TwiddleLab / Artificial Intelligence / How a Neural Network Thinks

Learning from Examples

Lesson 1 of 10 · 5 min · Free, no account needed

What you will learn

• AI learns from labeled examples, not from rules someone typed in.
• All the examples together are the training set. The machine copies every pattern in it, even accidental ones.
• One wrong label teaches a wrong habit. To fix the machine, fix the labels and add better examples.

Start the lesson

How the lesson goes

1. Your phone just moved 3 of these messages to Spam.

   Nobody wrote a rule for every scam. Guess which 3 it moved: tap them, then Check.

2. Be the teacher. Sort 8 messages for the machine.

   Drag each card into Spam or Inbox. Watch what the machine picks up from you.

3. Now your machine meets mail it has never seen.

   Press Run. It sorts each message using only the clues it got from your piles.

4. What you just did has names.

   Tap to reveal them.

   Each message with its answer is a labeled example. Spam and Inbox are the labels.

   All your labeled examples together are the training set.

   Finding patterns in a training set and using them on new things is machine learning.

5. Break it on purpose: send the homework question to Spam.

   You may flip one label in your training set. Find the one that does it.

6. Game: get your filter to 10 out of 10 on new mail.

   It starts with your broken label. Flip cards and drag in extra examples. It retrains after every move.

7. A music app wants to learn your taste. Build its learning line.

   Put the 5 stations in order. It's the same recipe you just used for spam.

Next lesson: Features as Coordinates · Part of the course: How a Neural Network Thinks

Lessons are interactive and need JavaScript. Turn it on to play them.