Subsampling Sparse Graphons Under Minimal Assumptions

TL;DR

This paper develops a general theory for subsampling sparse network data generated by graphons, demonstrating validity under minimal assumptions and analyzing eigenvalue distributions in sparse regimes.

Contribution

It introduces a minimal-assumption framework for subsampling sparse graphons and proves validity for vertex and p-subsampling procedures.

Findings

Validity of subsampling under minimal conditions

Conditions for uniform validity of subsampling methods

Limiting distributions for eigenvalues in sparse graphons

Abstract

We establish a general theory for subsampling network data generated by the sparse graphon model. In contrast to previous work for networks, we demonstrate validity under minimal assumptions; the main requirement is weak convergence of the functional of interest. We study the properties of two procedures: vertex subsampling and $p$-subsampling. For the first, we prove validity under the mild condition that the number of subsampled vertices is $o(n)$. For the second, we establish validity under analogous conditions on the expected subsample size. For both procedures, we also establish conditions under which uniform validity holds. Furthermore, under appropriate sparsity conditions, we derive limiting distributions for the nonzero eigenvalues of the adjacency matrix of a low rank sparse graphon. Our weak convergence result immediately yields the validity of subsampling for the nonzero eigenvalues under suitable assumptions.