David Hübne: The role of pairwise and higher-order correlations in feedforward inputs to neural networks

When presented with a task or stimulus, the ongoing activity in the brain is perturbed in order to process the new information of the environment. Typical characteristics of this evoked activity are (1) an increase in firing rate of neurons, (2) a decrease in trial-by-trial variability and (3) an increase or decrease in spike count correlations. Considering the importance of variability and correlations within the rate coding paradigm, it is crucial to understand the origin of these modulations. Different networks in the brain are typically connected through divergent-convergent connections. In a recent work, the correlations in the convergent connections of the feed-forward input have been found to be able to reproduce the above characteristics. We expand this work by also considering correlations in the divergent connections. Through large-scale network simulations, we can show that correlations in the divergent connections have a significant impact on the output correlation coefficient and a small impact on the output firing rate. Moreover, we 
investigate the question whether keeping the pairwise correlations constant and varying the higher-order correlation structure can influence the network dynamics. We find that this influence is very small suggesting that higher-order correlations in the divergence connections carry only a limited amount of information. These findings can significantly simplify the analysis of neural data.