Algorithm 2 Semi-Supervised Active Learning

Input: initial train set L, unlabelled validation set U, independent test set T

Output: predicted labels for the test set

1:$\Theta \leftarrow clf.fit(L)$▹ Learns model on initial training set 2:while SC not met do 3:    selection by Q, of most informative sample: $x^{*}$ 4:    ask Oracle for $x^{*}$’s label 5:    $L\leftarrow L\cup \left\{x^{*}\right\}$▹ Augments the model’s training set with $x^{*}$ 6:    $U\leftarrow U\setminus \left\{x^{*}\right\}$▹ Removes $x^{*}$ from unlabelled samples 7:    $\Theta \leftarrow clf.fit(L)$▹ Updates model 8:    automatically label confident samples C in U 9:    $L\leftarrow L\cup \left\{C\right\}$▹ Augments the model’s training set with C 10:    $U\leftarrow U\setminus \left\{C\right\}$▹ Removes C from unlabelled samples 11:    $\Theta \leftarrow clf.fit(L)$▹ Updates model 12:    return $clf.predict(T)$▹ Returns predicted labels for the test set 13:end while