Comparision between spike and rate models in networks of integrate-and-fire neurons
F. Chapeau-Blondeau
in Biophysical Neural Networks: Foundations of Integrative Neuroscience
Chap. 11, R. R. Poznanski, ed., Mary Ann Liebert Inc. (New York) 2001.

Abstract.
A neural network model is developed, based on a description of neuron membrane conductances to govern the dynamics of action potentials. The variables are endowed with realistic physical units and plausible numerical values in order to reach both qualitative and quantitative significance. At the same time, the model is kept simple enough to enable its application to the study of collective behaviors among large populations of interacting neurons. This model is used here to investigate the relations between spike and rate representations of neural activities. First, the conditions for the derivation of a neuron transfer function when the spikes are replaced by firing rates are examined, and it is shown that such a transfer function on rates is usually not invariant for a given neuron, but it may vary with the statistics of the inputs. Next, a complete reduction of the network model operating with spikes to a model operating with firing rates is carried out. The spike and the rate models are then explicitly compared for the description of various neural phenomena. These phenomena include self-sustained network dynamics, the locking of a neural chain into an oscillatory activity, and noise-enhancement of neural signal transmission via stochastic resonance. In all cases, the comparison is done in stationary states of the networks, and it examines neuron firing rates either as they are directly produced by the rate model or as they are computed by explicit time averaging on spike trains in the spike model. Dissimilarities between the descriptions based on spikes or rates are especially pointed out. The present study contributes to a better appreciation of the possibilities of spike and firing rate models, and more generally, of different modeling strategies existing for neural networks.

back to the List of Publications