Statistics > Computation
[Submitted on 15 Aug 2022 (this version), latest version 22 Jul 2023 (v3)]
Title:Nesterov smoothing for sampling without smoothness
View PDFAbstract:We study the problem of sampling from a target distribution in $\mathbb{R}^d$ whose potential is not smooth. Compared with the sampling problem with smooth potentials, this problem is more challenging and much less well-understood due to the lack of smoothness. In this paper, we propose a novel sampling algorithm for a class of non-smooth potentials by first approximating them by smooth potentials using a technique that is akin to Nesterov smoothing. We then utilize sampling algorithms on the smooth potentials to generate approximate samples from the original non-smooth potentials. With a properly chosen smoothing intensity, the accuracy of the algorithm is guaranteed. For strongly log-concave distributions, our algorithm can achieve $\mathcal{E}$ error in Wasserstein-2 distance with complexity $ \widetilde{\mathcal{O}} \left(
\frac{ d^{1/3}}{ \mathcal{E}^{5/3}}
\right) .$ For log-concave distributions, we achieve $\mathcal{E}$ error in total variation with complexity $\mathcal{O} \left(\frac{ M_\pi d }{ \mathcal{E}^{3}} \right) $ in expectation with $M_\pi$ being the second moment of the target distribution. For target distributions satisfying the logarithmic-Sobolev inequality, our algorithm has complexity $\widetilde{\mathcal{O}} \left( \frac{ d }{\mathcal{E}}\right)$.
Submission history
From: Jiaojiao Fan [view email][v1] Mon, 15 Aug 2022 22:31:55 UTC (432 KB)
[v2] Thu, 16 Mar 2023 04:28:42 UTC (605 KB)
[v3] Sat, 22 Jul 2023 21:26:10 UTC (599 KB)
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.