Algorithm 1.

Train confident classifiers

Input: Data ${\{({\mathbf{x}}_i,y_i)\}}_{i=1}^n$ and its index set 𝒟 = {1, 2, . . . , n}
Output: Confident classifiers (w+,w−), easy samples ℰ and hard samples ℋ
1:	Init: ℰ ←𝒟; ℋ←∅;
2:	repeat
3:	(w+,w−) ← solve (2) with fixed ℰ and ℋ;
4:	Update easy and hard samples (ℰ,ℋ) using (1);
5:	Update relabels ${\eta }_j^{+},{\eta }_j^{-}\forall j\in \mathcal{H}$;
6:	until convergence or exceed max iteration