Counting Stars Is Constant-Degree Optimal for Detecting Any Planted Subgraph: Extended Abstract

Abstract

We prove that whenever $p=\Omega(1)$ and for any graph $H$, counting $O(1)$-stars is optimal among all constant degree polynomial tests in terms of strongly separating an instance of $G(n,p),$ from the union of a random copy of $H$ with an instance of $G(n,p).$ Our work generalizes and extends multiple previous results on the inference abilities of $O(1)$-degree polynomials in the literature.

Cite

Text

Yu et al. "Counting Stars Is Constant-Degree Optimal for Detecting Any Planted Subgraph: Extended Abstract." Conference on Learning Theory, 2024.

Markdown

[Yu et al. "Counting Stars Is Constant-Degree Optimal for Detecting Any Planted Subgraph: Extended Abstract." Conference on Learning Theory, 2024.](https://mlanthology.org/colt/2024/yu2024colt-counting/)

BibTeX

@inproceedings{yu2024colt-counting,
  title     = {{Counting Stars Is Constant-Degree Optimal for Detecting Any Planted Subgraph: Extended Abstract}},
  author    = {Yu, Xifan and Zadik, Ilias and Zhang, Peiyuan},
  booktitle = {Conference on Learning Theory},
  year      = {2024},
  pages     = {5163-5165},
  volume    = {247},
  url       = {https://mlanthology.org/colt/2024/yu2024colt-counting/}
}