TwiddleLab / Courses / Artificial Intelligence

How a Neural Network Thinks

Teach a tiny network, one example at a time, and see how it learns.

Beginner · 10 lessons · Free, no account needed

Start the course

Lessons

1. Learning from Examples (5 min): AI learns from labeled examples, not from rules someone typed in.
2. Features as Coordinates (7 min): A feature is a number measured from a thing, like weight or redness.
3. A Line Separates the Classes (7 min): A line that splits the feature space into classes is a decision boundary.
4. Measuring Wrongness (6 min): The gap between a point and the line is an error.
5. Rolling Down the Loss (7 min): Feel the slope, step downhill, repeat: that's gradient descent.
6. The Learning Loop (7 min): One training step is predict, measure, adjust, always in that order.
7. A Neuron Is a Weighted Vote (7 min): A neuron adds up its inputs, each times its weight, and fires when the sum reaches the threshold.
8. Networks Solve XOR (7 min): One neuron draws one straight line, so it can't solve XOR.
9. Why Networks Get Fooled (7 min): A model that learns its training data too closely, noise and all, is overfitting: perfect on old data, bad on new.
10. Project: Train Your Own (8 min): Accuracy is the share of points a network gets right. The test set gives the honest one.

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