All labs

Machine learning 1

Let the computer find the spam line you drew by hand in Linear algebra 2, one small correction at a time — the perceptron, the first machine that learned from its mistakes.

Briefing

You already drew a spam line by hand in Linear algebra 2. Here you hand that job to the machine and watch it find the line itself, fixing one misclassified email at a time — the perceptron, the first thing that learned from its mistakes.

You'll learn to

  • Apply the perceptron's one rule to correct a wrong guess
  • Explain why each nudge reduces the mistake
  • Pick which mistakes to fix, and in what order
  • Show how one bad label can stop learning, and why test data is kept back
0 of 5 steps done0%
Start: One nudge
Before the machine learnsOptional warm-upLabelled examples, scores, what training changes, and why some examples are kept back. Ten minutes.

  1. One nudge

    Up next

    The whole learning rule, by hand

  2. Why a nudge helps

    The score always moves the right way

  3. Pick the nudges

    Which mistake should it learn from?

  4. One bad label

    When no line can do it

  5. What the machine learned

    From 1958 to today