Guest editorial—Uncertainty and parameter space analysis in visualization

Information

• Publication Type: Journal Paper (without talk)
• Workgroup(s)/Project(s):
  • Visualization Group
• Date: June 2014
• Journal: Computer & Graphics
• Volume: 41
• Pages: A1 – A2

Abstract

Within the past decades visualization advanced to a powerful means of exploring and analyzing data. Recent developments in both hard- and software contributed to previously unthinkable evaluations and visualizations of data with strongly increasing sizes and levels of complexity.

Providing just insight into available data of a problem seems not to be sufficient anymore: Uncertainty and parameter space analyses in visualization are becoming more prevalent and may be found in astronomic, (bio)-medical, industrial, and engineering applications. The major goal is to find out, at which stage of the pipeline - from data acquisition to the final rendering of the output image - how much uncertainty is introduced and consequently how the desired result (e.g., a dimensional measurement feature) is affected. Therefore effective methods and techniques are required by domain specialists, which help to understand how data is generated, how reliable is the generated data, and where and why data is uncertain.

Furthermore, as the problems to investigate are becoming increasingly complex, also finding suitable algorithms providing the desired solution tends to be more difficult. Additional questions may arise, e.g., how does a slight parameter change modify the result, how stable is a parameter, in which range is a parameter stable or which parameter set is optimal for a specific problem. Metaphorically speaking, an algorithm for solving a problem may be seen as finding a path through some rugged terrain (the core problem) ranging from the high grounds of theory to the haunted swamps of heuristics. There are many different paths through this terrain with different levels of comfort, length, and stability. Finding all possible paths corresponds in our case to doing an analysis of all possible parameters of a problem solving algorithm, which yields a typically multi-dimensional parameter space. This parameter space allows for an analysis of the quality and stability of a specific parameter set. In many cases of conventional visualization approaches the issues of uncertainty and parameter space analyses are neglected. For a long time, uncertainty - if visualized at all - used to be depicted as blurred data. But in most cases the uncertainty in the base data is not considered at all and just the quantities of interest are calculated. And even to calculate these quantities of interest, too often an empirically found parameter set is used to parameterize the underlying algorithms without exploring its sensitivity to changes and without exploring the whole parameter space to find the global or a local optimum.

This tutorial aims to open minds and to look at our data and the parameter sets of our algorithms with a healthy skepticism. In the tutorial we combine uncertainty visualization and parameter space analyses which we believe is essential for the acceptance and applicability of future algorithms and techniques. The tutorial provides six sessions starting with an overview of uncertainty visualization including a historical perspective, uncertainty modeling and statistical visualization. The second part of the tutorial will be dedicated to structural uncertainty, parameter space analysis, industrial applications of uncertainty visualization and an outlook in this domain.

Additional Files and Images

Weblinks

• http://www.sciencedirect.com/science/article/pii/S0097849314000375

BibTeX

@article{Groeller_2014_UPS,
  title =      "Guest editorial—Uncertainty and parameter space analysis
               in visualization",
  author =     "Christoph Heinzl and Stefan Bruckner and Eduard Gr\"{o}ller",
  year =       "2014",
  abstract =   "Within the past decades visualization advanced to a powerful
               means of exploring and analyzing data. Recent developments
               in both hard- and software contributed to previously
               unthinkable evaluations and visualizations of data with
               strongly increasing sizes and levels of complexity. 
               Providing just insight into available data of a problem
               seems not to be sufficient anymore: Uncertainty and
               parameter space analyses in visualization are becoming more
               prevalent and may be found in astronomic, (bio)-medical,
               industrial, and engineering applications. The major goal is
               to find out, at which stage of the pipeline - from data
               acquisition to the final rendering of the output image - how
               much uncertainty is introduced and consequently how the
               desired result (e.g., a dimensional measurement feature) is
               affected. Therefore effective methods and techniques are
               required by domain specialists, which help to understand how
               data is generated, how reliable is the generated data, and
               where and why data is uncertain.  Furthermore, as the
               problems to investigate are becoming increasingly complex,
               also finding suitable algorithms providing the desired
               solution tends to be more difficult. Additional questions
               may arise, e.g., how does a slight parameter change modify
               the result, how stable is a parameter, in which range is a
               parameter stable or which parameter set is optimal for a
               specific problem. Metaphorically speaking, an algorithm for
               solving a problem may be seen as finding a path through some
               rugged terrain (the core problem) ranging from the high
               grounds of theory to the haunted swamps of heuristics. There
               are many different paths through this terrain with different
               levels of comfort, length, and stability. Finding all
               possible paths corresponds in our case to doing an analysis
               of all possible parameters of a problem solving algorithm,
               which yields a typically multi-dimensional parameter space.
               This parameter space allows for an analysis of the quality
               and stability of a specific parameter set. In many cases of
               conventional visualization approaches the issues of
               uncertainty and parameter space analyses are neglected. For
               a long time, uncertainty - if visualized at all - used to be
               depicted as blurred data. But in most cases the uncertainty
               in the base data is not considered at all and just the
               quantities of interest are calculated. And even to calculate
               these quantities of interest, too often an empirically found
               parameter set is used to parameterize the underlying
               algorithms without exploring its sensitivity to changes and
               without exploring the whole parameter space to find the
               global or a local optimum.  This tutorial aims to open minds
               and to look at our data and the parameter sets of our
               algorithms with a healthy skepticism. In the tutorial we
               combine uncertainty visualization and parameter space
               analyses which we believe is essential for the acceptance
               and applicability of future algorithms and techniques. The
               tutorial provides six sessions starting with an overview of
               uncertainty visualization including a historical
               perspective, uncertainty modeling and statistical
               visualization. The second part of the tutorial will be
               dedicated to structural uncertainty, parameter space
               analysis, industrial applications of uncertainty
               visualization and an outlook in this domain. ",
  month =      jun,
  journal =    "Computer & Graphics",
  volume =     "41",
  pages =      "A1--A2",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2014/Groeller_2014_UPS/",
}