Usually Back propagation NN has the problem of vanishing gradients. I found that Convolutional NN (CNN) some how get rid of this vanishing gradient problems (why?).

Also in some papers some pretraining approaches have been discussed for CNN. Could somebody explain to me the following?

  1. the resons for pretraining in CNN and
  2. what are the problems/limitations with CNN?
  3. any relavent papers talking about the limitation of CNN?
  • $\begingroup$ Without knowing which particular papers you have in mind, it will be hard to answer this question. Can you elaborate on where you found these claims, and in what context? $\endgroup$ – Sycorax Aug 13 '18 at 1:09
  • $\begingroup$ What kind of "limitations of CNNs" do you have in mind? $\endgroup$ – Jan Kukacka Aug 13 '18 at 9:38