We consider a randomly connected neural network with linear threshold
elements which update in discrete time steps. The two main features of
the network are: (I) equally distributed and purely excitatory connec
tions and (2) synaptic depression after repetitive firing. We focus on
the time evolution of the expected network activity. The four types o
f qualitative behavior are investigated: singular excitation, converge
nce to a constant activity, oscillation, and chaos. Their occurrence i
s discussed as a function of the average number of connections and the
synaptic depression time. Our model relies on experiments with a slic
e culture of disinhibited embryonic rat spinal cord. The dynamics of t
hese networks essentially depends on the following characteristics: th
e low non-structured connectivity, the high synaptic depression time a
nd the large EPSP with respect to the threshold value. Copyright (C) 1
996 Elsevier Science Ltd