Algorithm 1 DDAE Weight Training Algorithm

Input: DIFF value, network architecture, max_epoch, dropout_rate p, and learning rate $\alpha$; Output: Trained weights $\omega$ and b;  1: Randomly initialize $\omega$ and b;  2: while $epoch<Max\_epoch$ do  3:    Randomly select a mini-batch from inputs;  4:    // Forward propagation;  5:    // L is the number of layers of the DDAE;  6:    for l = 2:L-2 do  7:      if the current layer is a dropout layer then  8:         $r_j^{\left(l\right)}\sim Bernoulli\left(p\right)$  9:         ${\tilde{y}}^{\left(l\right)}=r^{\left(l\right)}·y^{\left(l\right)}$  10:       $z_i^{(l+1)}={\omega }_i^{(l+1)}{\tilde{y}}^{\left(l\right)}+b_i^{(l+1)}$  11:       $y_i^{(l+1)}=f\left(z_i^{(l+1)}\right)$  12:      else  13:         // The current layer is a hidden layer;  14:         $z^{\left(l\right)}={\omega }^{\left(l\right)}{y}^{\left(l\right)}+b^{\left(l\right)}$  15:         $y^{\left(l\right)}=f_l\left(z^{\left(l\right)}\right)$  16:      end if  17:    end for  18:    //Loss function;  19:    $Loss=-{\sum }_{i=1}^{output\_size}{y}_i·log\widehat{y_i}$  20: end while