This article is the first in a series of wrap-ups from VizWorld reader T.J. Jankun-Kelly (@dr_tj) of the EuroVis2010 conference currently underway.
EuroVis 2010 is a three day conference supported by EuroGraphics and IEEE; it was the first of VisWeek’s “sister conferences.” The first VisSym was held in 1999 and became EuroVis in 2005. This year, it is held outside of Bordeaux, France by LaBri. Herein are a few brief glimpses at this year’s conference.
Keynote: Matt Ward Challenges, Partial Solutions, and Open Problems in Multivariate Data Visualization
Matt Ward is a Professor at Worcester Polytechnic Institute who has focused on multivariate visualization for the majority of his career. As one of the chief instigators of XmdvTool, a well established multivariate visualization tool, he has contributed significantly to information and scientific visualization. In his keynote, he shared his perspective on the field of multivariate visualization and how it has changed over the last 20 years.
Read the rest of TJ’s Review after the break.
Matt described his research in multidimensional visualization as occurring over three phases. First, he discussed mapping. In the beginning, multidimensional visualization focused on how to show what was going on—the mapping aspect of the problem. Matt did a quick survey of many different visual dimensions (based upon Bertin) and how they have been adapted to multivariate visualization via XmdvTool and others. But even in mapping, the best studied part of multidimensional visualization, several challenges remain. Chief amongst them, he feels, is making multivariate visualization ubiquitous—to move beyond just visual communication of 1 & 2 dimensions. He also called out how motion as a mapping tool (not an interaction tool) has yet to be largely exploited.
Moving past mapping, Matt discussed interaction—the source of power in visualization. He then provided several examples of how interaction has strongly benefited multidimensional visualization. The hard part of interaction is making it intuitive, visible, and direct. The chief challenges in interaction is moving beyond the desktop and the dashboard—novel devices are everywhere—and to find interactive paradigms beyond focus+context. It is a fine tool, but there are likely others to be discovered.
The final phase of Matt’s multivariate visualization research was Analytics. Matt opined that there should be an seamless integration of computation and visualization and that we are only beginning to utilize this potential. Significant more work needs to be done in this area: Its all a challenge.
Matt closed with some final thoughts on the need for a science of visual communication and analysis. We need task & pattern taxonomies. We need to measure information content: How it is lost (akin to some work of his in information loss in the visualization pipeline) and how it is “gained” (to use a term from Chris Weaver, i.e., when visualization provides misinformation or bias). A measure for the worth of visualization, and methods to measure its success is needed to continue this work into the future.
Overall, an interesting overview of the field of multivariate visualization and some calls to future work.
Session 1: Volume Visualization
Isosurface Similarity Maps; Stefan Bruckner, Torsten Möller
Bruckner and Möller, from Simon Frasier University in Canada, introduced a novel method or exploring isosurface space utilizing mutual information between distance transforms of isosurfaces in a volume. The insight of their work is that the mutual information here provides details of how the shapes of the isosurfaces differ; information that a histogram measure alone cannot always provide. They provides a mathematical foundation and discussed how the method can be used to both improve user interaction with isosurfaces and how the technique can be used for automatic “best” finding of isosurfaces.
An Exploratory Technique for Coherent Visualization of Time-varying Volume Data; Anna Tikhonova, Carlos D. Correa, Kwan-Liu Ma
Tikhonova, from Ma’s ViDi group at UC Davis, discussed the issue of trying to find appropriate transfer functions for data that is streaming. They propose a method utilizing ray attenuation calculations along each ray (pixel) in a volume rendering; thus, as more information (timestemps, larger data ranges, etc) becomes available, the transfer function can be on the fly adjusted without the use of the original data. Three classes of transfer function changes and methods for dealing with each were discussed leading to a method that has almost no overhead in many cases and small overhead in for complex changes with high image fidelity.
Session 2: Multivariate Data
Evaluation of Cluster Identification Performance for Different PCP Variants; Danny Holten, Jarke J. van Wijk
van Wijk, from Eindhoven, discussed an empirical evaluation of several different visual adaptations for parallel coordinates in order to improve their ability to visually communication clusters with the data. A combination of colors, animation, density mapping, and annotation via scatterplots were explored. While the hypothesis was that color would communicate the data best, scatterplot augmentation proved the only significantly different (and better) method with the others being equal aside from animation which always hurt. An interesting “negative” result, and suggests future work in the area of non-synthetic data and data of higher dimensionality to see if the result continues in such settings.
Supporting Exploratory Analysis with the Select & Slice Table; Yedendra Babu Shrinivasan, Jarke van Wijk
Another paper by van Wijk’s group, Shrinivasan explores the use of selection as a first class entity in visualizations. Using an interactive table metaphor, selections in data are labeled, cached, and allowed to be modified, queried, or adapted to other visualization to dynamically understand trends within multivariate visualization. Examples from socioeconomics dominated; on included looking at the factors of poverty in India via different classifications of income, housing, and so on.
Brushing Moments in Interactive Visual Analysis; Johannes Kehrer, Peter Filzmoser, Helwig Hauser
Kehrer, from Hauser’s group in Bergen, uses different estimators of statistical moments (mean, median, variance, skew, and kurtosis) in order to explore and classify data in multivariate scatterplots. Both the “traditional” and “robust” (resistant to outlier) versions of these moments were utilized, and brushing between these traditional and robust views were used to understand the shape of the distributions and to find outliers.
Visualizing Summary Statistics and Uncertainty; Kristin Potter, Joe Kniss, Richard Riesenfeld, Chris R. Johnson
Potter presented this work from SCI in Utah examining ways to extend the boxplot to include additional statistical features with the goal of reading all of these values at once. The “simularity plots” encode the measures of central tendency along with statistical moments, distribution, and model fit information. A 2D method was also introduced. The work has some interesting ideas and would benefit from a user study to validate the effectiveness of the different goals of the depiction.
Posters
The final session of the afternoon and early evening was the poster session. Seventeen posters were presented in the program. Highlight included some initial work of using cognitive workload as a metric in visualization effectiveness, GPU accelerated Voronoi treemaps with dynamic updates, and two community based efforts for visualization: One in a social context, another focused on web-based use.
