Summary: 688 nature neuroscience · volume 2 no 8 · august 1999
news and views
modulated, the recurrent connectivity of
the network will change.
The authors apply the idea of pointer
neurons to a network that selects a target
from the feedforward visual input it
receives. The neurons in the target-selec-
tion network are arranged in a spatial map.
A selected target is represented within this
map by a `hill' of activity in a localized
population of neurons (Fig. 2a and b). The
cluster of active neurons can be centered
at different places on the map to represent
different target locations. The selection of
a unique desired target requires this net-
work to generate a single hill of activity
even if the input is complex (as in Fig. 2c),
and this hill must correspond to a region
of the visual field that is of interest.