Membership Inference Attacks for Machine Learning Models

Description

Conducted research on privacy vulnerabilities in machine learning models by implementing membership inference attacks to evaluate reliability and disparities in privacy.
This work has led to a paper, Membership Inference Attacks as Privacy Tools: Reliability, Disparity, and Ensemble, which has been submitted to ACM CCS 2025 and is currently under review.
This project was ongoing for more than a year. I contributed to it from January 2024 to January 2025, focusing on building data pipeline and conducting comprehensive evaluations.

Built data pipelines for processing attack predictions across multiple seeds.
Created visualizations, including Venn diagrams and upset diagrams, to identify patterns and disparities in attack outcomes.
Automated data processing workflows with Shell Scripts to handle large datasets.
Calculated metrics such as Jaccard similarity, set variance, and entropy for reliability analysis.

Professor Lei Yu