arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2103.10060cs.LGstat.ML
收藏

使用 Wasserstein 生成对抗网络逼近概率分布

Approximating Probability Distributions by using Wasserstein Generative Adversarial Networks

The University of Hong Kong(香港大学)

Yihang Gao, Michael K. Ng, Mingjie Zhou

AI总结:

本文研究了以GroupSort神经网络为判别器的WGANs,建立了生成分布与目标分布间Wasserstein距离的量化泛化界限,揭示了判别器容量不足时高容量生成器反而表现更差的现象,并通过Swiss roll和MNIST数据集验证了理论。

摘要:

本文研究了以 GroupSort 神经网络作为判别器的 Wasserstein 生成对抗网络(WGANs)。结果表明,目标分布逼近的误差界限取决于生成器和判别器的宽度与深度(即容量)以及训练中的样本数量。本文为生成分布与目标分布之间的 Wasserstein 距离建立了一个量化的泛化界限。根据理论结果,WGANs 对判别器容量的要求高于对生成器容量的要求,这与一些现有结果一致。更重要的是,如果判别器不够强大,使用过深和过宽(高容量)生成器的结果可能比使用低容量生成器的结果更差。使用 Swiss roll 和 MNIST 数据集获得的数值结果证实了这些理论结果。

英文摘要:

Studied here are Wasserstein generative adversarial networks (WGANs) with GroupSort neural networks as their discriminators. It is shown that the error bound of the approximation for the target distribution depends on the width and depth (capacity) of the generators and discriminators and the number of samples in training. A quantified generalization bound is established for the Wasserstein distance between the generated and target distributions. According to the theoretical results, WGANs have a higher requirement for the capacity of discriminators than that of generators, which is consistent with some existing results. More importantly, the results with overly deep and wide (high-capacity) generators may be worse than those with low-capacity generators if discriminators are insufficiently strong. Numerical results obtained using Swiss roll and MNIST datasets confirm the theoretical results.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑