We present a data-driven framework called generative adversarial privacy (GAP). Inspired by recent advancements in generative adversarial networks (GANs), GAP allows the data holder to learn the privatization mechanism directly from the data. Under GAP, finding the optimal privacy mechanism is formulated as a constrained minimax game between a privatizer and an adversary. We show that for appropriately chosen adversarial loss functions, GAP provides privacy guarantees against strong information-theoretic adversaries. We also evaluate GAP's performance on the GENKI face database.
@article{arxiv.1807.05306,
title = {Generative Adversarial Privacy},
author = {Chong Huang and Peter Kairouz and Xiao Chen and Lalitha Sankar and Ram Rajagopal},
journal= {arXiv preprint arXiv:1807.05306},
year = {2019}
}
Comments
Talk presentation at Privacy in Machine Learning and Artificial Intelligence (PiMLAI) Workshop, ICML 2018