juniorDeep Learning
What is backpropagation in neural networks?
Updated May 16, 2026
Short answer
Backpropagation is the algorithm used to calculate gradients and update neural network weights during training.
Deep explanation
Backpropagation applies the chain rule of calculus to compute how much each weight contributes to prediction error. The process involves a forward pass to compute predictions, loss calculation, backward propagation of gradients, and weight updates using optimization algorithms like gradient descent. Efficient gradient computation makes training deep neural networks feasible.
Real-world example
Backpropagation is used in image recognition, speech synthesis, and recommendation systems.
Common mistakes
- Confusing backpropagation with gradient descent
- backprop computes gradients while gradient descent updates weights.
Follow-up questions
- Why is the chain rule important?
- What happens during the forward pass?
- Can backpropagation work without differentiable activations?