning
  9. 9What they cannot do
  10. 10Learning it yourself

Chapter 2 of 10

Layers and depth

Building representations.

Stacking

Each layer transforms the output of the previous one. Early layers detect simple patterns and later ones combine them into complex features, without anyone specifying what those should be.

  • In vision: edges, then textures, then parts, then objects.
  • Width is how many units per layer; depth is how many layers.

Learned features

The network discovers its own representation of the input. That is the real shift from earlier approaches, where features were designed by hand.