An Exploration for Generative Adversarial Networks Via Adding A Screening Model

Abstract

As an emerging deep learning model, generative adversarial networks has enough creativity and potential in the application and advanced studies. However, many problems should be tackled in its training process. Based on the exist studies of generative adversarial networks and relevent ideas of games theories, this article adds a new screening model for the framework of generative adversarial networks to solve vanishing gradient in the training process, which contains the function of filter and measurer. This model does not interfere training process, only find and delete discriminators which might cause vanishing gradient, and output a value to represent the progress of training process.