Step 1: $\forall l$, set $△{\mathit{W}}^{\left(l\right)}=0,△{\mathit{b}}^{\left(l\right)}=0;$

Step 2: for $i=1$ to m,

a. Calculate the partial derivatives $▽{}_{{\mathit{W}}^{\left(l\right)}}\mathcal{L}\left(\mathbf{x},{\mathbf{x}}^{\prime }\right)$ and $▽{}_{{\mathit{b}}^{\left(l\right)}}\mathcal{L}\left(\mathbf{x},{\mathbf{x}}^{\prime }\right);$

b. Partial differential superposition:

$△{\mathit{W}}^{\left(l\right)}=△{\mathit{W}}^{\left(l\right)}+▽{}_{{\mathit{W}}^{\left(l\right)}}\mathcal{L}\left(\mathbf{x},{\mathbf{x}}^{\prime }\right),$ $△{\mathit{b}}^{\left(l\right)}=△{\mathit{b}}^{\left(l\right)}+▽{}_{{\mathit{b}}^{\left(l\right)}}\mathcal{L}\left(\mathbf{x},{\mathbf{x}}^{\prime }\right);$

Step 3: Update weight parameters:

a. Renew ${\mathit{W}}^{\left(l\right)}={\mathit{W}}^{\left(l\right)}-\alpha ·△{\mathit{W}}^{\left(l\right)}$, ${\mathit{b}}^{\left(l\right)}={\mathit{b}}^{\left(l\right)}-$ $\alpha ·△{\mathit{b}}^{\left(l\right)}$;

b. Disrupt the order of patches in the dataset and finish the current iteration epoch;