Anatomical surveys show that 80% of the synapses of long-range lateral connections connect directly between pyramidal cells, which are thought to make excitatory synapses only (Gilbert et al., 1990; McGuire et al., 1991). The other 20% of the connections target inhibitory interneurons which in turn contact the pyramidal cells, and thus represent inhibitory connections. Even though the inhibitory connections are outnumbered, the net effect at the columnar level has been difficult to establish conclusively with anatomical studies. For instance, the interneurons often synapse at regions such as the soma where their effects may be larger than those of excitatory neurons, which synapse farther out on the dendrites (Gilbert et al., 1990; McGuire et al., 1991).
Physiological and psychophysical evidence now indicates that the balance between these two types of connections is actually contrast-dependent: the influence of the lateral connections impinging upon a neuron is mildly excitatory when the surrounding area is activated by a low-contrast stimulus, and strongly inhibitory when the surround is activated by a high-contrast stimulus. This has been demonstrated conclusively at the cellular level in tissue slices (Hirsch and Gilbert, 1991), and more recently in vivo at the level of cortical columns (Weliky and Kandler, 1995). As discussed below, it has been hypothesized that these complex connections help enhance the ability to detect low-contrast inputs while suppressing redundant activation for high-contrast inputs.
However, it remains unclear how exactly this contrast dependence is implemented in the cortex. An early proposal was that the inhibitory interneurons are inherently more effective than the direct excitatory connections, but have a higher threshold for activation (Sillito, 1979). At very low stimulus levels, the excitatory effects would predominate, but at high levels the inhibitory interneurons would become progressively more active and eventually would suppress the response of the target cell.
Douglas et al. (1995) have presented a more detailed proposal which takes into account the recurrent circuit provided by the type of short-range excitatory lateral connections found in LISSOM. That paper shows how the relatively few lateral inhibitory connections could be effective enough to dominate the response even though they are fewer in number. Simplified versions of such circuits have been modeled by Stemmler et al. (1995) and Somers et al. (1996). They propose that these complex connections help enhance the ability to detect weak, large-area stimuli while suppressing spatially redundant activation for strong stimuli.
The current RF-LISSOM model emphasizes the suppressive effects only. Since all inputs used for self-organization have been high-contrast, the assumption that the effects are primarily inhibitory is well-founded. Because the amount of synaptic change due to the learning mechanisms is very small for low-contrast inputs, the behavior for such inputs does not significantly affect self-organization. Furthermore, since all the inputs used in this thesis are high-contrast, the results presented for the tilt aftereffect should not depend upon this simplifying assumption.
Future versions of RF-LISSOM may be extended so that the effect of the long-range connections varies according to circumstances. However, it is not yet clear what type of extension is necessary to account for the phenomena. One possibility is to multiply each lateral connection impinging upon a cell by a factor that depends upon the activation of the cell. This would treat any lateral input as a mild excitation if the cell is inactive, and as strong inhibition if the cell is strongly active (Sillito, 1979; Somers et al., 1996). Another alternative extension is to use the net lateral input to the cell to compute this scaling factor. Future work may help to clarify which of these two, or others not yet formulated, is most appropriate.