Exercise

How many parameters in a CNN?

We need to know how many parameters a CNN has, so we can adjust the model architecture, to reduce this number or shift parameters from one part of the network to another. How many parameters would a network have if its inputs are images with 28-by-28 pixels, there is one convolutional layer with 10 units kernels of 3-by-3 pixels, using zero padding (input has the same size as the output), and one densely connected layer with 2 units?

Instructions

50 XP

Possible answers