Redwood Neuroscience
Title: “Possible Functional Purpose of Spike-Timing
and Frequency Dependence of Neocortical Synapticplasticity”
Michael
Eisele
Abstract:
What is the computational function of neocortical synaptic plasticity? Some of its features resemble a Hebbian learning rule, but the function of other features, like the dependence on spike timing, is less clear. Here we show that many of these features could be useful for a quite basic task, which we call “mending”. This task does not consist of creating new synaptic connections between neurons, as in associative, Hebbian learning, but in maintaining and repairing the existing ones. Any Hebbian learning rule can achieve some mending, but to fulfill this function well, the rule should also show other features, including a dependence on the relative timing between pre- and postsynaptic spike and also a dependence on the relative timing between postsynaptic spikes. To show this, we will introduce a simple, 5-parameter learning rule that agrees quantitatively with the known spike-timing dependence of neocortical synaptic plasticity in superficial layers of cat visual cortex and qualitatively with the frequency dependence that is known from other parts of the neocortex. We will demonstrate that this rule is better at mending than various other rules and derive the conditions under which it is close to being the optimal rule. This suggests that mending is one function of neocortical synaptic plasticity, but certainly not the only one.