  |
The key distinguishing characteristic of the VAN process is the duality of dynamic partitioning (clusterization) and integration: Every new piece of information is added to the growing structure, in the form of a new link or link weight adjustment, followed by a global structure re-organization such that the degree of changes in every entity (bounded node group) remains commensurate with the entire preceding history of that entity as well as the preceding history of the overall hierarchical structure. The process is modeled explicitly on the psychological mechanisms of assimilation (adding new information to the existing memory structure) and accommodation (adaptive structural re-organization in response to the addition). The concept of re-grouping as the mechanism of accommodation comes from the recent neurobiological findings demonstrating that adaptive partitioning of neuronal populations into bounded ensembles constitutes the basic neuronal mechanism of adaptive behavior (Georgopoulos, 1993).
The VAN process of structure discovery resolves some of the main limitations of the current technologies.
Evidence Extraction
- VAN is designed to represent relational/structural facts.
- VAN complements statistics, provides what statistics cannot give.
- VAN methods are domain independent.
- VAN has unlimited combinatorial capacity.
- VAN employs dynamic Data Schema.
- VAN has unlimited scalability.
Link Detection
- VAN is domain Independent.
- VAN offers methods for precise and VERY FAST fact Searching.
- VAN is all about detecting Complex Structures.
- VAN is based on the notion of Dynamic Adaptive Granularity.
- VAN has Multiple Levels of Abstraction.
- VAN inherently integrates Multiple Relational Types, e..g., temporal, organizational, transactional.
VAN learns Structures, not Patterns.
|
|
