13 Modeling Feature Selectivity in Local Cortical Circuits

Neuronal representations of the external world are often based on the selectivity of the responses of individual neurons to external features. For example, many neurons in visual cortex respond preferentially to visual stimuli that have a specific orientation (Hube1 and Wiesel 1959), spatial frequency (Campbell et al. 1969), color (Hube1 and Wiesel 1968), velocity and direction of motion (Orban 1984). In motor systems, neuronal activities are tuned to parameters of a planned action such as the direction of an arm reaching movement (Georgopoulos, Taira, and Lukashin 1993), or the direction of a saccadic eye movement (for a review, see Sparks and Mays 1990). It is often assumed that the primary mechanism underlying the response properties of a neuron resides in the transformations of sensory signals by feedforward filtering along afferent pathways (e.g., Hubel and Wiese11962). Although, in some cases, the feedforward model is consistent with our understanding of the nature of afferent inputs (Reid and Alonso 1995;, Chapman, Zahs, and Stryker 1991), in others, particularly in motor areas, the relation between afferent inputs and cortical neuronal response properties 'is not obvious. Moreover, neurons in cortex, even in the input stages of primary sensory areas, receive most of their excitatory inputs from cortical sources rather than from afferent thalamic nuclei These facts and other experimental and theoretical considerations suggest that local cortical circuits may play an impor'-tant role is shaping neuronal responses in cortex. , In this chapter we review the theoretical study of the function of local networks in cortex in relation to feature selectivity. By "local network" we mean an ensemble of neurons that respond to the same patch of the external world and are interconnected by recurrent synaptic connections. Typically, a local network sp.ans roughly 1mm2 of cortical surface and is assumed to consist of subgroups of neurons each of which is tuned to a particular feature of an external stimulus. These subgroups will be called "feature columns" and the whole network a "hypercolumn," in analogy with. the "ice cube" model of primary visual cortex (Hubel 1988; for a review of local cortical circuitry, see Martin 1988; Gilbert 1992; Abeles 1991). The complexity of neuronal dynamics and circuitry in cortex precludes systematic investigation of t4e properties of realistic large-scale neuronal models of local cor