Algorithm 2 Fine-tuning Procedure
1: $f_{target}$	▹ Pretrained student model
2: D	▹ Target dataset
3: $N_{fold}$	▹ Total number of folds
4: $N_{epoch}$	▹ Total number of training epochs
5: $N_{stop}$	▹ Patience number for early stopping
6: procedure Fine-tune($f_{target},D,N_{fold},N_{epoch},N_{stop}$)
7:       for fold in range [1, $N_{fold}$] do
8:             $X_{train},Y_{train},X_{val},Y_{val},X_{test},Y_{test}$ = LoadData($D,\mathit{fold}$)
9:             for $epoch$ in range [1, $N_{epoch}$] do
10:                 for i in range [1, $\left|X_{train}\right|$] do
11:                       $L_{train}=L_{focal}(f_{target}\left(X_{train}\left(i\right)\right),Y_{train}\left(i\right))$
12:                       Backpropagate$(L_{train},f_{target})$
13:                 $L_{Val}=\frac{1}{\left|X_{val}\right|)}{\sum }_{j=1}^{\left|X_{val}\right|}{L}_{focal}(f_{target}(X_{val}\left(j\right),Y_{val}\left(j\right))$
14:                 CheckForEarlyStopping($L_{val},N_{stop}$)
15:           $A_{test}=\frac{1}{\left|X_{test}\right|}{\sum }_{k=1}^{\left|X_{test}\right|}$Accuracy$(f_{target}\left(X_{test}\left(k\right)\right),Y_{test}\left(k\right))$
16:      $A_{average}$ = $\frac{1}{N_{fold}}{\sum }_{fold=1}^{N_{fold}}{A}_{test}\left(fold\right)$
17:      return $A_{average}$
* In this study, $N_{fold}=5$, $N_{epoch}=200$, and $N_{stop}=20$.