Decorrelation of Neural-Network Activity by Inhibitory Feedback

Correlations in spike-train ensembles can seriously impair the encoding ofinformation by their spatio-temporal structure. An inevitable source ofcorrelation in finite neural networks is common presynaptic input to pairs ofneurons. Recent studies demonstrate that spike correlations in recurrent neuralnetworks are considerably smaller than expected based on the amount of sharedpresynaptic input. Here, we explain this observation by means of a linearnetwork model and simulations of networks of leaky integrate-and-fire neurons.We show that inhibitory feedback efficiently suppresses pairwise correlationsand, hence, population-rate fluctuations, thereby assigning inhibitory neuronsthe new role of active decorrelation. We quantify this decorrelation bycomparing the responses of the intact recurrent network (feedback system) andsystems where the statistics of the feedback channel is perturbed (feedforwardsystem). Manipulations of the feedback statistics can lead to a significantincrease in the power and coherence of the population response. In particular,neglecting correlations within the ensemble of feedback channels or between theexternal stimulus and the feedback amplifies population-rate fluctuations byorders of magnitude. The fluctuation suppression in homogeneous inhibitorynetworks is explained by a negative feedback loop in the one-dimensionaldynamics of the compound activity. Similarly, a change of coordinates exposes aneffective negative feedback loop in the compound dynamics of stableexcitatory-inhibitory networks. The suppression of input correlations in finitenetworks is explained by the population averaged correlations in the linearnetwork model: In purely inhibitory networks, shared-input correlations arecanceled by negative spike-train correlations. In excitatory-inhibitorynetworks, spike-train correlations are typically positive. Here, the suppressionof input correlations is not a result of the mere existence of correlationsbetween excitatory (E) and inhibitory (I) neurons, but a consequence of aparticular structure of correlations among the three possible pairings (EE, EI,II).