Invited Talk, TU Delft, Delft, Netherlands
This invited talk explored the privacy-interpretability tradeoff in graph machine learning, focusing on how feature explanations can leak graph structure and what this means for trustworthy GNN deployment.
Invited Talk, CISPA, Hannover, Germany
This talk examined privacy risks in explanation methods for graph neural networks and discussed how extracting private graphs through explanations exposes tradeoffs between accuracy, interpretability, and privacy.
Seminar, University of Gottingen, Gottingen, Germany
This seminar focused on the privacy risks introduced by model explanations in graph neural networks, showing how explanatory signals can help an attacker reconstruct private graph structure.
Workshop Talk, Workshop on Privacy Preserving Analysis, Hannover, Germany
This talk connected privacy risks in anonymized healthcare data with privacy risks in trained machine learning models, including reconstruction attacks and membership inference attacks on graph neural networks.
Invited Talk, Tsinghua University Science Park (TusPark), Hong Kong
This talk discussed how NLP methods can improve review helpfulness prediction by extracting sentiment, context, and informative structure from user-generated feedback.