Sample-Spacings-Based Density and Entropy Estimators for Spherically Invariant Multidimensional Data
Intae Lee · Neural Computation · 2010
While the sample-spacings-based density estimation method is simple and efficient, its applicability has been restricted to one-dimensional data. In this letter, the method is generalized such that it can be extended to multiple dimensions in certain circumstances. As a consequence, a multidimensional entropy estimator of spherically invariant continuous random variables is derived. Partial bias of the estimator is analyzed, and the estimator is further used to derive a nonparametric objective function for frequency-domain independent component analysis. The robustness and the effectiveness of the objective function are demonstrated with simulation results.