Non–Parametric Estimation of Mutual Information through the Entropy of the Linkage

摘要: A new, non–parametric and binless estimator for the mutual information of a d–dimensional random vector is proposed. First all, an equation that links to entropy suitable with uniformly distributed components deduced. When d = 2 this reduces well known connection between copula function associated original variables. Hence, problem estimating reduced estimation obtained through multidimensional transformation. The we propose two–step method: first estimate transformation obtain transformed sample, then its entropy. properties new are discussed simulation examples performances compared those best estimators in literature. precision converges values same order magnitude tested. However, unbiased even larger dimensions smaller sample sizes, while other tested show bias these cases.