1
2
3

Fast or Slow?

Control how fast the network learns.

Step 1 of 3

What you'll learn in this level

  • What the learning rate is and what it controls
  • Why steps that are too big and too small both cause trouble
  • What gradient descent means

How Big Is One Step?

Imagine walking down a foggy mountain to find the bottom. If you take huge steps, you might overshoot the lowest point. If you take tiny steps, it will take forever.

In AI, this step size is called the 'Learning Rate'. It's one of the most important settings!

Finding the Right Size

  1. Too Fast

    Big steps easily jump right past the target.

  2. Too Slow

    Tiny steps are safe but take a very long time.

  3. Just Right

    The right step size gets you to the goal fast and exactly.

  4. Optimization

    Looking for the smallest error is like looking for the bottom of a valley.

Can you find the perfect learning rate?

Guess first, then experiment

Big steps get you to the goal faster. So should you just turn the learning rate up to the maximum?