if you apply (taking dot product) a 4 x 4 x 3 filter to an image of 32 x 32 x 3. how many float operations (excludes adding the bias term) would be needed to compute one node of the next layer?