A Family of Statistical Divergences Based on Quasiarithmetic Means
share
This paper proposes a generalization of Tsallis entropy and Tsallis relative entropy and discusses some of their properties. This generalization is defined with respect to a deformed exponential φ, which is the same used in defining φ-families of probability distributions, and generalizes important classes of distributions. The generalized relative entropy is then shown to be a φ-divergence with some conditions and related to a φ-family by a normalizing function. We determine necessary and sufficient conditions for the generalized relative entropy so it satisfies the partition inequality and jointly convexity. Further, such conditions are fulfilled we derive the conditions about the inverse of the deformed exponential φ which results from such assumption. We also show the Pinsker's inequality for the generalized relative entropy that provides a relationship between the generalized relative entropy and the statistical distance.
READ FULL TEXT