1: \begin{abstract}

2: Recent results in quantization theory show that the mean-squared

3: expected distortion can reach a rate of convergence of $\mathcal{O}(1/n)$, where $n$ is

4: the sample size [see, e.g.,

5: \textit{IEEE Trans. Inform. Theory} \textbf{60} (2014) 7279--7292

6: or

7: \textit{Electron. J. Stat.} \textbf{7} (2013) 1716--1746].

8: This rate is attained

9: for the empirical risk minimizer strategy, if the source distribution

10: satisfies some

11: regularity conditions. However, the dependency of the average

12: distortion on other

13: parameters is not known, and these results are only valid for

14: distributions over

15: finite-dimensional Euclidean spaces.

16: 

17: This paper deals with the general case of distributions over separable, possibly

18: infinite dimensional, Hilbert spaces. A condition is proposed, which

19: may be thought

20: of as a margin condition [see, e.g.,

21: \textit{Ann. Statist.} \textbf{27} (1999) 1808--1829], under which a

22: nonasymptotic

23: upper bound on the expected distortion rate of the empirically optimal

24: quantizer is

25: derived. The dependency of the distortion on other parameters of

26: distributions is

27: then discussed, in particular through a minimax lower bound.

28: \end{abstract}