Algorithm 1: Label recovery algorithm based on subset label iterative propagation and replacement (SLIPR).
Require: dataset $X$, noisy label set $Y\text{'}$, parameters $\gamma$, $T$, $S$.
Ensure: recovered label set $Y^{\ast }$.
1: for$s$ = $1$ to $S$do
2:   for$t$ = $1$ to $T$do
3:    $m=\mathrm{randperm}(N)$
4:    $l=\mathrm{round}(N\ast \gamma )$
5:    $Y_L{\text{'}}^{(t,s)}=Y\text{'}(m(1:l))$
6:    $Y_U{\text{'}}^{(t,s)}=0$
7:    ${X_L}^{(t,s)}=X(m(1:l),:)$
8:    ${X_U}^{(t,s)}=X(m((l+1):\mathrm{e}\mathrm{n}\mathrm{d}),:)$
// Construct the neighborhood graph
9:    $W=\mathrm{construct}({X_L}^{(t,s)})$
10:    while not converged do
11:     Update $Q$ by Eqs. (3), (4), (8)
12:    end while
13:    $X_U{\text{'}}^{(t,s)}=Q^T{X_U}^{(t,s)}$
14:    $X_L{\text{'}}^{(t,s)}=Q^T{X_L}^{(t,s)}$
// Classify samples by KNN
15:    $Y_U{\text{'}}^{(t,s)}=\mathrm{knnclassify}(X_U{\text{'}}^{(t,s)},X_L{\text{'}}^{(t,s)},Y_L{\text{'}}^{(t,s)})$
16:    $L=[Y_L{\text{'}}^{\left(t,s\right)};Y_U{\text{'}}^{(t,s)}]$
17:   end for
18:   for$i$ = $1$ to $N$do
19:    ${F_i}^{\ast (s)}=\{{\bigcup }_{t=1}^T{L_{i,t}}^{(s)}\}\cup l_i^0$
20:    ${y_i}^{(s)}={\arg \max F_i^{\ast }}^{(s)}$
21:   end for
22:   $Y\text{'}=\{{y_1}^{\left(s\right)};{y_2}^{\left(s\right)};\cdots ;{y_N}^{(s)}\}$
23: end for
24:$Y^{\ast }=Y\text{'}$
25: return$Y^{\ast }$