Table 1. Algorithm of HDBN.

Input:

data ,

number of training data ; number of test data ;

number of layers ; number of epochs ;

number of units in every hidden layer ;

number of groups in every convolutional hidden layer ;

hidden layer ;

convolutional hidden layer ;

parameter space ;

biases , ; momentum and learning rate ;

Output:

deep architecture with parameter space

1. Greedy layer-wise unsupervised learning

for ; ; do

for do

for do

Calculate the non-linear positive and negative phase:

if then

Normal calculation.

else

Convolutional calculation according to Eq. 6 and Eq. 7.

end if

Update the weights and biases:

end for

end for

end for

2. Supervised learning based on gradient descent