Algorithm 2 Multi-label learning (ML)
Input: Training data $D={\{({\mathbf{x}}_i,{\mathbf{y}}_i)\}}_{i=1}^N$, PL matrix W1, Lagrange multiplier of ADMM ρ and U, learning rate η2, penalty parameter β2, and accuracy control parameter μ.
Output: ML matrix W2 ∈ ℝD×L.
1:	${\mathbf{W}}_2^{(0)}=\frac{1}{D}{\mathbf{1}}_{D\times L},{\mathbf{V}}^{(0)}=\frac{1}{D}{\mathbf{1}}_{D\times L},a^{(0)}=1,t=0$; // Init.
2:	while not convergence do
3:	${\mathbf{U}}^{(t)}=(1-a^{(t)}){\mathbf{W}}_2^{(t)}+a^{(t)}{\mathbf{V}}^{(t)}$;
4:	Hμ = 0L×D;
5:	for i = 1, … , N do
6:	${\mathbf{z}}_i=\min (1,\max (-1,\frac{{\mathbf{U}}^{(t)┬}{\mathbf{x}}_i}{\mu }))$;
7:	$q_i={\mathbf{z}}_n^{┬}{\mathbf{U}}^{(t)┬}{\mathbf{x}}_i-\frac{\mu }{2}\Vert {\mathbf{z}}_i{\Vert }_2^2$;
8:	${\mathbf{H}}_{\mu }={\mathbf{H}}_{\mu }+q_i({\mathbf{z}}_i{\mathbf{x}}_i^{┬})$;
9:	end for
10:	${\mathbf{V}}^{(t+1)}={\mathbf{V}}^{(t)}-\frac{1}{{\eta }_2}({\mathbf{H}}_{\mu }^{┬}-\mathbf{u}+\rho ({\mathbf{W}}_1-{\mathbf{U}}^{(t)}))+\nabla PC({\mathbf{U}}^{(t)})$;
11:	${\mathbf{W}}_2^{(t+1)}=(1-a^{(t)}){\mathbf{W}}_2^{(t)}+a^{(t)}{\mathbf{V}}^{(t+1)}$;
12:	$a^{(t+1)}=\frac{2}{t+1}$;
13:	t = t + 1;
14:	end while