Generative Adversarial Network for Generating Different Types of Data

Abstract

A generative adversarial network (GAN) is one of the popular deep generative models. It generates new data similar to the data of a dataset but is not intended to generate different data from them. In this paper, we propose a GAN that generates such different types of data, which a user desires to obtain. In the proposed method, some data of the dataset are iteratively exchanged for ones generated by the generator if the generated data are more helpful in generating the user's desirable ones. The performance of the proposed method is evaluated by comparing it with some other GANs.