Spike trains from individual neurons are highly variable in response to repeated stimulus. The shared variability among neurons reflects the underlying circuitry and the dynamical state of the network. By studying neural variability in network models, we identify circuit features that shape the dimensional structure of neuronal responses. Recently, we developed a spatially extended network of spiking neuron models, which can internally generate shared variability matching the low dimensional structure widely reported across cortex. We show that when the inhibitory projection width is not broader than the excitatory projections, the turbulent spiking dynamics can propagate across the entire network, giving rise to global fluctuations.
In addition, we are interested in how neural population responses are modulated by behavior contexts. We found that a top-down depolarization of inhibitory neurons in our network recapitulates several properties of attentional modulation on neural population responses. Further, we study the implication of neuronal variability on neural coding in spiking neuron networks. By combining information theory and network modeling, we investigate the change of information flow under different contexts.