juniorNeural Networks
What is backpropagation?
Updated May 17, 2026
Short answer
Backpropagation is an algorithm for computing gradients and updating weights in neural networks.
Deep explanation
It uses chain rule of calculus to compute gradients from output layer back to input layer, enabling optimization via gradient descent.
Real-world example
Used in training deep learning models like GPT and CNNs.
Common mistakes
- Confusing backpropagation with gradient descent.
Follow-up questions
- What is gradient descent?
- Why is chain rule important?