Designing structure of CNN

by Shenghao Wu   Last Updated April 15, 2017 18:19 PM

Does anyone know any rules of thumb (i.e.

  • How many convolutional layers?
  • How many normalization layer?
  • Area of the receptive fields?
  • etc)

to design a convolutional neural network given a specific dataset?

If I change my out put from two-class to three-class, how should I modify the structure accordingly?



Related Questions


K-Fold CV: Errant fold refusing to train

Updated April 16, 2018 08:19 AM


Evaluating neural network for certain task

Updated May 10, 2018 17:19 PM