visual perception and mixed-initiative interaction for assisted visualization design
TRANSCRIPT
Visual Perception and Mixed-Initiative Interaction for Assisted Visualization Design
by Christopher G. Healey, etc.
Info863Kai Li
10/27/2015
What the paper is about?
• The paper is about the design of a visualization system called “ViA” using the techniques of– Perceptual guidelines from human vision, and– AI-based mixed-initiative search strategy
• and an evaluation of the system on these two aspects.
What’s new in ViA?
• Shortcomings of existing visualization systems: – They cannot suggest users how to best represent
their data based on their needs, and, – They have limited abilities to integrate perceptual
rules in the creation of visualizations.• Users who are not visualization researchers
often repeat the same design principles, which causes a number of inefficiencies, especially for multidimensional datasets.
ViA’s architecture• The goal of the ViA project is to help users getting access to a
collection of visualization designs using the same data, and construct visualizations for their own data, that are effective, multidimensional, transparent, application independent, and extensible.– ViA seems to create only 2D and static visualizations.
• Components of ViA:– User input: attribute importance, spatial frequency, value type, task.– Evaluation engine– Hints: feature swap, importance weight modify, discretize, task
removal.– Search algorithm
ViA’s architecture
Perceptual foundations of information visualization
• Visualization is the mapping between data properties to visual properties. – M(V, Φ)
• Previous studies on how human visual system actually “sees” fundamental properties of color and texture in an image formed the foundation of the visualization in this study.– Color:
• Luminance, hue, saturation…– Texture:
• Size, spatial packing density, orientation, regularity…– Interactions between visual properties
• Visualization of 2D flow in a simulated supernova collapse– Stroke orientation: flow direction– Color: magnitude– Size: pressure
Mixed-Initiative Interaction
• Mixed-initiative interaction “[combines] automated services and user control to form mixed-initiative interaction.”– Participants are enabled to contribute their
unique strengths towards solving a common problem.
– Controls are shifted between parties who are the most qualified to solved the problems in each step.
How mixed-initiative interaction is applied to ViA
• Consideration: how to manage the uncertainty about a user’s goals during problem solving.
• A common solution is “to use Bayesian agents to model goals and construct utility measures based on probabilistic relationships”
Evaluation of ViA
• Three separate components are considered for evaluating ViA’s performance:– ViA’s ability to locate the best visualization mappings,
compared with an exhaustive search and simulated annealing and reactive tabu search algorithms
– Mixed-initiative interaction’s influences on ViA’s visualization recommendation
– How ViA system is applied in another project focusing on E-commerce auction environments
Evaluation of search performance
• Three metrics were calculated to evaluate the search performance:– Optimality: the evaluation weight of the best
mapping found by an algorithm– Efficiency: the number of visualization evaluated
before an algorithm finds the first optimal mapping– Completeness: the total number of visualizations
found by an algorithm relative to the total number of mappings with the maximum evaluation weight.
Evaluation of search performance (cont.)
• ViA (Hint) produced the highest efficiency when it found the first optimal result, but also got the least number of optimal mappings among the three algorithms.
• The authors argue that hint-based search, because its searching procedure is based on hints rather than local regions (RTS), has the potential to generate diverse results.
Evaluation of Mixed-initiative interaction’s influences on recommendation
• Experiments were conducted in three modes:– Without mixed-initiative interaction (ViA-N). System was run by
fixed settings.– With mixed-initiative interaction. Decisions are controlled solely
by expected utility (ViA-MI)– With mixed-initiative interaction. Decisions made by users (ViA-UI)
Limitations of ViA
• Visualizations created ViA are based on geometrics glyphs, which might not fit certain types of datasets and analysis.
• ViA is application-independent, thus may not support some domains very well.
• No controlled experiments have been conducted to prove ViA’s advantages in real-world contexts.
• Users might not agree with ViA about which visualization is the “best” choice.
Compare and contrast
Compare and contrast (cont.)
Compare and contrast (cont.)
Questions
• Is there better (or best) visualizations?– What criteria are we using to evaluate?• This paper seems to evaluate visualizations based on
perceptual theories (how different data properties should be mapped to visual cues, and how these visual cues interact with each other), which is sort of an etic perspective and only necessary conditions.
– What other criteria other than those in this paper can we use to evaluate visualizations?• http://hint.fm/wind/
Reference
• Healey, C., Kocherlakota, S., Rao, V., Mehta, R., & St Amant, R. (2008). Visual perception and mixed-initiative interaction for assisted visualization design. IEEE Transactions on Visualization and Computer Graphics, 14(2), 396–411. http://doi.org/10.1109/TVCG.2007.70436
Thank you!