Parameter Values and Fatigue Mechanisms for FLIF Neurons

A typical human brain consists of roughly 100 billion neurons, and one key aim of Biological Cybernetics is to simulate neural systems. A good model of a neuron accurately represents the behaviour of biological neurons, typically the spiking behaviour. For cybernetic systems that aim to function in real time with thousands, millions, or even billions of simulated neurons, it is also important that the model is computationally efficient. One Fatiguing Leaky Integrate and Fire neuron is a model that has four free parameters per neuron. This model has been used in cybernetic agents, but there have been few links to actual biological behaviour. A model of a rat neocortical neuron is developed with four specific parameter settings. This model is tuned to a particular input regime. When compared to a biological neuron it gets 90% of spikes roughly correct. Further modifications of the fatigue model enables the FLIF neuron to account for spontaneous neural firing, a known neural property, that is not present in the data. These modifications provide other FLIF models with a similar fit to the biological data. The best of these models correctly predicts over 94% of the spikes.