Computing with the Integrate and Fire Neuron: Weber's Law, Multiplication and Phase Detection

Abstract

The integrate and fire model (Stein, 1967) provides an analytically tractable formalism of neuronal firing rate in terms of a neuron's membrane time constant, threshold and refractory period. Integrate and fire (IAF) neurons have mainly been used to model physiologically realistic spike trains but little application of the IAF model appears to have been made in an explicitly computational context. In this paper we show that the transfer function of an IAF neuron provides, over a wide parameter range, a compressive nonlinearity sufficiently close to that of the logarithm so that IAF neurons can be used to multiply neural signals by mere addition of their outputs. Thus, although the IAF transfer function is not explicitly logarithmic, its compressive parameter regime supports a simple, single neuron model for multiplication. A simulation of the IAF multiplier shows that under a wide choice of parameters, the IAF neuron can multiply its inputs to within a 5% relative error. We also show that an IAF neuron under a different, yet biologically reasonable, parameter regime can have a quasi-linear transfer function, acting as an adder or a gain node. We then show an application in which the compressive transfer function of the IAF model provides a simple mechanism for phase-detection: multiplication of 40Hz phasic inputs followed by low-pass filtering yields an output that is a quasi-linear function of the relative phase of the inputs. This is a neural version of the heterodyne phase detection principle. Finally, we briefly discuss the precision and dynamic range of an IAF multiplier that is restricted to reasonable firing rates (in the range of 10-300 Hz) and reasonable computation time (in the range of 25-200 milliseconds).