1: \begin{abstract}

2: In many data analysis applications the following scenario is commonplace:

3: we are given a point set that is supposed to sample a hidden ground truth

4: $K$ in a metric space,

5: but it got corrupted with noise so that some of the 

6: data points lie far away from $K$ creating 

7: outliers also termed as {\em ambient noise}. One of the main

8: goals of denoising algorithms is to eliminate such noise 

9: so that the curated data lie within a bounded Hausdorff distance

10: of $K$. Popular denoising approaches such as deconvolution and thresholding 

11: %the two prevailing techniques for this problem suffer from the difficulty that they burden the user with setting 

12: often require the user to set several parameters and/or to choose an appropriate

13: noise model while guaranteeing only asymptotic convergence. 

14: Our goal is to lighten this burden as much as possible while

15: ensuring theoretical guarantees in all cases.

16: 

17: Specifically, first, we propose a simple denoising algorithm that requires only

18: a single parameter but provides a theoretical guarantee on the quality of the output on general input points. 

19: %under a sampling condition of the input points that is not any more

20: %restrictive than the known prevailing noise models. Under such sampling

21: %conditions, this parameter cannot be avoided. 

22: We argue that this single parameter cannot be avoided. 

23: We next present 

24: a simple algorithm that avoids even this parameter by paying

25: for it with a slight strengthening of the sampling condition on the input points which is not unrealistic.

26: We also provide some preliminary empirical evidence that our algorithms

27: are effective in practice. 

28: \end{abstract}

29: