Subscribe to DSC Newsletter

Grand Challenge: Unified Visual Data Representation

All creatures have the ability to sense the surrounding world, but in various ways and degrees. You might envy the bloodhound’s exceptional nose, but humans possess visual prowess that (although it doesn’t match the eagle’s eye in distance) is unsurpassed in the ability to detect and make sense of patterns. Our eyes and brains work as a team to discover meaningful patterns that help us make sense of the world [1].

Digital computers take input in direct quantitative form constructed from digits. Human extract most of quantitative information from 3D visual environment: distances between observable objects, sizes of objects, colors intensity and hue, proximity, similarity, symmetry … “A striking fact about human cognition is that we like to process quantitative information in graphic form” [2].

A pattern recognition (sense-making) stage comes after low-level extraction stage of human visual perception.

A relatively recent in human history process of visual symbolic information reading (e.g. letters & numbers) and their derivatives (text, tables, etc.) severely limit the amount of quantitative information extracted from this visual input. Therefore we disable substantial part of information directed to pattern recognition (sense-making) stage.

This is why Graphs (e.g. line chart, bar chart, etc.) are very powerful tools for understanding digital data: they mimic human visual environment by encoding digital data as locations, distances, sizes, colors, etc., therefore enabling power of human standard pattern recognition (sense-making) process. In short, the following information conversions sequence is taking place:

Quantitative Information in Pure Symbolic Form --> Graphic Representation --> Extraction of Quantitative Information from Graphic Representation --> Pattern Recognition

Do the same by using derivatives of symbolic representation (e.g. tables, text) makes sense-making difficult and sometimes impossible (cases of medium to large amount of symbolic data) due to limitation of human cognitive capabilities to memorize symbolic information. This is why presenting information in graphic form or information visualization is so important for its understanding.

The Grand Challenge

The exponential technological development created overwhelming amounts of disparate, conflicting, and dynamic information and, therefore, huge needs to analyze and understand information. Despite vast amount of newly created effective digital algorithms for information analysis, attempts to completely remove human from a decision loop been unsuccessful. Altogether overwhelming analyzing needs, deficiency of automated algorithms and previous visualization methods created enormous demands for information visualization and visual analytics.

You can find short review of information visualization and visual analytics in attached review [4], MILESTONES part (page 3) and in research agenda [3]. Different Charts or visual representations correspond to different data types and designed to solve specific problems. But how to combine vast amounts of disparate data types together in unified visual representation suitable for discovery and satisfying visual information seeking mantra?

“The holy grail of information visualization is for users to gain insights. In general, the notion of insight is broadly defined, including unexpected discoveries, a deepened understanding, a new way of thinking, eureka-like experiences, and other intellectual breakthroughs”[4]. To make search for insight feasible between other requirements, we must have:

  • A) Reduce the search per se, such as by representing a large amount of data in a small space [3]
  • B) Create methods to synthesize information of different types into a unified VISUAL data representation (VUDR) [3]
  • C) Create a new science of interaction to support visual analytic [1]. Part of this challenge required increase level of dynamism in presenting relevant information; create synchronous multi-level interaction space that enhances human association capabilities.

Data transition B) to VUDR cannot be done without moving to high level of abstraction. It will compact the data and positively affect A). Then interaction C) will be done on high level of abstraction. It will let to have synchronous high capacity multi-level information presentation from high abstraction level to connected low levels including raw data.

Therefore the critical to the above is to find a solution of challenge B) or to find unified visual data representation (VUDR).

Interestingly VUDR design can be prompted by the process of finding insight. Let’s consider the following: the burst of recognition often happened then one arrives at insights by linking previously unconnected thoughts. The theory is computational and it is possible to formulate the search for insights as a problem of searching for the potential linkage between even the most unthinkable relations. Initial studies of transformative discoveries such as Nobel Prize winning discoveries are particularly promising. This approach is particularly relevant to visual analytics and insight-based evaluative studies because they can characterize insightful patterns in terms of structural and temporal properties [4][5].

References

[1] Stephen Few: “Visual Pattern Recognition”, COGNOS, Innovation Center, 2006.
[2]Pinker, S.:“A theory of graph comprehension”; In R. Freedle (Ed.) Artificial intelligence and the future of testing, (pp. 73–126). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc., 1990.
[3] “Illuminating the Path: The Research and Development Agenda for Visual Analytic”. January 1, 2005. James J. Thomas (editor), Kristin A. Cook (editor). 
[4] Chaomei Chen: “Information visualization”, 2010 John Wiley & Sons, Inc. WIREs Comp, Stat 2010 2 387–403. 
[5]Chen, C. | Chen, Y. | Horowitz, M. | Hou, H. | Liu, Z. | Pellegrino, D : “Towards an explanatory and computational theory of scientific discovery” ; Journal of Informetrics, Volume 3, Issue 3, July 2009, Pages 191-209.

Views: 1084

Tags: Analytics, Data, Discovery, Intelligence, Visual, Visualization

Comment

You need to be a member of Data Science Central to add comments!

Join Data Science Central

Comment by Georges Grinstein on June 30, 2014 at 9:09am

Your grand challenge is somewhat on target but also somewhat off. First at a high level it is correct but so are many high level views.A UVDR without incorporating the data, the tasks, the user (and even the medium) is weak and that's what has been done to date (however there is research to address this).

The first item (Reduce the search per se, such as by representing a large amount of data in a small space) is right on - but again incorporation of the complete parameters is key. Big data must be reduced but there are lots of issues in data reduction - and there are only 1-4 million pixels on the screen.

The second (Create methods to synthesize information of different types into a unified VISUAL data representation) is also on but here's where some misinterpretation occurs. I've been co-chairing with PNNL the VAST Challenges (with Kris Cook and Mark Whiting as co-chairs) and a unified view is valuable but still fails often. Unification just does not hack it unless lots of things are paid attention to, especially the task (think situational awareness vs surgery vs budgetary planning).

The third (Create a new science of interaction to support visual analytic [1]. Part of this challenge required increase level of dynamism in presenting relevant information; create synchronous multi-level interaction space that enhances human association capabilities.) is on target with respect to harnessing human association.

However all this is vague without evaluation and measurements that actually involve the human and in more details than has been done to date. Such measurements require a theory of visualization. Such a theory should provide predictions that can be tested to invalidate the theory and help evolve it. There are a few theory proposals out there, such as the one by Chen you mention. What is a grand challenge is to develop such a theory. Each item you mention should be involved in that theory (somehow) and measurably predictable (somehow). I am being vague here as it would depend on what metrics are proposed developed. Robert Patterson and others at the Air Force, AFRL and WPAFB, have developed a first pass at a theory (A human cognition framework for information visualization - see Computers and Graphics April 2014). I was a co-author but it's their insight and our collaborative efforts over 3 years that led to this, what I consider, breakthrough in opening up the possibility of measuring various cognitive aspects.

Thanks for putting this down on paper and I hope researchers (and analysts) design and use human cognition metrics as proposed in the paper.

Videos

  • Add Videos
  • View All

© 2019   Data Science Central ®   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service