1: \begin{abstract}

2: It has been recently shown that, under the margin (or low noise)

3: assumption, there exist classifiers attaining fast rates of

4: convergence of the excess Bayes risk, i.e., the rates faster than

5: $n^{-1/2}$. The works on this subject suggested the following two

6: conjectures: (i) the best achievable fast rate is of the order

7: $n^{-1}$, and (ii) the plug-in classifiers generally converge slower

8: than the classifiers based on empirical risk minimization. We show

9: that both conjectures are not correct. In particular, we construct

10: plug-in classifiers that can achieve not only the fast, but also the

11: {\it super-fast} rates, i.e., the rates faster than $n^{-1}$. We

12: establish minimax lower bounds showing that the obtained rates

13: cannot be improved.

14: \end{abstract}

15: