1: \begin{abstract}

2: This paper establishes the consistency of  a family of graph-cut-based algorithms for clustering of data clouds. We consider point clouds obtained as samples of a ground-truth measure.  We investigate approaches to clustering based on minimizing objective functionals defined on proximity graphs of the given sample. Our focus is on functionals based on graph cuts like the Cheeger and ratio cuts. We show that minimizers of the these cuts converge as the sample size increases to a minimizer of a corresponding continuum  cut (which partitions the ground truth measure). Moreover, we obtain sharp conditions on how the connectivity radius can be scaled with respect to the number of sample points for the consistency to hold. We provide results for two-way and  for multiway  cuts. Furthermore we provide numerical experiments that illustrate the results and explore the optimality of scaling in dimension two.

3: %

4: %This paper establishes consistency for minimizers of balanced graph cut objective functionals. Specifically, we prove the following result. Draw $n$ data points independently distributed according to some density on a domain $D$ and construct an (appropriately scaled) neighborhood graph from these data points. Then the balanced cut of this graph converges almost surely  toward the continuum balanced cut of the domain $D$ in the limit of infinite sample size. The proof uses $\Gamma$-convergence techniques combined with a proper extension mechanism between the data points and the continuum domain. 

5: \end{abstract}

6: