. 2020 Aug 14;8:867. doi: 10.3389/fbioe.2020.00867

Table 3.

The algorithm details of Adversarial Learning.

Algorithm 2: Adversarial learning
For each X in training samples: 1. Calculate the forward loss of X and get the gradient g by back propagation; g=∇_XL(θ,X,Y) 2. Calculate r_adv according to the gradient of the embedding matrix X and add it to the current embedding, which is equivalent to X+r_adv; r_adv=ϵ∙g/\|\|g\|\|2 X_adv=X+r_adv 3. Calculate the forward loss of X_adv, backpropagate to obtain the gradient of the confrontation, and add to the gradient of step 1; 4. Restore embedding to the value at step 1; 5. Update the parameters according to the gradient of step 3.

Algorithm 2: Adversarial learning

For each X in training samples:
1. Calculate the forward loss of X and get the gradient g by back propagation;
     g=∇_XL(θ,X,Y)
2. Calculate r_adv according to the gradient of the embedding matrix X and add it to the current embedding, which is equivalent to X+r_adv;
     r_adv=ϵ∙g/||g||2
     X_adv=X+r_adv
3. Calculate the forward loss of X_adv, backpropagate to obtain the gradient of the confrontation, and add to the gradient of step 1;
4. Restore embedding to the value at step 1;
5. Update the parameters according to the gradient of step 3.