A rule-based firing model for neural networks

Full multi-compartment multi-channel neu- ron models are state of the art for single neuron modeling but are CPU intensive. This makes them unsuitable for network modeling, where simulation of 10,000 or more neurons is desirable. For this reason, most network mod- els utilize highly simplified models such as single state- variable integrate-and-fire units. This compromise has the disadvantage of eliminating most biological detail, much of which can be expected to lead to interesting and important network behavior. To reconcile these opposing computational and biological demands, we developed a rule-based firing (RBF) model incorporating a number of synaptic and cellular responses which are activated as needed. The rules produce effects that include adaptation, bursting, depolarization blockade, Mg-sensitive NMDA conductance, and post-inhibitory rebound. By utilizing pre-calculated waveforms and avoiding linked differential equations, network simulations are entirely event-driven, with no integration overhead. The model has been further optimized by use of table look-ups in lieu of run-time calculation.