Algorithm 1. Node Random Pruning Algorithm.

Given the parameters to be deleted, $\mathit{W}=[w_1,\dots ,w_L]$,$b=[b_1,\dots ,b_L]$,$\mathit{\beta }=[{\beta }_1,\dots ,{\beta }_L]$,

where L is the number of node; the initial accuracy of the network, $\alpha$; test data $X,Y$;

setting an acceptable accuracy $ϵ=\alpha -0.01$; and the maximum number of nodes that

can be deleted, $N_{max}=\lfloor 0.2\ast L\rfloor$.

1. Initialize $states\leftarrow zeros(N_{max}+1)$,$states\left[0\right]\leftarrow \alpha$,$diclist\leftarrow$$\left[\right]$

2. For $n=1$ To $N_{max}+1$ Do

3.  $num=C_L^n$,$accuracy\_and\_indices\leftarrow$$\left\{\right\}$,$accurac{y}_{l}ist\leftarrow$$\left[\right]$

4.   If  $num>10000$

5.      $num=10000$

6.   For  $k=1$  To  $num$  Do

7.     Randomly choice $indices$ of n nodes from original nodes.

8.     Remove the corresponding nodes from the original list

based on the $indices$ of the nodes to be deleted.

9.     ${\mathit{W}}^{*}=$ delete $(\mathit{W},indices)$

10.     $b^{*}=$ delete $(b,indices)$

11.     ${\mathit{\beta }}^{*}=$ delete $(\mathit{\beta },indices)$

12.     Calculate the accuracy of the model after removing the nodes.

13.     $accuracy=$ accuracy_score $({\mathit{W}}^{*},b^{*},{\mathit{\beta }}^{*},X,Y)$

14.     If  $accuracy\ge ϵ$

15.      $accuracy\_and\_indices\leftarrow$ append $\{accuracy:indices\}$

16.      $accuracy\_list\leftarrow$ append $\left[accuracy\right]$

17.  End For (corresponds to Step 4)

18.  $diclist\leftarrow$ append $[accuracy\_and\_indices]$

19.  $states\left[n\right]=$ max $\left(state\right[n-1],$ max $(accuracy\_list))$

20. End For (corresponds to Step 2)

21. Return $states,diclist$.