Train your own model

Now it's your turn to train an AI model. You don't need to be a scientist to train an A I model. Google's free Teachable Machine runs in your web browser. You teach it with examples from your camera, and it learns to recognise them.

How Teachable Machine works

First, go to teachable machine dot with google dot com. It works best on a computer with a camera. Second, choose Image project. Third, name your classes, the groups you want it to tell apart, like thumbs up and thumbs down. Fourth, record lots of examples of each, from different angles. And fifth, click train, then test it by showing it something new. Here, you've recorded about fifty pictures of thumbs up, and fifty of thumbs down. Click Train model, and wait while it learns. Now hold up your thumb. The model shows how sure it is for each class. Try tricky tests too, like a different background, or a friend's hand.

Tips for a good model

Record lots of examples for every class. Vary the background, the light and the angle, so it learns the real pattern, not the room. Keep your classes balanced, with about the same number of examples each. And test it with new things it hasn't seen. Don't worry if your model makes mistakes. Notice when it happens, and think about why. Is a class missing examples? Is the background confusing it? That's exactly how real A I engineers work.

Quick check

Quick check. Your model mixes up thumbs up and thumbs down when you change rooms. What helps? Record more examples in different rooms and light, so it learns the hand, not the background. Options: Record more examples in different rooms and light; Delete the model; Use fewer examples; Only test in the dark.

Recap

Teachable Machine trains real models in your browser. Classes are the groups you want it to tell apart. Use lots of varied, balanced examples. Then test, notice the mistakes, and improve.

I've read this lesson