We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

math

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Mathematics > Statistics Theory

Title: Rates of convergence for density estimation with GANs

Abstract: In this work we undertake a thorough study of the non-asymptotic properties of the vanilla generative adversarial networks (GANs). We derive theoretical guarantees for the density estimation with GANs under a proper choice of the deep neural networks classes representing generators and discriminators. In particular, we prove that the resulting estimate converges to the true density $\mathsf{p}^*$ in terms of Jensen-Shannon (JS) divergence at the rate $(\log{n}/n)^{2\beta/(2\beta+d)}$ where $n$ is the sample size and $\beta$ determines the smoothness of $\mathsf{p}^*$. To the best of our knowledge, this is the first result in the literature on density estimation using vanilla GANs with JS convergence rates faster than $n^{-1/2}$ in the regime $\beta > d/2$. Moreover, we show that the obtained rate is minimax optimal (up to logarithmic factors) for the considered class of densities.
Comments: 57 pages
Subjects: Statistics Theory (math.ST); Machine Learning (stat.ML)
Cite as: arXiv:2102.00199 [math.ST]
  (or arXiv:2102.00199v2 [math.ST] for this version)

Submission history

From: Nikita Puchkin [view email]
[v1] Sat, 30 Jan 2021 09:59:14 GMT (27kb)
[v2] Sun, 7 Nov 2021 16:22:03 GMT (82kb)

Link back to: arXiv, form interface, contact.