Discovery of occluded structures and their behavior is inherently difficult due to data volume and fragmentation, but is exacerbated when the behavior must be identified before it culminates in unacceptable actions. This means that the analyst must integrate multiple pieces of data, bearing no explicit indication of their degree of interrelatedness, into a coherent model of the domain representing all the essential objects and interactions between them. As the new data arrives, this model must be adjusted. Coherent representation of essential domain entities and interrelations (structure) constitutes domain ontology (or ontological model). The difficulty in the development of such models lies in combinatorics and circularity (Heuer, 1999):
A) Multiple data admit large (often astronomical) numbers of combinations.
B) In the analysis, some combinations are amplified and some other ignored, depending on the model adopted by the analyst.
C) Initial model’s confirmation biases affect subsequent data search and interpretation, entailing circularity.
The role of the Structure Discovery Engine (SDE) is to augment analysis and aid the analyst in overcoming the difficulties posed by combinatorics and circularity. Figure 1 helps to appreciate the scale of such difficulties.
 |
| A | B | C |
|
Figure 1.
A. Data gathering starts by circumscribing a number of geographically dispersed objects subject to observation.
B. Observation of co-variation in the object’s behavior produces a complex relational network (co-varying objects are connected by links).
C. The number of interrelations is a power function of the number of objects.
|
The VAN process is directed towards discovering a coherent spatial structure underlying the growing relational network in Figure 1B. The structure is coherent to the extent that the main components preserve their composition in the course of subsequent observations. Discovery of such bounded and stable components enables inference about causal interdependencies between the objects’ behavior, and behavior prediction.
Structure does not manifest in patterns, and VAN is not about pattern recognition.
Patterns capture some discrete event sequences (episodes) that are cut out and disconnected from the flow of changes experienced by the objects of interest. By contrast, connected structures (such as relational networks comprising all the objects) allow representation of continuous flows. Conceptualizing investigation or analysis as pattern recognition (template matching) prevents discovery of emergent structure and consigns one to the past.. Figure 2 compares pattern recognition and structure discovery.
Figure 2.
Pattern recognition has inherent limitations, in that only representation and comparison of discrete episodes involving object subsets are allowed. By contrast, connected structures can represent continuous history of multiple objects, and allow analysis unavailable in pattern recognition.
The VAN model asserts that analysis of structural transformations is the “missing link” in the current decision technology, in particular, in computer-aided intelligence analysis. This assertion is motivated by neurobiological evidence concerning memory organization and processes in the humans and other higher animals.
|
