[2102.00199] Charges of convergence for density estimation with generative adversarial networks

0
33


Obtain a PDF of the paper titled Charges of convergence for density estimation with generative adversarial networks, by Nikita Puchkin and 4 different authors

Obtain PDF

Summary:On this work we undertake an intensive examine of the non-asymptotic properties of the vanilla generative adversarial networks (GANs). We show an oracle inequality for the Jensen-Shannon (JS) divergence between the underlying density $mathsf{p}^*$ and the GAN estimate with a considerably higher statistical error time period in comparison with the beforehand recognized outcomes. The benefit of our sure turns into clear in utility to nonparametric density estimation. We present that the JS-divergence between the GAN estimate and $mathsf{p}^*$ decays as quick as $(log{n}/n)^{2beta/(2beta + d)}$, the place $n$ is the pattern measurement and $beta$ determines the smoothness of $mathsf{p}^*$. This charge of convergence coincides (as much as logarithmic elements) with minimax optimum for the thought-about class of densities.

Submission historical past

From: Nikita Puchkin [view email]
[v1]
Sat, 30 Jan 2021 09:59:14 UTC (27 KB)
[v2]
Solar, 7 Nov 2021 16:22:03 UTC (82 KB)
[v3]
Thu, 19 Jan 2023 08:29:23 UTC (38 KB)
[v4]
Thu, 25 Jan 2024 10:04:05 UTC (41 KB)



Supply hyperlink

LEAVE A REPLY

Please enter your comment!
Please enter your name here