Title:Sample-based high-dimensional convexity testing

Abstract:In the problem of high-dimensional convexity testing, there is an unknown set $S \subseteq \mathbb{R}^n$ which is promised to be either convex or $\varepsilon$-far from every convex body with respect to the standard multivariate normal distribution $\mathcal{N}(0, 1)^n$. The job of a testing algorithm is then to distinguish between these two cases while making as few inspections of the set $S$ as possible.
In this work we consider sample-based testing algorithms, in which the testing algorithm only has access to labeled samples $(\boldsymbol{x},S(\boldsymbol{x}))$ where each $\boldsymbol{x}$ is independently drawn from $\mathcal{N}(0, 1)^n$. We give nearly matching sample complexity upper and lower bounds for both one-sided and two-sided convexity testing algorithms in this framework. For constant $\varepsilon$, our results show that the sample complexity of one-sided convexity testing is $2^{\tilde{\Theta}(n)}$ samples, while for two-sided convexity testing it is $2^{\tilde{\Theta}(\sqrt{n})}$.