Neural models are modeled at a variety of grains of resolution,
from the simpler McCulloch-Pitts neurons, the leaky integrator model
in which firing rate is a simple nonlinear function of membrane
potential (each neuron is modeled as a single compartment), to multiple
compartmental models that incorporate active membranes and a variety
of channels. Such models may or may not incorporate mechanisms of
synaptic plasticity. In addition, some modelers use mechanisms of
synaptic weight adjustment which are not intended to represent learning
but are simply identification procedures - i.e., given an overall
"architecture" (connectivity pattern) for a neural network,
these procedures seek settings of weights (and, possibly, other
parameters) which allow the architecture to approximate desired
One simulation system developed to support modeling and simulation
of general purpose neural networks is NSL
- Neural Simulation Language.
Multi-Level Neural Networks
A current research thrust is the extension of large neural networks
and behavioral systems to include modeling of detailed neurons.
This extension focuses in the integration of NSL with subneural
modeling tools, such as GENESIS (Wilson and Bower 1989), and NEURON
(Hines 1994). This task is part of the research project entitled
Multi-Level Modeling in Neural
Networks: A Computational and Experimental Approach, which integrates
ASL and NSL, besides linking to neurobiological data base systems.