Counterbalance: A Declarative Tool for Privacy-Preserving Statistical Chart Authoring based on Differential Privacy

Authors

Xumeng Wang (Nankai University), Shuangcheng Jiao (Nankai University), Chris Bryan (Arizona State University), Yining Wang (Nankai University)

Presentation

Session: Story time
Time: Thursday, Nov 12, 08:24 – 08:36 (US/Eastern) · session 08:00 – 09:30
Location: Hall America north

Keywords

Chart authoring, privacy preservation, declarative tool.

Abstract

Visualizations are widely used to communicate insights about data, and many authoring tools now exist to simplify the chart creation process. Unfortunately, when datasets contain sensitive information, such charts and tools can risk privacy exposure. For many dataset owners, it is a significant challenge to balance (i) effectively representing desired insights and (ii) ensuring privacy when visualizing their data. We investigate this by studying how to integrate privacy mechanisms into the chart authoring process. Based on a pre-study with domain experts, we design a compact-yet-inclusive declarative grammar for specifying differentially private visualizations. Using this, we develop Counterbalance, an authoring tool that supports creating privacy-preserving charts in a declarative manner. Notably, Counterbalance employs a novel scheme recommendation model to identify the implementation details of differential privacy approaches that successfully achieve user-defined goals such as preserving sensitive information while still effectively communicating desired insights. The tool also supports post-generation feedback on goal achievement as a way to validate goal achievement and mitigate potential misuse (such as inadvertent deductive disclosure) based on ambiguous declarations. Based on our process of designing, implementing, and empirically evaluating Counterbalance, we discuss how declarative approaches can support data owner accessibility, provide security guarantees, and promote human-in-the-loop privacy-preserving visualization.

For Practitioners

Data scientists.