Private graph extraction via feature explanations

Published in Proceedings on Privacy Enhancing Technologies (PETS), 2023

This work studies the tension between interpretability and privacy in graph machine learning. We develop reconstruction attacks that recover private graph structure from model explanations and compare privacy leakage across gradient-based, perturbation-based, and surrogate explanation methods.

Recommended citation: Iyiola E. Olatunji, Mandeep Rathee, Thorben Funke, and Megha Khosla. (2023). "Private graph extraction via feature explanations." Proceedings on Privacy Enhancing Technologies (PETS).
Download Paper