Obscuring and Analyzing Sensitive Information With Generative Adversarial Networks

Abstract

Much of the data collected by corporations and public institutions is too sensitive to share publicly or with a third party. Strict rules govern who may access medical records, financial information and other confidential data. However, there is a great potential for this data to be analyzed in new ways, if only it could be shared with the right researchers or business analysts. Generative adversarial networks (GANs) are an advance in artificial intelligence which may provide a solution to this problem. A well-trained GAN will create new data that is representative of the original data. This output could be analyzed by a third party while obscuring any sensitive or confidential information from the original data. In this paper, we assess the potential of using GANs to generate representative data and build insightful models without the original data.