The role of temporal sensitivity of synaptic plasticity in representation learning
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Synaptic plasticity, the process by which synapses change in an activity-dependent manner, is assumed to be the basis of learning. Experiments demonstrated that synaptic plasticity not only depends on excitatory activity but also on the rate and timing of inhibitory events. Hypothesising that the regulatory effect of inhibition is mediated by membrane potential hyperpolarisation, we identify fast fluctuation sensitivity as a contributing factor to the inhibitory regulation of plasticity. Fast fluctuation sensitivity characterises the influence of fast changes in the membrane potential on the plasticity model predictions, including (short) hyperpolarising events. Furthermore, by introducing a novel plasticity model, the Voltage-Dependent Pathway model, we show that fast fluctuation sensitivity enables a precise temporal regulation of plasticity via inhibition tightly locked to stimulus presentation. In recurrent networks, receptive fields develop independent of the degree of fluctuation sensitivity, yet the temporal precision of inhibitory regulation is critical for the heterogeneity and quality of the resulting representation.
Author Summary
Synaptic plasticity plays a central role in learning and enables us to interact within an ever-changing environment flexibly. The term governs all processes which leads to a change in connection strength between two neurons. Previously, studies of synaptic plasticity strongly focused on the role of excitatory activity. However, experiments excluding GABA blockers implicate inhibitory neurons to regulate synaptic plasticity. We further investigate the role of inhibitory neurons as regulators of synaptic plasticity in a model study. Specifically, we focus on the decrease in membrane potential due to inhibitory activity as a key actor in plasticity regulation. This comparative modelling study considers two well-established models and a novel model which differ in the order in which filtering and thresholding is applied to the membrane potential. We find that while all models develop connectivity structures resembling receptive fields characteristics of the visual cortex, the quality and heterogeneity of the developed feature detectors is model-specific. Heterogeneity and representation specificity is promoted by a high degree of fast fluctuation sensitivity of the model enabling temporally precise inhibitory regulation of plasticity determined by the model’s filter and threshold.
