Consensus versus competition in neural networks: a comparative analysis of three models

✍ Scribed by F.S. Montalvo

• Publisher: Elsevier Science
• Year: 1975
• Weight: 556 KB
• Volume: 7
• Category: Article
• ISSN: 0020-7373
• DOI: 10.1016/s0020-7373(75)80016-2

✦ Synopsis

Three neural network models are compared in terms of underlying mechanisms. One is a visual model that segments a scene in terms of disparities. A second is one of the frog tectum designed to pick out the position of maximum "foodness" for the frog. The third is one of the reticular formation involved in the commitment of an organism to one gross mode of behavior or another. The parts that inhibition, unit coupling, network length, input features, nonlinearities, and time-varying functions play in the final results of these networks are discussed. A state-reduction scheme is proposed that enables an analysis of a network's behavior in terms of meaningful groups of units rather than single unit activity. This involves a decomposition into a dimension along which units compete vs. a dimension along which units reach consensus. Units reaching consensus tend to group themselves into aggregates that in turn compete to gain dominance in mode decisions.