All labs

Machine learning 2

Hit the perceptron's wall on XOR, get past it with a new feature and then a hidden layer, and meet the trap waiting for any model that tries too hard: memorising.

Briefing

You hit a wall: no straight line can separate XOR, the problem that stalled neural networks for years. You get past it twice — first with a cleverer feature, then with a hidden layer — and meet the trap that catches any model trying too hard.

You'll learn to

  • Show why one straight line can't solve XOR
  • Add a feature that makes the problem separable
  • Stack neurons into a hidden layer that solves it
  • Recognise overfitting, and tell memorising from learning
0 of 6 steps done0%
Start: One line is not enough
Before the networkOptional warm-upThe logic gates again, what makes a neuron fire, what a feature is, and why a perfect practice score can mislead. Ten minutes.

  1. One line is not enough

    Up next

    The perceptron meets XOR

  2. A new feature

    Measure something else, and a line works again

  3. The memoriser

    A perfect score on what it has seen

  4. Two lines

    The pieces of a hidden layer

  5. Solve XOR

    A tiny neural network, set by hand

  6. What "deep" means

    From XOR to today's networks