1: \begin{abstract}

2:     In this work we study the semi-supervised framework of confidence set classification with controlled expected size in minimax settings.

3:     We obtain semi-supervised minimax rates of convergence under the margin assumption and a H\"older condition on the regression function.

4:     Besides, we show that if no further assumptions are made, there is no supervised method that outperforms the semi-supervised estimator proposed in this work.

5:     We establish that the best achievable rate for any supervised method is $n^{-1/2}$, even if the margin assumption is extremely favorable.

6:     On the contrary, semi-supervised estimators can achieve faster rates of convergence provided that sufficiently many unlabeled samples are available.

7:     We additionally perform numerical evaluation of the proposed algorithms empirically confirming our theoretical findings.

8:     \end{abstract}

9: