Bayesian Analysis in Visual Analytics (BAVA)

The goal of this research is to combine two areas, Visual Analytics and Bayesian Statistics. Currently, visualizations display inflexible deterministic transformations of data that inherently separate data visualization from visual synthesis. Analysts cannot manipulate displays to inject domain-specific knowledge to formally assess the merger of their expert judgment with the data. However, by changing the nature of the data transformation from deterministic to probabilistic Bayesian methods, manipulations to a display are possible to interpret quantitatively. Thus, a new visualization model is developed which offers editable representations to promote bidirectional flow between analyst and data.

This approach enables analysts to quantify data uncertainty, formally include expert judgment into analyses, rapidly generate and test new hypotheses, and allows multi-source and multi-scale data to contribute to one data display. The developed software is applied and evaluated with analysts in multiple fields, including intelligence analysis.

While the research immediately impacts how analysts discover new information in large datasets, the methodology also transitions smoothly into the classroom. Bayesian courses are currently limited to advanced graduate students, but learning Bayesian statistics at a more intuitive level can overcome this barrier. The combination of Bayesian methodology and visual analytics enables students to assess the impact of various levels of prior information on probability models through visual representations and the sense-making feedback loop. Students with varying backgrounds, interests, and talents, including undergraduates and underrepresented groups with limited academic history, will now have the opportunity to learn 21st century statistics. The project Web site (http://infovis.cs.vt.edu/BAVA) will be used for results dissemination.