Loading slide

Why Shallow Networks Failed

  1. 01Why Shallow Networks Failed
  2. 02Memory card
  3. 03What "deep" actually means
  4. 04The vanishing gradient
  5. 05A small step inside each neuron
  6. 06Why bother bending the number?
  7. 07The bend they chose
  8. 08How the flat ends starve the signal
  9. 09The absurdly simple fix
  10. 10Take a breath
  11. 11Other fixes
  12. 12Depth vs. width
  13. 13Memory card
  14. 14The research gap
  15. 15Reinforce your understanding
  16. 16Question: The vanishing gradient
  17. 17Question: Why does a simple fix work?
  18. 18Question: Why does depth matter?
  19. 19Quiz: answer
  20. 20The one idea to keep
  21. 21The gradient problem was solved. What was still missing was scale.
  22. 22Want to go deeper?
5 / 21
BackNext

A small step inside each neuron

To see why the signal faded, we have to zoom in on a single neuron. And to do that, we need one small idea first: a function.

A function is just a fixed rule for turning one number into another. You put a number in, you get a number out, and the rule never changes its mind. "Double it" is a function: put in 3, you get 6. Put in 10, you get 20. That is all a function is. A little machine with a number going in one end and a number coming out the other.

Now back to the neuron. Remember what it does from the last module: it adds up its inputs into a single total. But it has one more step after that, before it passes its answer along. It runs that total through a function. The number that comes out is what it hands to the next layer.

This final rule has a name, the . Every neuron has one. It is a tiny step, easy to overlook.

Which raises a fair question. If the neuron already added everything up, why not just pass that total straight on? Why run it through anything at all?

Citations(2)↓
  1. 1. nature.com
  2. 2. deeplearningbook.org