Formation of working memory in networks of spiking neurons and spike-driven synapses

We study the evolution of the collective behaviour of a recurrent network of integrate-and-¯re neurons, connected by synapses whose e±cacies are subjected to a spike-driven, Hebbian long-term potentia-tion and a homosynaptic depression, as well as a short-term, frequency dependent depression. Through a mean ¯eld analysis, and numerical simulations of the coupled neural and synaptic dynamics, we show how repeated stimulations of the network can dynamically produce a synaptic structure able to support persistent, collective neural states of stimulus-selective activity, that have been proposed since long as a substrate for `working memory'. We discuss constraints relevant to the stability of the `learning ' process, and the related role of synaptic depression.