DSF-GAN: DownStream Feedback Generative Adversarial Network

arXiv:2403.18267v1 Announce Type: new
Abstract: Utility and privacy are two crucial measurements of the quality of synthetic tabular data. While significant advancements have been made in privacy measures, generating synthetic samples with high utility remains challenging. To enhance the utility of synthetic samples, we propose a novel architecture called the DownStream Feedback Generative Adversarial Network (DSF-GAN). This approach incorporates feedback from a downstream prediction model during training to augment the generator's loss function with valuable information. Thus, DSF-GAN utilizes a …

abstract adversarial architecture arxiv cs.ai cs.lg data feedback gan generative generative adversarial network network novel privacy quality samples synthetic tabular tabular data type utility