Visualizing Uncertainty: Computer Science Perspective

Transcription

1 Visualizing Uncertainty: Computer Science Perspective Ben Shneiderman, Univ of Maryland, College Park Alex Pang, Univ of California, Santa Cruz National Academy of Sciences Workshop, Washington, DC

2

3 What do we mean by uncertainty? Why is this an issue now?

4 Sources of uncertainty Measurement problems Ranges/Summaries Missing data Human ratings Potential deceptions Privacy protection Risk assessments Forecasts Scientific data Intelligence sources Statistical analyses Medical images Gene expression Simulations Financial models Weather Consumer ratings 4

5 Visualizations Text, statistical measures InfoViz SciViz 1D: Lists, documents, numeric ranges 2D: Geographic Info 3D: Scientific Visualization Multi-Variate: Information Visualization Temporal: Patient histories, web logs Tree: Taxonomies, org charts, directories Network: Social, communication 5

6 Text: Statistical uncertainty Text 1D 2D 3D Poll Dems Bush 14% Kerry 79% Margin of error +/- 3% Reps 81% 7% Multi-V Temporal Tree Network Time Variables Time Lost (Incident) Time to Fix (Incident) Computer Years Hours per Week * = p<.05 ** = p<.01 Frustration (N=372) **.293 ** *-.124 SW SE NW NE Rain inches Reliability low high low med 6

7 Risk/Danger vs. Trust/Validity/Security Text 1D 2D 3D Multi-V Temporal Tree Network Municipal Bonds Blue Chip Stocks Tech Stocks Real Estate Risk low med high med Terror Threat levels Highly About Highly Unlikely Unlikely Even Likely Likely 7

8 Text 1D: Ranges, variations & forecasts 1D 2D 3D Multi-V Temporal Tree Network Average Completion Times (with standard deviations) Type 1 Type 2 Type 3 Question Type 8

9 2D: Ball glyphs Delta (observation - forecast) Text 1D 2D 3D Multi-V Temporal Tree Network 9

10 2D: Arrow glyphs (Direction & velocity) 10

11 2D: Box glyphs Schmidt et al., 2004, Underwater Environmental Uncertainty, IEEE CG&A 11

12 2D: Grid with transparency/shading Lefevre, Pfautz & Jones 12

13 2D: Isolines with missing values Off-the-shelf software can give incorrect contours on data with lots of missing values. Modifications to contouring algorithm to account for large number of missing values. 13

14 2D: Pseudocolor shows mean values Luo, Kao & Pang, 2003, EuroVis 14

15 2D: Darkness = uncertainty (high stddev) Mean = hue Skew = 1/saturation Stddev= 1/value 15

16 2D: Separate layer 16

17 2D: Streamlines with binwise + 17 Luo, Kao & Pang,

18 2D: Weather forecast 18

19 2D: NOAA Storm Prediction Center 19

20 2D: Dual views & grid lines Dark shows pollution (Howard & MacEachren, 1996) Dark shows Certainty (Cedlink & Rhenigas, 2000) 20

21 Dual maps for Rate and reliability Bivariate color scheme Double hatch shows unreliable 21 (MacEachren et al., 1998)

22 2D: Gray for missing & interpolation Gray shows missing & interpolated value, superior to using black only Twiddy, R., Cavallo, J., and Shiri, S Restorer: A visualization technique for handling missing data. IEEE Visualization 94,

23 3D: Fuzzy molecular surface: HIV protease Text 1D 2D 3D Multi-V Temporal Tree Network Crisp molecular surface Probe radius 1.4 Fuzzy molecular surface Probe radius Crisp molecular surface Probe radius 5.0 Fuzzy molecular surface Probe radius 5.0 Lee & Varshney (2002), UM Graphics and Visual Informatics Lab

24 3D: Fuzzy molecular densities 24

25 3D: Uncertainty dust 25

26 3D: Color & opacity 26

27 Multi-V: Database/spreadsheet tables Text 1D 2D 3D Multi-V Temporal Tree Network 27

28 Multi-V: Database/spreadsheet tables Text 1D 2D 3D Multi-V Temporal Tree Network 28

29 Temporal: Granularity of time Text 1D 2D 3D Multi-V Temporal Tree Network 29

30 Temporal: Granularity of time Text 1D 2D 3D Multi-V Temporal Tree Network 30

31 Temporal: Granularity of time Text 1D 2D 3D Time Uncertainty Hi Med Multi-V Temporal Tree Network Low 31

32 Temporal: Granularity of time Text 1D 2D 3D Multi-V Temporal Tree Network 32

33 Tree: Topology, values & names Text 1D 2D 3D Certainty Hi Med Low Very Low Multi-V Temporal Tree Network -Gene Ontology -Tree of Life -Medical Subject Heading (MeSH) -Chain of command/org chart 33

34 Tree: Topology, values & names Text 1D 2D 3D Certainty Hi Med Low Capacity Hi Multi-V Temporal Tree Network Med Low 34

35 Tree: Topology, values & names Text 1D 2D 3D Multi-V Temporal Tree Network Leland or Lee Mike or Michael Alex or Alan Barbara Scott Ben or Benjamin Diane or Di Ed or Edward or Eddie Certainty Hi Med Low 35

36

37 Text Network: Social relationship 1D 2D 3D Multi-V Temporal Tree Network 37

38 Network: Communication capacity 38 J.A. Brown, McGregor A.J and H-W Braun.

39 Network: Node & edge uncertainty Text 1D 2D 3D Certainty Hi Med Low Flow Hi Med Low Multi-V Temporal Tree Network Capacity Hi Med Low 39

40 Next steps Explore novel approaches to: Text: standard terms, percent, probabilities Box plots, whiskers, ranges Uncertainty glyphs, isoclines, Contours, surfaces, volume rendering Hue, saturation, value, focus, haze, dust, Dual views, probes Animation, blinking, shaking, flipping, Sound, haptics, 40

41 Next steps Heighten awareness of the problem among public, professionals, researchers & developers Support multi-valued data representation standards Explore techniques for each data type Develop guidelines for implementers: Data formats Interactive interfaces Visual presentations Develop human performance evaluation methods Publish benchmark datasets & evaluation metrics Form guidelines for how to propagate/integrate uncertainty markers 41