Collective properties of neural populations: beyond higher order correlations

ROMAN BARAVALLE; FERNANDO MONTANI

Conferencia; XX Conference on Nonequilibrium Statistical Mechanics and Nonlinear Physics; 2018

The dichotomized Gaussian model generates spike trains by thresholding a multivariateGaussian random variable. Importantly its spiking mechanism is similar to the one assumedin balanced networks, where the mean input to each neuron is smaller than the thresholdand thus spikes are induced by fluctuations in the input. Higher-order correlations can takeplace due to this threshold operation of common input to a population of neurons. Wepropose a precise quantification of the higher order correlations estimating all the ordersof these resulting interactions. We investigate the behavior of the critical fluctuations bymeans of the higher order cumulant as a function of the number of neurons. Our approachallows us to characterize the emergent properties of the system showing that the modelexhibits a second-order phase transition at the critical point.