In the framework of information geometry, we explore the structure of statistical manifold Sn related to the Rényi entropy. Starting from a new family of generalized exponential distributions given by the equilibrium distribution of a canonical system described by the Rényi entropy, we introduce on Sn a dually-flat geometry endowed by a Hessian structure derived from a generalization of Fisher metric and a flat affine connection. This geometrical structure admits a Legendre transform that links the thermodynamic potentials of the underlying statistical mechanics to the dual potentials of the corresponding information geometry. A canonical divergence function à la Bregman is naturally derived in this framework. Further geometric structures, derived from other contrast functions strictly related to the Rényi entropy are explored and their link with the ?*-geometry of Amari is highlighted.
A study of Rényi entropy based on the information geometry formalism
Antonio M Scarfone;
2020
Abstract
In the framework of information geometry, we explore the structure of statistical manifold Sn related to the Rényi entropy. Starting from a new family of generalized exponential distributions given by the equilibrium distribution of a canonical system described by the Rényi entropy, we introduce on Sn a dually-flat geometry endowed by a Hessian structure derived from a generalization of Fisher metric and a flat affine connection. This geometrical structure admits a Legendre transform that links the thermodynamic potentials of the underlying statistical mechanics to the dual potentials of the corresponding information geometry. A canonical divergence function à la Bregman is naturally derived in this framework. Further geometric structures, derived from other contrast functions strictly related to the Rényi entropy are explored and their link with the ?*-geometry of Amari is highlighted.| File | Dimensione | Formato | |
|---|---|---|---|
|
prod_418496-doc_147799.pdf
solo utenti autorizzati
Descrizione: A study of Rényi entropy based on the information geometry formalism
Tipologia:
Documento in Post-print
Licenza:
NON PUBBLICO - Accesso privato/ristretto
Dimensione
227.49 kB
Formato
Adobe PDF
|
227.49 kB | Adobe PDF | Visualizza/Apri Richiedi una copia |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


