Workshop Goals: Good visual displays of data and of complex concepts are known to facilitate
understanding, thinking, and inferences about the data or concepts. The process of using flexible, perhaps
interactive, visualization tools has the potential to multiply the power of visual data display. Creating
effective visual representations is challenging, all the more when the information is vast, uncertain, or
poorly understood. A major goal of this workshop is to convene experts from various communities to
share perspectives on these challenges. An additional goal, prompted by the dramatic increase in the size
of data sets in many fields over recent years, is to spark increased research into innovative approaches for
visualizing uncertainty for large data sets.
The workshop is structured around three examples of contexts within which people need to evaluate
information with uncertainties. These three contexts are characterized by differing amounts of data;
differing structure, mode, certainty, and relevance of data; differing time pressures; and differing goals,
expertise, and motivation of users. They are not meant to be a hard-and-fast taxonomy of the problem
space, but rather a way to sample the range of problems that would benefit from better means of
visualizing uncertain information.
Thursday, March 3
8:00 am Breakfast in meeting room
8:30 Welcome; goals of workshop (Leland Wilkinson, SPSS, Inc.)
Overview Presentations
8:40 Best practices for visualizing uncertain information in computer science
Ben Shneiderman, University of Maryland, and Alex Pang, University of California at
Santa Cruz
9:30 Best practices for visualizing uncertain information in geography
Michael Goodchild, University of California at Santa Barbara
10:20 Break
10:35 Best practices for visualizing uncertain information in statistics
Andreas Buja, University of Pennsylvania
11:25 Best practices for visualizing uncertain information in psychology
Barbara Tversky, Stanford University
12:15 Lunch
1:30 Context 1: Large, structured data sets with ample time for interpretation. Examples include
science, engineering, and medical research. The challenge for decision-makers is
to comprehend and interpret complex visualizations.
1:30-1:45 Discussant (Alex Szalay, Johns Hopkins University) explains the context and gives an
example of real data
1:45-2:15 David Scott, Rice University
2:15-2:45 Bill Cleveland, Purdue University
2:45-3:15 Michael Friendly, York University
3:15-3:45 Discussant presents reactions to ideas given by the speakers.
3:45 Break
4:15 Plenary discussion
5:30 Reception
6:30 Dinner
Friday, March 4
8:00 am Breakfast in meeting room
8:30 Context 2: Large, unstructured, changing data sets where the relevance, significance, and
conceptual links among the data have yet to be discovered. Examples include
intelligence, homeland security, military and business planning, market trading, waging
war. The data are not well structured and may be very varied (e.g., text, images, reports
of varying credibility), uncertainties might be difficult to quantify, and the data may be
updated during while analysis is underway. The signal-to-noise ratio is low in that the
important information is only a small fraction of the data, and there is no good conceptual
model to guide the analysis, so the analysis is exploratory or driven by ad hoc, uncertain
hypotheses. Users are motivated to learn new visualization tools, but and the breadth and
variety of data make them dependent on the way the data are presented because they
cannot invest time into understanding all the uncertainties in the input data. Analyses
might have a timescale of hours to days.
8:30-8:45 Discussant (David Harris, NSA) explains the context and gives an example of real data
8:45-9:15 Ronald Coifman, Yale University
9:15-9:45 Andre Skupin, University of New Orleans
9:45-10:15 Stephen Eick, University of Illinois at Chicago and SSS Research
10:15-10:45 Discussant presents reactions to ideas given by the speakers.
10:45 Break
11:00 Plenary discussion
11:45 Lunch
12:45 Context 3: Data sets that are reduced for rapid understanding in time-pressured
situations. Examples include pilots, executive decisions in business and policy,
newspaper pollsand economic data. Decision-makers have limited time to study the
information, because of lack of time, motivation, or expertise. The visual display must
summarize a complex landscape of phenomena for quick interpretation. The raw data
may be precise (e.g., measures of the state of an aircraft) or imprecise (e.g., measures of
the “health” of a business unit). Users might be motivated to learn special symbology (e.g., pilots) or may expect the displays to be self-evident (e.g., newspaper readers).
Because the data are condensed, ambiguity is difficult to avoid.
12:45-1:00 Discussant (Robert Frey, Director, Program in Quantitative Finance Applied
Mathematics and Statistics, SUNY-Stony Brook) explains the context and gives an
example of real data
1:00-1:30 Chris Wickens, University of Illinois at Urbana-Champaign
1:30-2:00 Peter Fisher, City University, London
2:00-2:30 Henry Rolka, Centers for Disease Control
2:30-3:30 Discussant presents reactions to ideas given by the speakers.
3:30 Break
3:45 Plenary discussion. Develop list of the following:
Lessons learned from the workshop.
Most promising research directions.
Next steps for developing this topic.
4:30 Adjourn
|
Subscribe to e-newsletters |
Feedback | Back to Top