An idea proposed in the early 80s called synfire chains proposes that self-assembling networks of neurons can form in which small clusters or groups of neurons communicate with each other, possibly in a sequential linear fashion, more than other neurons. This concept is somewhat interesting if it can play a role in explaining how small networks of neurons spontaneously choose to communicate selectively with each other and possibly form cortical ensembles.
In a paper entitled “Development of Neural Circuitry for Precise Temporal Sequences through Spontaneous Activity, Axon Remodeling, and Synaptic Plasticity” by Jun and Jin, they explore the formation of these synfire chains with a computational model that utilizes leaky integrate and fire neurons along with two activity dependent learning rules. The more well known of these rules is the Hebbian type spike timing dependent plasticity (STDP), a phenomenological model that takes into account the relative timing between pre- and postsynaptic pairs of spikes.
Additionally, employing various thresholds of synaptic potentiation, the second rule they employ caps the number and strength of axonal remodeling — that is to say that when their neurons make strong connections with certain synapses (so-called supersynapses), they prune away other, weaker connections. It is this additional contribution that allows their model to spontaneously exhibit the synfire chain like behavior.