The feedforward model with push-pull inhibition and same-phase
excitation is consistent with a number of observations obtained from
intracellular recordings of simple cells and described above. First
there is the motivating observation that excitation and inhibition are
spatially opponent (push-pull inhibition). Second, the orientation
tuning of inhibition to simple cells in the model is identical to that
of excitation: Both peak at the preferred orientation and fall off to
small values at the orthogonal orientation, as observed in
vivo. Third, the model is consistent with the cortical inactivation
experiments (Chung and Ferster, 1998, Ferster et al., 1996), which show
that the orientation tuning of visually-evoked responses does not
change when cortical cells are silenced, and show additionally that
the cortex amplifies responses to the LGN inputs 2- to 3-fold. Fourth,
the model explains why push-pull inhibition must be so strong relative
to feedforward excitation. The behavior of
Troyer et al. (1998)'s model
conforms to that of simple cells in two other important
ways. Iontophoresis of GABA
antagonists into the visual
cortex have long been known to degrade or even abolish orientation
selectivity (Tsumoto et al., 1979, Sillito, 1975), whereas intracellular
blockade of GABA
inhibition in a single cell has little
effect on the cell's orientation selectivity (Nelson et al., 1994). The
model closely mimics these behaviors. Because it is a variant of the
feedforward model and because it relies on the spatial organization of
the geniculate input for establishing orientation selectivity, Troyer
et al's model also exhibits a decrease in orientation tuning width
with increasing stimulus spatial frequency
(Jones et al., 1987, Hammond and Pomfrett, 1990, Vidyasagar and Sigüenza, 1985, Webster and De Valois, 1985).
Again, the existence of interneurons with the proposed
response properties is a key test of the model.