Convolutional kernels are matrix-multiplied over and over again in a sliding window across the feature map. The fully connected layer is matrix-multiplied exac

Convolutional kernels are matrix-multiplied over and over again in a sliding window across the feature map. The fully connected layer is matrix-multiplied exactly once. So the number of times each element of the weight matrix is used depends on the size of the layer for FC, but depends on the size of the feature map for Conv layers. Usually the feature map is much larger than the kernel, for the majority of layers, so the kernel weights are re-used a lot. The other possible interpretation is that typically an FC is only applied on a reduced dimensional embedding, after reducing the input data dimensionality iteratively by several convolutional layers. So there is overall more computation done by Conv layers, to get that final embedding, which his then fed to one or two FC layers for taking global context into account -- so usually you'll find more Conv layers than FC layers in a given ANN architecture, thus they make up a larger percentage of the overall computation.