Bayesian Nonparametric Learning using the Maximum Mean Discrepancy Measure for Synthetic Data Generation

Forough Fazeli-Asl, Michael Minyi Zhang, Lizhen Lin

Keywords: Bayesian nonparametric learning, Dirichlet process, GAN

TL;DR: 4-pages

Abstract: We introduce a Bayesian estimator for maximum mean discrepancy (MMD), enabling a novel approach to measure-based data generation. To demonstrate the adaptability of our method, we embed this estimator within a generative adversarial network (GAN) framework. This integration offers a powerful avenue for Bayesian nonparametric (BNP) learning, showcasing the estimator's broad applicability. Our BNP-driven GAN not only enhances sample diversity but also improves inferential accuracy, surpassing the performance of traditional methods. Further theoretical properties, proofs, and experiments are given by the Appendix.

Submission Number: 62