1: \begin{abstract}%   <- trailing '%' for backward compatibility of .sty file

2: %\textcolor{blue}{}

3: We study universal consistency and 

4: convergence 

5: rates

6: %of convergence 

7: of simple nearest-neighbor prototype rules for the problem of multiclass classification in metric spaces. 

8: We first show that a novel data-dependent partitioning rule,

9: named \proto{}, is universally

10: % strongly 

11: consistent in any 

12: metric space that admits a universally consistent rule.

13: %separable metric 

14: %space. 

15: %This rule, which we call 

16: %This rule, which we call 

17: \proto{} is a significant simplification of \optinet{}, a recently proposed compression-based algorithm that, 

18: to date,

19: was 

20: %was the first 

21: the only algorithm known to be universally consistent in such a general setting.

22: Practically, \proto{} is simpler to implement and enjoys reduced computational complexity.

23: 

24: We then 

25: proceed to 

26: study convergence rates of the excess error probability.

27: We first obtain 

28: %derive convergence

29: rates for the standard $k$-NN rule under a margin condition and a new generalized-Lipschitz condition.

30: The latter is an extension of a recently proposed modified-Lipschitz condition from $\R^d$ to metric spaces. Similarly to the modified-Lipschitz condition, the new condition avoids any boundness assumptions on the 

31: %support of the underlying 

32: data distribution.

33: %It avoids any boundness assumptions on the support of the underlying data distribution. 

34: %As sucand generalizes recent results to more general metric spaces.

35: While obtaining rates for 

36: %the universally consistent 

37: \proto{} 

38: %rule 

39: %under these conditions 

40: is left open,

41: we show that a second prototype rule that hybridizes between $k$-NN and \proto{}

42: %we call \hybrid{}, 

43: achieves the same rates as $k$-NN while enjoying similar computational advantages as \proto{}.

44: \textcolor{black}{However, as $k$-NN, 

45: this hybrid rule

46: %\hybrid{} 

47: is not consistent in general.}

48: \end{abstract}