Dynamic estimation of causal influences in sparsely-interacting neuronal ensembles

In this paper, we consider a neuronal ensemble under spontaneous activity where each neuron modulates the activity of the others through its spiking history. Assuming that the cross-history dependence parameters of the ensemble are sparse and time-varying, we perform adaptive system identification using sparse point process filters. We then provide a novel filtering and smoothing algorithm for estimating the Granger causality with high temporal resolution and with recursively computed statistical confidence intervals. We provide simulation studies which reveal significant performance gains obtained by our proposed technique in describing the causal influences in neuronal ensemble activity.