A PROBABILISTIC MODEL OF NEURONAL INTEGRATION

A probabilistic estimation of an integrate and fire model commonly used for performing biologically plausible neurocomputing is proposed. It processes the temporal information of the input stimuli as a group of cells — the stellate or Chopper cells of the cochlear nucleus. The main properties are the same as the ones of the common Hodgkin–Huxley system. This model has a limited set of parameters: time constant of the membrane potential, threshold level, refractory durations. It is defined by a delay differential system having a low cost of computation. Original properties such as blocking and reset are introduced, providing a control of the oscillatory behaviour useful for building functional neural nets. In order to recover the parameters of neurons, we also analyze the first interspike interval histogram.

A TEST FOR A BIOLOGICAL SIGNAL ENCODED BY NOISE

We consider here a simple example of stimulated sensory neurons operating under the influence of their own internal noise: the hair mechanoreceptor of the crayfish stimulated by a weak, periodic, hydrodynamic signal. Action potential spike trains from the sensory neuron are recorded and assembled into two objects for analysis: the interspike interval histogram (ISIH) and the cycle histogram of the spike density. A time transformation is carried out on the ISIH’s in order to test the hypothesis that the spike train is basically random and that the probability of coherent spike generation is related to the instantaneous stimulus amplitude. Moreover it is shown that the physiological spike train data can be qualitatively mimicked by an electronic Fitzhugh-Nagumo model, operated in the subcritical mode, driven by noise and a weak periodic signal. A discussion of how the Fitzhugh-Nagumo model is properly operated to mimic noisy data from sensory neurons is included.