Statistical manifolds, self-parallel curves and learning processes
The paper treats statistical manifolds and \(\alpha\)-Chentsov-Amari connections. The case of \((\pm 1)\)-flat manifolds is considered and some interesting properties are given. Self-parallel curves are defined and their role in the minimization of the Kullback-Leibler relative entropy is studied. These results are applied to the learning of Boltzmann machines. The authors introduce some formalism of information geometry, a new domain relating differential geometry to probability theory. Analysis and examination of the structure of some special spaces called statistical manifolds is given. For the new geometry, covariant properties are introduced. For that the new formalism, the so-called \(\alpha\)-geometry, provides a natural interpretation for the learning process, the learning rules are discussed. The Boltzmann machine is studied using the previously described analysis of the manifold. The second section treats the construction of statistical manifolds and the Chentsov-Amari connections are introduced.~ Section 3 is devoted to the important case of (+1)-flat manifolds and some interesting properties are given. In Sect. 4 self-parallel curves are defined. The role of (+1)-self-parallel curves in the minimization of the Kullback-Leibler relative entropy is studied in Sect. 5. Finally in Section 6 these results are applied to the learning of the Boltzmann machine. For the entire collection see [Zbl 0914.00071].
This page was built for publication: Statistical manifolds, self-parallel curves and learning processes
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4263609)
