We begin by reviewing the assumptions of the now classical, connectionist-PDP
neural networks. For concreteness, we present one example of McClelland
and colleagues showing how a PDP network can model attentional focus
and interference in the Stroop color-naming task.
With this concrete example, we point out aspects of the neural code
that a connectionist neuron does - and does not - represent. In particular,
a connectionist neuron does not model the timing of a neural spike.
We cite empirical evidence that indicates that timing of neural spikes
are important for cognitive models. Spike timing seems to be needed
to explain processing speed in the brain (cf., Thorpe, Sereno) and to account
for data involving synchronized firing as the neural correlate
of attention.
We describe a minimal abstract neuron that does code for timing of neural
spikes. We then show simulation results that confirm the theoretical
work of Hopfield, Gerstner, and colleagues. These simulations show
how synchronized firing can emerge from networks of unsynchronized neurons.