English

Selecting the Best in GANs Family: a Post Selection Inference Framework

Machine Learning 2018-06-26 v2 Machine Learning

Abstract

"Which Generative Adversarial Networks (GANs) generates the most plausible images?" has been a frequently asked question among researchers. To address this problem, we first propose an \emph{incomplete} U-statistics estimate of maximum mean discrepancy MMDinc\mathrm{MMD}_{inc} to measure the distribution discrepancy between generated and real images. MMDinc\mathrm{MMD}_{inc} enjoys the advantages of asymptotic normality, computation efficiency, and model agnosticity. We then propose a GANs analysis framework to select and test the "best" member in GANs family using the Post Selection Inference (PSI) with MMDinc\mathrm{MMD}_{inc}. In the experiments, we adopt the proposed framework on 7 GANs variants and compare their MMDinc\mathrm{MMD}_{inc} scores.

Keywords

Cite

@article{arxiv.1802.05411,
  title  = {Selecting the Best in GANs Family: a Post Selection Inference Framework},
  author = {Yao-Hung Hubert Tsai and Makoto Yamada and Denny Wu and Ruslan Salakhutdinov and Ichiro Takeuchi and Kenji Fukumizu},
  journal= {arXiv preprint arXiv:1802.05411},
  year   = {2018}
}
R2 v1 2026-06-23T00:23:07.080Z