3,384 research outputs found

    The exponential statistical manifold: mean parameters, orthogonality and space transformations.

    No full text
    Let (X,CalX,mu)(X, Cal X, mu) be a measure space, and let CalM(X,CalX,mu)Cal M(X,Cal X,mu) denote the set of the mumu-almost surely strictly positive probability densities. It was shown by G. Pistone and C. Sempi (1995) that the global geometry on CalM(X,CalX,mu)Cal M(X,Cal X,mu) can be realized by an affine atlas whose charts are defined locally by the mappings CalM(X,CalX,mu)supsetCalUpiqmapstolog(q/p)+K(p,q)inBpCal M(X,Cal X,mu)supset Cal U_p i q mapsto log(q/p) + K(p,q)in B_p, where CalUpCal U_p is a suitable open set containing pp, K(p,q)K(p,q) is the Kullback-Leibler relative information and BpB_p is the vector space of centered and exponentially (pcdotmu)(pcdotmu)-integrable random variables. In the present paper we study the transformation of such an atlas and the related manifold structure under basic transformations, that is measurable transformation of the sample space. A generalization of the mixed parameterization method for exponential models is also presented

    Analytical and geometrical properties of statistical connections in Information Geometry

    No full text
    Information Geometry is a field where one can measure the deep impact of geometry and analysis in statistics, information theory and related applied fields. The present contribution has the goal of showing also the impact that statistics and information theory can have in geometry and analysis. Indeed it is clear that the development of the non-parametric and non-commutative versions of Information Geometry need a massive use of mathematical instruments of infinite-dimensional analysis, geometry and operator theory. On the other side there is an increasing interest of mathematicians for the non-trivial problems that are suggested by the application of Information Geometry. Even in the elementary case of a finite state space, the geometrical approach adds considerable insight to the modeling of applied problems
    corecore