ABSTRACT
Mathematical description of lateral inhibition makes it possible to relate and compare the network models with those investigated for their computational properties. This chapter describes some of the unique processing capabilities of these networks which make them of great computational interest and introduces the closely related class of additive models. The classification of network models into additive and shunting, which follows Grossberg's convention is meant to highlight the distinctions by providing a comparative framework. A general-purpose model of preattentive vision which includes shunting nets as a primary module of the architecture is described by Grossberg et al. The chapter reviews some of the processing capabilities unique to networks with multiplicative inhibition. The crucial feature of the model is the control of conductances by neighboring cells which gives rise to the multiplicative terms, and hence the name multiplicative lateral inhibition. The stability, automatic gain control, and wide dynamic range properties of such feedback systems are well known in control theory.
