SPECTRAL NORMALIZATION FOR GENERATIVE ADVERSARIAL NETWORKS阅读记录

1，A number of works (Ue-hara et al., 2016; Qi, 2017; Gulrajani et al., 2017) advocate the importance of Lipschitz continuity in assuring the boundedness of statistics

1.1  Ue-hara et al., 2016;Generative Adversarial Nets from a Density Ratio Estimation Perspective

1.1.1 density ratio estimation 

 Divergence estimation and two-sample homogeneity test under semiparametric density-ratio models

1.1.1.1  semiparametric density ratio estimator；

 semiparametric density ratio estimator

1.1.1.1.1  K b (?) ¼ b ?1 K(?=b)

Density Estimation for Statistics and Data Analysis

 

2，W-GAN

2.1  It is then unlikely that the model manifold and the true distribution’s support have a non-negligible intersection

T OWARDS P RINCIPLED M ETHODS FOR T RAINING GENERATIVE A DVERSARIAL N ETWORKS

2.1.1 One possible cause for the distributions not to be continuous is if their supports lie on low dimen-
sional manifolds. There is strong empirical and theoretical evidence to believe that P r is indeed
extremely concentrated on a low dimensional manifold

Optimal Transport: Old and New