Description
Synthetic data allow us to explore situations beyond what available observations capture or what data privacy policies allow. Yet selected statistical properties may be compatible with many different datasets, of which we normally pick the most "plausible." But synthetic datasets can satisfy the same conditions while exhibiting different patterns and behaviors, thus supporting different analytical conclusions. We call this analytical underdetermination: the information available leaves room for multiple alternatives whose differences matter for analysis.
This topic investigates how Visual Analytics can make data synthesis an interactive, exploratory process. You will design and implement a visual application that connects a generation method with tools for visualizing alternative generated datasets. Users should be able to specify properties to preserve or vary, generate and compare alternatives, and explore how their distributions, relationships, and resulting insights differ.
The aim is to help users explore what is the space of the possible under stated conditions and distinguish findings that persist across generated alternatives from those that depend on particular generation choices in an interactive loop. The topic provides an entry into generative Visual Analytics through a self-contained prototype and concrete experiments.
Tasks
- Review relevant approaches to synthetic data generation and visual exploration, and select a bounded generation scenario.
- Design visualizations and interactions for configuring generation, exploring outputs, and comparing alternatives.
- Implement a prototype using an existing generation method, keeping its settings and the origin of displayed data inspectable.
- Evaluate the approach through controlled examples and user feedback, examining whether users can identify patterns that depend on generation choices.
Requirements
- Good programming skills and an interest in interactive visualization.
- Basic knowledge of statistics and data analysis.
- Experience with visualization libraries or generative models is advantageous.
Environment
The project can be implemented as a standalone web or desktop application using suitable visualization tools and existing generation libraries. Controlled examples or openly available datasets will support development and evaluation. A bachelor thesis will focus on one generation method and a compact exploration workflow; a master thesis can investigate alternative interaction designs through a comparative evaluation.
References
- Pérez-Messina, I., Miksch, S., and Tominski, C. (2026). From Data to Ficta: A Critical Reflection on Visual Analytics in the Age of Generative Models. EuroVis Workshop on Visual Analytics.
- Offenhuber, D. (2024). Shapes and frictions of synthetic data. Big Data & Society, 11(2).
Responsible
For more information please contact Ignacio Pérez-Messina.