Concept Visualization for Ontologies of Artificial Intelligence

1
Concept Visualization forOntologies of
Artificial Intelligence

• Yu Suo
• TJHSST Computer Systems Lab
• George Mason University

2
Background

• The Disciple Learning Agent
• An artificial intelligence that can be adapted
  for many purposes
• military operations
• Teaching
• traveling

3

• Decision aid
• Subject-matter experts can train and use when
  making decisions under stressful, complex, and
  constrained conditions
• Tool was developed and used under the Defense
  Advanced Research
• Rapid Knowledge Formation programs

4
Purpose

• research, create, and program a concept
  visualization tool
• easily browse small to large scale semantic
  networks and ontologies
• Used mainly by those who are not proficient at
  Computer Science

5

• Aims to distribute information about the
  artificial intelligence in an easy and usable
  way.

6
Procedure and Methodology

• written in JAVA and called JDisciple
• Eclipse is used to provide a basic organized
  structure
• Simple to advanced cases
• Small to large cases
• Programmed in iterations to keep track of
  progress

7
Algorithms

• Algorithm to arrange/select elements
• Greedy algorithm serves as a base/control
• Paired with one or more heuristic functions
• Dynamic Programming or more advanced methods may
  be used in the future

8
Algorithms

• Graph layout algorithm to place elements
• Well structured
• Find space, make space
• Clear and readable
• Process all types of graphs
• Minimize cross-links
• Accessible to end-users

9
JGraph for Visualizing Networks

• JGraph is an open source graph visualization
  library written in Java.
• JGraph is a fully Swing compatible component.
• The current DJGraph is derived from an older
  version of JGraph.
• Simplifies drawing process for networks, trees,
  and maps.
• Fairly versatile Vertices and edges may be
  shapes, images, other Swing components, or
  animations.

10
(No Transcript)
11
Current Progress

• Current algorithm only shows two types of
  relationships.
• Based on purely heuristic searching.
• Layout algorithm is limited to 50x50 pixel
  blocks.
• Very space dependent and unintelligent space
  searching

12
Work Needed

• Expand current algorithm to work on all general
  types of relationships.
• Add additional or support heuristics for better
  greedy selection.
• Optimize current space finding algorithm or
  program new method.

13
(No Transcript)
14
Usefulness?

• Current graph layout algorithms often fail at
  very general cases
• Graphs are used to show should be made clear
  and readable!
• Shows the complexity involved in a full fledged AI