Methods for deriving competitive activation mechanisms

Inhibitory interactions in connectionist models have traditionally been implemented using inhibitory connections. Recently, an alternative approach based on competitive activation mechanisms has been proposed. To make this alternative approach practical in large and complex models, it is necessary to formulate general methods for creating task-specific competitive activation mechanisms. The author presents two such methods that have been applied successfully: heuristic computer-assisted search and decomposition of global optimization criteria. The usefulness of these two methods is demonstrated through simulations that establish the effectiveness of the competitive activation mechanisms so produced.<>