Intrinsic Transfer Entropy

Quantifying information flow within a system is paramount to understanding its behavior. One common, though flawed, method of doing this is via the \emph{transfer entropy}. The transfer entropy is a particular form of conditional mutual information, which captures both \emph{intrinsic dependence} between variables as well as \emph{conditional dependence}. Here, we propose a new method of quantifying information flow, the \emph{intrinsic transfer entropy}. Rather than utilizing the conditional mutual information, intrinsic transfer entropy uses the \emph{intrinsic mutual information} from information-theoretic cryptography. This provides for the first time a concrete method of separately quantifying intrinsic information flow from conditional information flow. We apply this measure to a variety of systems to demonstrate its usefulness.