Skip to yearly menu bar Skip to main content


Workshop

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

Yao-Hung Hubert Tsai · Yi Wu · Makoto Yamada · Ruslan Salakhutdinov · Ichiro Takeuchi · Kenji Fukumizu

East Meeting Level 8 + 15 #13

Mon 30 Apr, 11 a.m. PDT

"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 $\textnormal{MMD}_{inc}$ to measure the distribution discrepancy between generated and real images. $\textnormal{MMD}_{inc}$ enjoys the advantages of asymptotic normality, computation efficiency, and model agnosticity. We then propose a GANs analysis framework to select the "best" member in GANs family using the Post Selection Inference (PSI) with $\textnormal{MMD}_{inc}$. In the experiments, we adopt the proposed framework on 7 GANs variants and compare their $\textnormal{MMD}_{inc}$ scores.

Live content is unavailable. Log in and register to view live content