Novel techniques for model-free and fast computation of mutual information

In this study, two new approaches are proposed to calculate mutual information between two random variables from data. These approaches are constructed in a way to use the properties of the differential entropy under linear transformations and to try to minimize conditional entropy in a model-free manner. In comparisons with a widely used mutual information estimator, the Kraskov method, the methods that we termed as unit vector parametrization and data fitting based estimators, offered an advantage in terms of computation speed.