Efficient multivariate entropy estimation via \(k\)-nearest neighbour distances (Q1731757): Difference between revisions

The main object of analysis of this paper is a generalization of the so-called Kozachenko-Leonenko (K,L) estimator of entropy \(H\) [\textit{L. F. Kozachenko} and \textit{N. N. Leonenko}, Probl. Inf. Transm. 23, No. 1--2, 95--101 (1987; Zbl 0633.62005); translation from Probl. Peredachi Inf. 23, No. 2, 9--16 (1987)]: \[ H_n=\frac{1}{n}\sum^n_{i=1} \log\left(\rho^d_{(k)}\frac{V_d(n-1)}{e^{\Psi(k)}}\right), \] where \(V_d:=\pi^{d/2}/\Gamma(1+d/2)\) denotes the volume of the \(d\)-dimensional Euclidean unit ball and \(\Psi\) denotes the digamma function. This generalization is formed as a weighted average of the (K,L) estimator for different values of \(k\), where the weights are chosen to try to cancel the dominant bias terms.