Learning excitatory-inhibitory neuronal assemblies in recurrent networks

In sensory circuits with poor feature topography, stimulus-specific feedback inhibition necessitates carefully tuned synaptic circuitry. Recent experimental data from mouse primary visual cortex (V1) show that synapses between pyramidal neurons and parvalbumin-expressing (PV) inhibitory interneurons tend to be stronger for neurons that respond to similar stimulus features. The mechanism that underlies the formation of such excitatory-inhibitory (E/I) assemblies is unresolved. Here, we show that activity-dependent synaptic plasticity on input and output synapses of PV interneurons generates a circuit structure that is consistent with mouse V1. Using a computational model, we show that both forms of plasticity must act synergistically to form the observed E/I assemblies. Once established, these assemblies produce a stimulus-specific competition between pyramidal neurons. Our model suggests that activity-dependent plasticity can enable inhibitory circuits to actively shape cortical computations.