Gradients
A real model has millions of weights, not one. A gradient is a list of slopes, one for each weight. It shows which way to turn all the weights at once. In this lesson you find a gradient, walk downhill with it, and see why the step size matters.
New words
- Partial derivative
- The slope when you change one input and hold the others fixed.
- Gradient
- A list of all the partial derivatives. It points uphill.
- Learning rate
- The size of each step you take against the gradient.
- Contour map
- A top-down map where each line joins points of equal height.
- Gradient descent
- Taking many small steps against the gradient to lower the error.
What you will learn
- Why do AI models need gradients?
- What if there are two inputs?
- What is a partial derivative?
- What is a gradient?
- How do you walk downhill?
- How big should a step be?
- See the path on a map
- Try it: roll the ball down the bowl
- Common mistakes
- Explain it to a friend
- Key takeaways
10 quiz questions at the end.
The full lesson, with interactive demos and the quiz, opens after you sign in with Google. It is free.