English

On Scalable Testing of Samplers

Data Structures and Algorithms 2023-06-27 v1 Probability

Abstract

In this paper we study the problem of testing of constrained samplers over high-dimensional distributions with (ε,η,δ)(\varepsilon,\eta,\delta) guarantees. Samplers are increasingly used in a wide range of safety-critical ML applications, and hence the testing problem has gained importance. For nn-dimensional distributions, the existing state-of-the-art algorithm, Barbarik2\mathsf{Barbarik2}, has a worst case query complexity of exponential in nn and hence is not ideal for use in practice. Our primary contribution is an exponentially faster algorithm that has a query complexity linear in nn and hence can easily scale to larger instances. We demonstrate our claim by implementing our algorithm and then comparing it against Barbarik2\mathsf{Barbarik2}. Our experiments on the samplers wUnigen3\mathsf{wUnigen3} and wSTS\mathsf{wSTS}, find that Barbarik3\mathsf{Barbarik3} requires 10×10\times fewer samples for wUnigen3\mathsf{wUnigen3} and 450×450\times fewer samples for wSTS\mathsf{wSTS} as compared to Barbarik2\mathsf{Barbarik2}.

Keywords

Cite

@article{arxiv.2306.13958,
  title  = {On Scalable Testing of Samplers},
  author = {Yash Pote and Kuldeep S. Meel},
  journal= {arXiv preprint arXiv:2306.13958},
  year   = {2023}
}

Comments

Appeared at NeurIPS 2022

R2 v1 2026-06-28T11:13:28.079Z