How do I derive the back-propagation algorithm in Convolutional Neural Network?

Answer (1 of 5): For a simple CNN like the one below: If the stride of filter (F) is 2 and assuming c_{10} is the output of the MaxPool function, the weights w_{00}, w_{10}, w_{01},w_{11} of F are updated as: Note that we assume a true convolution operation and rotate the filter by 180°. Here e...