Class Conditional Time Series Generation with Structured Noise Space GAN

Class-conditional generative models represent significant progress in machine learning, empowering tasks that necessitate controlled data synthesis. Contrary to their unconditional counterparts, these models synthesize samples conditioned on specific class labels, facilitating a more focused generative process. This conditional generation enables the production of class-representative data, which has a multitude of practical applications. For example, such models can augment the diversity of a class in a training dataset, thereby enhancing the robustness of classifiers in subsequent tasks. Moreover, Generative Adversarial Networks (GANs) [1] stand to gain from integrating class labels during the training phase.