Theory of enhanced-by-coincidence neural information transmission

The activity of neurons within brain circuits has been ubiquitously reported to be correlated. The impact of these correlations on brain function has been extensively investigated. Correlations can in principle increase or decrease the information that neural populations carry about sensory stimuli, but experiments in cortical areas have mostly reported information-limiting correlations, which decrease the information encoded in the population. However, a second stream of evidence suggests that temporal correlations between the spiking activity of different neurons may increase the impact of neural activity downstream, implying that temporal correlations affect both the encoding of information and its downstream readout. The principle of how encoding and readout combine are still unclear. Here, we consider a model of transmission of stimulus information encoded in presynaptic input spike trains with information-limiting time-correlations to the output firing of a postsynaptic biologically plausible leaky integrate and fire (LIF) readout neuron. We derive an analytical solution of the model in the diffusion approximation, in which the encoding spiking activity is treated as a continuous-time stochastic variable. An ansatz based on a separation of timescales allows us compute the stimulus information transmitted to the readout over a broad range of parameters. Our analytical results reveal that, for sufficiently low input firing rates, large enough difference in input stimulus-specific activity, and moderately large input temporal correlations, the stimulus discriminability of the firing of the LIF readout neuron can be enhanced by the presence of input time correlations, despite their decreasing the stimulus information encoded in its inputs.