DGSAN: Discrete Generative Self-Adversarial Network

• DOI: 10.48550/arxiv.1908.09127
• Publication Date: 2019-01-01
• Authors: Ehsan Montahaei; Danial Alihosseini; Mahdieh Soleymani...
• Keywords: Discriminator; Generative adversarial network

Although GAN-based methods have received many achievements in the last few years, they not been entirelysuccessful generating discrete data. The most crucial challenge of these is difficulty passing gradientfrom discriminator to generator when outputs are discrete. Despite fact that several attemptshave made alleviate this problem, none existing improved performance oftext generation compared with maximum likelihood approach terms both quality and diversity. In thispaper, we proposed a new framework for data by an adversarial which there no need topass gradient generator. method has iterative manner each definedbased on discriminator. It leverages discreteness model real datadistribution implicitly. Moreover, supported theoretical guarantees, experimental results generallyshow superiority DGSAN other popular or recent generatingdiscrete sequential