Stimulus-Driven Unsupervised Synaptic Pruning in Large Neural Networks

We studied the emergence of cell assemblies out of locally connected random networks of integrate-and-fire units distributed on a 2D lattice stimulated with a spatiotemporal pattern in presence of independent random background noise. Networks were composed of 80% excitatory and 20% inhibitory units with initially balanced synaptic weights. Excitatory–excitatory synapses were modified according to a spike-timing-dependent synaptic plasticity (stdp) rule associated with synaptic pruning. We show that the application, in presence of background noise, of a recurrent pattern of stimulation let appear cell assemblies characterized by an internal pattern of converging projections and a feed-forward topology not observed with an equivalent random stimulation.