Inhibitory control of plasticity promotes stability and competitive learning in recurrent networks

Current voltage-dependent plasticity models assume that any plastic event needs a postsynaptic action potential and use induction mechanisms that are insensitive to postsynaptic voltage fluctuations. However, experiments have demonstrated that inhibitory inputs can affect plasticity updates, likely through voltage-dependent effects. To address such findings, here we introduce the novel Voltage-Dependent Pathway model (VDP) where the induction variables track the membrane potential, increasing its sensitivity to fast fluctuations. We first confirm that this model produces spike-timing dependent plasticity (STDP) kernels. In addition, inhibitory inputs temporally close to plasticity-inducing events change the potentiation/depression ratio as reported experimentally, an effect that cannot be reproduced with previous models. This allows inhibition to control the direction of plasticity events. We next show that this has a stabilising effect in recurrent networks, and induced circuit-level competitive effects. In spontaneously active recurrent networks where both excitatory and inhibitory synapses are plastic, stronger inhibition due to increasing activity in the excitatory population slows weight growth, thus stabilising the population activity. Furthermore, a network receiving inputs from two or more afferent pathways that are correlated within each pathway, but not between them, will develop neurons selective to only one of the pathways. This effect, which simulates developmental processes such as eye-specific segregation, depends on competition through recurrent inhibition: a selective neuron will activate inhibitory feedback, which in turn causes depression of afferent synapses from the same pathway to other neurons. This work predicts that plasticity is sensitive to fast membrane potential fluctuations, and that this leads to competitive effects that do not rely on weight normalisation, but are instead driven by recurrent inhibition.