Loading slide

The Neuron Idea

  1. 01The Neuron Idea
  2. 02A familiar idea
  3. 03Same era, different bet
  4. 04A neuron, at its simplest
  5. 05Adding things up
  6. 06The first artificial neuron
  7. 07Strength, not just presence
  8. 08The signal itself has a strength
  9. 09Multiply, then add
  10. 10Just a recipe
  11. 11What a weight really is
  12. 12Where the numbers come from
  13. 13Back to the real thing
  14. 14The Perceptron
  15. 15How it learned
  16. 16What training actually is
  17. 17Why this was different
  18. 18What Rosenblatt claimed
  19. 19The shape of hype
  20. 20Twelve years of real work
  21. 21The verdict
  22. 22What the proof didn't say
  23. 23What it could do
  24. 24What it could not do
  25. 25The reason
  26. 26The limit was real
  27. 27Before we go further
  28. 28Memory card
  29. 29What if we added another?
  30. 30Many dumb things
  31. 31Layers
  32. 32More Layers
  33. 33The whole thing
  34. 34A network of neurons
  35. 35A small group kept going
  36. 36Reinforce your understanding
  37. 37Question: What is learning?
  38. 38Question: The limit
  39. 39Question: Training a network
  40. 40Quiz: answer
  41. 41The idea didn't die. It waited.
  42. 42Want to go deeper?
27 / 41
BackNext

Before we go further

Take a breath. A lot has happened in this chapter. Let's make sure the foundations are solid before we build on them.

What a neuron does. It takes in several signals, multiplies each one by a weight, adds them all up, and fires if the total is high enough. That's the whole operation.

What a weight is. A number that controls how much a signal counts. High weight means "I take this seriously." Low weight means "I barely listen to this." The weights are where the neuron keeps its memory of what matters.

What training is. Making a guess, measuring how wrong it was, and nudging every weight slightly in the direction of a better answer. Do that thousands of times, and the weights slowly settle into something that works. Nobody sets them there. The machine corrects its way there, the same way you corrected your way to balance on a bike.

What the Perceptron was. One neuron with adjustable weights, learning from examples. A real advance, but with a hard ceiling built into the design.

What that ceiling was. One neuron can only score inputs independently. It has no way to ask what two inputs mean together. That isn't fixable with more training. It's a limit of the design itself.

Okay. Now we can go further.

Citations