The research in this area is dedicated to development of an efficient estimator of entropy in high dimensions. The concept is to include manifold detection in the algorithm such that the k-nearest neighbour entropy estimator is efficient in high dimensions.

kNN

Depiction of k-nearest neighbours and associated epsilon-balls.

Performance of the estiamtor

EE_Gamma3D

Performance of the entropy estimator for varying N, k, and p parameters on a 3D-Gamma distribution. N: number of samples; k: number of neighbours; and p: length scale parameter (integer)