How does backpropagation work in CNNs?

Backpropagation in CNNs computes gradients of filters and updates them using chain rule to minimize loss.

Backpropagation in CNNs works by computing gradients of the loss function with respect to each filter weight. The chain rule is applied layer by layer, propagating errors from output to input. Each convolution filter is updated based on how much it contributed to prediction error, enabling learning of meaningful feature detectors.