Dear mentor! we have seen different variants that CNN is working better because they reduced the number of parameters. One reason is " Sparsity connection "
First of all that what i understand from sparsity connection is that ( the pixel at the next layer is not connected to all input image means only local group is connected to that pixel or if i say like it is not trying to get information from the full image every time )
but how sparsity connection is reduces the number of total parameters .