A Regularization Approach to Fractional Dimension Estimation
Abstract
We propose a new way of evaluating the regularity of a graph of a function f. Our approach is based on measuring the growth rate of the lengths of less and less regularized versions of f. This leads to a new index, that we call regularization dimension, dim_R . We derive some analytical properties of dim_R and compare it with other fractional dimensions. A statistical estimator is derived, and numerical experiments are performed, which suggest that dim_R may be computed in a robust way. Finally, we apply the regularization dimension to the study of Ethernet traffic.
Domains
Probability [math.PR]
Fichier principal
A_Regularization_Approach_to_Fractional_Dimension_Estimation.pdf (7.43 Mo)
Télécharger le fichier
Origin : Files produced by the author(s)