68 — Point Neurons with Conductance-Based Synapses in the Neural Engineering Framework

The Neural Engineering Framework, or NEF, is a simplified neural simulation system to simulate spiking neuron models (as opposed to continuous-value “neurons” common in computational neural networks). This pseudo-biofidelic approximates neurons using a point-neuron model, which, although flawed, more closely simulates neural activity than a CNN, for example.

This paper contributes a conductance-based synapse model to the NEF, which enables more biologically-plausible neural networks. This system is “modelled (sic) by conductance-based synapse models”, rather than neuron models; this means that in the conventional system, synapses are represented as a linear function, which is not quite biofidelic.

In other words, the current model simulates all synapses as a direct injection of current into its membrane; this is obviously not how synapses actually work (voltage is a gradual, continuous function of ion flow).

The current system is described as

where $J$ is the incident synaptic current, $g$ is membrane resistance, $E$ is driving voltage, and $C$ is the membrane capacitance in a leaky integrate-and-fire neuron.

This paper demonstrates that while a new conductance-based synapse model can account for most feed-forward dynamic models, more complex structures, such a integrators, begin to deviate from the expected, simple formulaic model.