Generative Adversarial Networks (GAN) is designed as an unsupervised learning model that refers to the process of automatically identifying and learning the regularities or patterns in input data, and then the model generates new examples from the original dataset. The GAN model comprises of two sub-models includes a generator model for generating new examples and a discriminator model for classifying whether generated examples are real from the domain or not.
Classic GAN models experience problems such as both generator and discriminator are not optimal at the same time and inability of the generator to control the explication of generated samples. Other issues of GAN are limited accuracy in classification, condition generation, and predictions. The emergence of Triple Generative Adversarial Networks (Triple-GAN) is to address such problems. Triple-GAN comprises three components, namely a generator, a discriminator, and a classifier. The generator and the classifier distinguish the conditional distributions to perform conditional generation and classification. The discriminator focuses on recognizing fake pairs. Triple-GAN accomplishes excellent classification results and produces meaningful samples. Triple-GAN is flexible to integrate different GAN architectures and semi-supervised classifiers.