BriefGPT.xyz
Jun, 2020
通过流形拓扑评估深度生成模型的解缠结性
Evaluating the Disentanglement of Deep Generative Models through Manifold Topology
HTML
PDF
Sharon Zhou, Eric Zelikman, Fred Lu, Andrew Y. Ng, Stefano Ermon
TL;DR
本文提出了一种利用生成模型度量解缕编表示的方法。通过量化学习表示中条件子流形的拓扑相似性，该方法不依赖于外部模型或特定数据集即可测量解缕编程度。我们通过多个数据集上的实验表明了该方法的有效性和适用性，发现与现有方法相比，该方法能够很好地对模型进行排序。
Abstract
Learning
disentangled representations
is regarded as a fundamental task for improving the generalization, robustness, and interpretability of
generative models
. However, measuring disentanglement has been challen
→