Dropout and max-norm constraint

I’m trying to re-implement some network to classify SVHN data according to this paper.
‘Dropout: A Simple Way to Prevent Neural Networks from Overfitting’ (http://jmlr.org/papers/volume15/srivastava14a.old/srivastava14a.pdf)
In this paper, I have 2 questions.

  1. This paper said they use 3 conv layer and 2 fc layer. but the number of dropout p is 6. (0.9, 0.75, 0.75, 0.5, 0.5, 0.5).
    Can anyone explain about this ?

  2. What is the max-norm constarint and how can use this in pytorch?