Examining heterogeneous weight perturbations in neural networks with spike-timing-dependent plasticity

摘要

Large-scale cortical networks employing homeostatic mechanisms and synaptic plasticity rules have been shown to differentiate into neural ensembles when common stimuli are applied in tandem to selected subsets of neurons. These ensembles were found to be stable in response to small perturbations to synaptic strengths—such ensemble stability is a critical feature for network-based memory. Previous studies applied relatively simple perturbations to probe the stability of the network—all synapses within a given population were lowered by a uniform percentage. The goal of this work has been to analyze whether more complex perturbations can reveal more information about network stability. Towards this aim, we constructed a reduced stochastic Wilson-Cowan model, which captures the same perturbation phenomenon observed in spiking simulations, but which is analytically much simpler. We found that when the mean self-excitatory synaptic weight for a population was preserved, perturbations that were distributed more evenly among synapses would lead to a more stable response than focused perturbations, and that this was caused by quantization of neural activity levels within a population.