Algorithm 1: The learning algorithm for SAMM

Input: Training set ${\left\{{\mathbf{A}}_i,y_i\right\}}_{i=1}^N$, low-rank coefficient $\beta$, loss penalty coefficient $\rho$, step size $\delta$. Output $\mathbf{W}$, b 1. Initialize. ${\mathbf{W}}^{\left(0\right)}=\mathbf{0},{\mathbf{Z}}^{\left(0\right)}=\mathbf{0},{\mathbf{\Lambda }}^{\left(0\right)}=\mathbf{0},k=1$ While not converging do 2. Update ${\mathbf{Z}}^k$ with Equation (8) 3. Update ${\mathbf{W}}^k$ and b with Equations (14) and  (17) 4. Update ${\mathbf{\Lambda }}^{\left(k\right)}$ with Equation (6) 5. $k=k+1$ End 6. Return $\mathbf{W}$, b