An information theoretic approach to identify the role of higher-order interactions between cortical neurons in stimulus coding

The massive ¿wiring¿ observed between cortical neurons suggests that these neurons are interacting in encoding external covariates. We propose an information theoretic approach to determine how this interaction may provide an optimal encoding strategy. Using a biologically plausible statistical learning model, we compared the performance of the proposed approach with an independent and a maximum entropy model in capturing the information in a motor task using a subset of neurons drawn randomly from a larger population encoding the task. We demonstrate that a substantial amount of information about the task is encoded in second order interactions, confirming in vitro experimental results using maximum entropy models. Additionally, a considerable amount of information was captured by third order interactions, suggesting that higher order interaction may be playing a larger role in cortical information processing in vivo. We believe that this framework may be useful for improving real-time decoding performance in Brain Machine Interfaces.