Step 1. Initialize, given parameters $\rho ,\sigma \in \left(0,1\right)$ , ${\eta }_0>0$ , ${\eta }_1>0$ , select $w_0$ , set $k=0$ , given error $\epsilon >0$ and maximum number of iterations N.

Step 2. PCA is used to reduce the dimension of the data set.

Step 3. By calculating (8), get $w^{k+1}$ and use Moreau-Yosida regularization obtain $s^{k+1}=s^{*}\left(w^{k+1}\right)$ .

Step 4. Update the Lagrange multiplier ${\lambda }^{k+1}={\lambda }^k-{\sigma }_k\left(s^{k+1}-B{w}^{k+1}-d\right)$ .

Step 5. Select ${\sigma }_{k+1},{\sigma }_k\le {\sigma }_{k+1}<+\infty$ , go to step 2.