Vigilance Parameter

The vigilance parameter is the control parameter of Adaptive Resonance Theory that determines how closely an input pattern must match an existing category before the system either accepts the match or creates a new category. It is not a learning rate, a threshold, or a confidence score in the conventional sense. It is a bifurcation parameter: a small change in vigilance can qualitatively alter the system's categorization behavior, switching it from coarse-grained to fine-grained classification.

At low vigilance, the system tolerates broad matches and produces few, general categories. At high vigilance, it demands precise matches and produces many, specific categories. The parameter is adjustable in real time, either by the network itself or by external modulatory signals — a form of meta-control in which the system regulates its own resolution. This makes the vigilance parameter a systems-theoretic primitive: it is the control knob that governs the scale at which a self-organizing system perceives structure. The same pattern can be one category or many, depending on the vigilance setting.