Theory and Applications of Human Interaction in Visual Analytics

• Activity Number: 678
• Type: Topic Contributed
• Date/Time: Thursday, August 8, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Scotland Charles Leman*+
• Companies: Virginia Tech

Abstract

We develop an interactive visual data exploration framework called Bayesian Visual Analytics (BaVA). BaVA combines methods in Visual Analytics (VA) with Bayesian statistics so that experts may guide model based, quantitative data analyses via their interactions with data visualizations. For example, experts may adjust observations in BaVA-driven displays of data to convey that the statistical model (which created the display) must account for similarities and/or differences in these observations. The novelty is that we consider display adjustments to represent cognitive feedback from experts that is reliable information concerning the underlying analytical approach. Such feedback will enable user-guided data explorations. This talk will focus on presenting the BaVA paradigm and several methods we use transform cognitive feedback into parametric feedback. Additionally, we provide illuminating examples of BaVA-enhanced data explorations based on both simulated and real datasets. Additionally, we will discuss the integration of interactive visualization into core statistics undergraduate and graduate programs, and how such topics complement and motivate formal statistical thinking.