Graphical Generative Adversarial Networks

摘要: We propose Graphical Generative Adversarial Networks (Graphical-GAN) to model structured data. Graphical-GAN conjoins the power of Bayesian networks on compactly representing dependency structures among random variables and that generative adversarial learning expressive functions. introduce a recognition infer posterior distribution latent given observations. generalize Expectation Propagation (EP) algorithm learn jointly. Finally, we present two important instances Graphical-GAN, i.e. Gaussian Mixture GAN (GMGAN) State Space (SSGAN), which can successfully discrete temporal visual datasets, respectively.