Lesson 19 of 20 · about 17 min
The chain rule and backpropagation
To train the park's wait-time model by gradient descent, you need the gradient of its error with respect to every setting. Backpropagation computes all of them in one sweep, by running the chain rule backwards through the calculation. It's how every neural network learns. You'll step gradients back through a small computational graph by hand, then code the forward and backward passes yourself.
🔓 Unlock this lesson
This lesson is part of the DoingCamp Pass. Members get every lesson in every course, with all the interactive visuals, coding exercises and challenges.
In this lesson you'll learn:
- Computational graph
- Backpropagation
Free to try in F09 Calculus for CS:
The first 3 lessons of every course are free, no account needed.