It is stipulated that all the following three categories of dynamic phenomena must occur in any realistically behaving neural-network model: (i) Activation effects, (ii) Adaptation effects, (iii) Plasticity control effects.

In most neural models one only has activation and adaptation present. The Self-Organizing Map (SOM) algorithm devised by this author (Kohonen, 1982, 1995a) is the only neural-network model that involves all the three phenomena. Its modeling laws include the following partial functions:

1.˜Some parallel computing mechanism for the specification of a cell in a piece of cell mass whose parametric representation matches with or which responds to the afferent input best. This cell is called the “winner.” 2.˜Control of some learning factor at the cells in the neighborhood of the “winner” so that only this neighborhood will be adapted to the present input. By virtue of the “neighborhood learning,” the SOM will form spatially ordered maps of sensory experiences, very much resembling the brain maps.

The newest version of the SOM is the ASSOM (Adaptive-Subspace SOM). The adaptive processing units of it are able to represent signal subspaces, not just templates for original patterns. A signal subspace is an invariance group: therefore the processing units of the ASSOM are able to respond invariantly, e.g., to moving and transforming patterns, in a somewhat similar way as the complex cells of the cortex do.