Authors: Alex Sosnkowski, Gabriel Gladstone
Course: CS6501 Spring 2025
ClearAudit is an intuitive web application designed to help data publishers create differentially private datasets and data releases. The tool guides users through the parameter tuning process with interactive data visualizations and attack simulations, making the privacy-utility tradeoff clear and actionable.
- Slow Adoption of Differential Privacy: Many organizations lack understanding of differential privacy (DP), slowing its adoption.
- Growing Need: As machine learning usage increases, so does the need for widespread adoption of DP to protect sensitive data.
- Balancing Interests: It's challenging to find the right balance between the needs of data analysts and the privacy concerns of data publishers.
- Differentially Private Data Visualization: Visual tools help users understand how privacy parameters affect data utility.
- Attack Simulation: Simulate potential privacy attacks to better characterize the privacy-utility tradeoff.
- Synthetic Data Publishing: Generate and release synthetic datasets with configurable privacy parameters (e.g., epsilon tuning) for ML applications.
- Secure Data Releases: Configure context-aware privacy settings for safer data sharing.
- Visualize the impact of privacy parameters on data and machine learning outcomes.
- Simulate attacks to understand privacy risks.
- Compare classic algorithms (e.g., PCA) against their differentially private equivalents.