Every gray box can represent the computations for one pixel (3 for one dimensional, 9 for 2D) - not all the pixels are necessarily dependent on all other pixels, so that is why not all the input boxes are connected to every output element. We can then represent a convolution as a dot product, the weights of the parameters of the pixel.
Every gray box can represent the computations for one pixel (3 for one dimensional, 9 for 2D) - not all the pixels are necessarily dependent on all other pixels, so that is why not all the input boxes are connected to every output element. We can then represent a convolution as a dot product, the weights of the parameters of the pixel.