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 2010.12059stat.MLcs.LG
收藏

归一化流中的有原则插值

Principled Interpolation in Normalizing Flows

University of Campinas(坎皮纳斯大学);Leuphana University(勒芬那大学);Norwegian University of Science and Technology(挪威科技大学)

Samuel G. Fadel, Sebastian Mair, Ricardo da S. Torres, Ulf Brefeld

AI总结:

针对归一化流中线性插值偏离数据流形的问题,本文提出将高斯基分布替换为概率单纯形上的 Dirichlet 分布和超球面上的 von Mises-Fisher 分布,从而在保持生成性能的同时显著提升插值质量。

摘要:

基于归一化流的生成模型在利用较简单分布建模复杂数据分布方面非常成功。然而,直接的线性插值会表现出意想不到的副作用,因为插值路径位于观测到样本的区域之外。这是由标准选择高斯基分布造成的,并且可以从插值样本的范数看出,因为它们位于数据流形之外。这一观察表明,改变插值方式通常应当能带来更好的插值结果,但如何以无歧义的方式做到这一点尚不清楚。在本文中,我们通过强制使用特定的流形并因此改变基分布来解决这一问题,从而实现一种有原则的插值方式。具体而言,我们分别在概率单纯形和超球面上使用 Dirichlet 基分布和 von Mises-Fisher 基分布。我们的实验结果表明,在保持生成性能的同时,插值在每维比特数、Fréchet Inception Distance(FID)和 Kernel Inception Distance(KID)分数方面均表现出优越的性能。

英文摘要:

Generative models based on normalizing flows are very successful in modeling complex data distributions using simpler ones. However, straightforward linear interpolations show unexpected side effects, as interpolation paths lie outside the area where samples are observed. This is caused by the standard choice of Gaussian base distributions and can be seen in the norms of the interpolated samples as they are outside the data manifold. This observation suggests that changing the way of interpolating should generally result in better interpolations, but it is not clear how to do that in an unambiguous way. In this paper, we solve this issue by enforcing a specific manifold and, hence, change the base distribution, to allow for a principled way of interpolation. Specifically, we use the Dirichlet and von Mises-Fisher base distributions on the probability simplex and the hypersphere, respectively. Our experimental results show superior performance in terms of bits per dimension, Fréchet Inception Distance (FID), and Kernel Inception Distance (KID) scores for interpolation, while maintaining the generative performance.

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。

↑