IEEE Visualization VisWeek Panel Friday, October 29, 2010
Visualization Theory: Putting the Pieces Together
Organizers: Caroline Ziemkiewicz and Peter Kinnaird
Panelists: Robert Kosara, Jock Mackinlay, Bernice Rogowitz, Ji Soo Yik.
Theory is an increasingly hot topic in visualization, expanding from its traditional origins in low-level perception and statistics to an ever-broader array of fields and subfields. Modern visualization theory includes color theory, visual cognition, visual grammars, interaction theory, visual analytics, information theory, and a growing but so far vaguely defined area of theory specific to visualization itself. In this panel, we bring together researchers who are studying visualization theory from these numerous different perspectives and ask how these disparate topics can combine and comment on one another to create a more unified body of theory and answer pressing research questions.
Bernice Rogowitz's Position Statement and Slides
A theory of visualization hinges on the question of semantics. In visualization, we map data onto visual elements in a way that we hope will help the user to perceive and reason about the structure in the data. We also develop interactive methodologies that we hope will allow the user to explore the data in a way that will help reveal structures that were previously hidden.
To preserve the semantic structure of the data, we need to have a theory that will provide a mapping from data structures to perceptual structures. For exampleexperiments have shown that magnitude information is well represented using color maps that have a monotonic luminance component. When equal steps in the data correspond to equal perceptual steps in the visualization, data magnitude information is directly available to the user, which can be very useful in revealing structure in the data. To pick another case, we have all seen examples of glyphs or textures that obfuscate the semantic meaning in the data. A theory of visualization needs to develop rules for characterizing which visual elements produce successful representations of different structures in the data.
In interactive visualization, the representation of the data is manipulated and transformed to reveal deeper structures in the underlying data. For example, the user may want to use color a semantic marker, to highlight certain ranges in the data, or may want to peel away unselected regions in order to identify hidden structures. Or, the analyst may want to transform the data mathematically, or represent structure that have been revealed through algorithmic manipulations. The goal of a theory of visualization would be to predict how our perception of the meaning in the data depends on these various transformations, and how to match these different representations to our analysis goals and tasks. This work will require us to develop a deeper understanding of how higher-level human capabilities, such as attention, decision-making, and aesthetics contribute to our ability to wrest meaning from our data.
In addition to developing a theory of visualization, we need methodologies that will integrate its rules into visualization software. In the same way a statistics or data mining program guide the user in selecting an appropriate test or ensuring that the data fit the algorithm’s assumptions, we need to provide training wheels for visualization users. To do so will require extracting and providing guidance based on metadata about the data, visual task, and the perceptual and cognitive capabilities of the user who will be interpreting it.