  |
Memory organization and processes underlie cognition and are responsible for the unique efficiency of human adaptive behavior. There are reasons to believe that representation of the environment in the human memory takes the form of globally connected associative networks, subject to dynamic self-partitioning into weakly coupled clusters.
Associative networks are formed in the cortex, having neuronal columns and other aggregates as the nodes are linked via axonal interconnections. Repetitive simultaneous activation of distant nodes strengthens the links between them and, vice versa, prolonged absence of activation causes association decay. In this way, the history of activation (that is, the history of organism-environment interaction) is recorded as variation of associative strength between the neurons responding to the states of environmental variables (stimulation). Global connectivity is the central definitive feature of the associative memory structure. The other definitive feature is continuous self-partitioning of the associative network carried out by the biological mechanism of lateral inhibition. Global connectivity allows self- partitioning into stable non-overlapping components (groupings), experienced as disambiguation of the environment, or comprehension (Yufik, 2002, 2001, 1998). Figure 3 illustrates the underlying biological architecture.
 |
 |
Figure 3.
Information is proceeded in a hierarchy of cortical layers (left). Lateral (horizontal) connections in the layers underline intra-layer associative structures amenable to adaptive self-partitioning. Sensory processing involves multi-stage integration and associative clusterization (above). Global integration is due to convergence of sensory streams in the hippocampus (after Eichenbaum et al, 1998) |
It has been demonstrated that operations of logical reasoning and syntactical composition (language) can be implemented as biologically feasible operations on connected self-partitioning networks (Yufik, 1998). The VAN model has motivated computational techniques proved to be efficient in a variety of DoD applications centered on large-scale real-time adaptive decision-making and resource optimization (e.g., sensor management). Computational advantages of the method are derived from clusterization procedures motivated by biological analogies and residing at the core of Structure Discovery Engine.
|
|
