Measurements of generalisation based on information geometry

Abstract

Neural networks are statistical models and learning rules are estimators. In this paper a theory for measuring generalisation is developed by combining Bayesian decision theory with information geometry. The performance of an estimator is measured by the information divergence between the true distribution and the estimate, averaged over the Bayesian posterior. This unifies the majority of error measures currently in use. The optimal estimators also reveal some intricate interrelationships among information geometry, Banach spaces and sufficient statistics.

Divisions: Aston University (General)
Additional Information: Copyright of SpringerLink
Uncontrolled Keywords: neural networks,Bayesian,information geometry,estimator,error,Banach
Publication ISSN: 1573-7470
Last Modified: 30 Oct 2024 08:04
Date Deposited: 14 Jul 2009 11:11
Full Text Link:
Related URLs: http://www.spri ... l/journal/10472 (Publisher URL)
PURE Output Type: Article
Published Date: 1995-07-02
Authors: Zhu, Huaiyu
Rohwer, Richard

Download

[img]

Version: Published Version


Export / Share Citation


Statistics

Additional statistics for this record