arXiv subjects

Naomichi Nakajima

Publications and source records attributed to Naomichi Nakajima.

3 recordsLinked to original sources

arXiv2025

The space of positive transition measures on a Markov chain

Information geometry of Markov chains has been studied using the dually flat structure of the space of transition probabilities. Although applications of this structure have been investigated, few attempts have examined its statistical meaning. In this paper, we construct a foundation for investigating the statistical meaning based on Amari's theory of positive measures. For the space of discrete distributions, Amari has introduced the space of positive measures by removing the constraint condition and investigated the extended space by finding the Bregman and $F$-divergence suitably. According to this, we introduce an extension of the space of transition probabilities equipped with suitable $F$-divergence for a given Markov chain. We regard it as the space of positive transition measures on a Markov chain, and study its dually flat structure. This provides new insight into the geometry of Markov chains and may lead to the development of the theory of Markov embeddings.

math.DG↗

arXiv2022

Local normal forms of em-wavefronts in affine flat coordinates

In our previous work, we have generalized the notion of dually flat or Hessian manifold to quasi-Hessian manifold; it admits the Hessian metric to be degenerate but possesses a particular symmetric cubic tensor (generalized Amari-Centsov tensor). Indeed, it naturally appears as a singular model in information geometry and related fields. A quasi-Hessian manifold is locally accompanied with a possibly multi-valued potential and its dual, whose graphs are called the $e$-wavefront and the $m$-wavefront respectively, together with coherent tangent bundles endowed with flat connections. In the present paper, using those connections and the metric, we give coordinate-free criteria for detecting local diffeomorphic types of $e/m$-wavefronts, and then derive the local normal forms of those (dual) potential functions for the $e/m$-wavefronts in affine flat coordinates by means of Malgrange's division theorem. This is motivated by an early work of Ekeland on non-convex optimization and Saji-Umehara-Yamada's work on Riemannian geometry of wavefronts. Finally, we reveal a relation of our geometric criteria with information geometric quantities of statistical manifolds.

math.DG↗

arXiv2021

The dually flat structure for singular models

The dually flat structure introduced by Amari-Nagaoka is highlighted in information geometry and related fields. In practical applications, however, the underlying pseudo-Riemannian metric may often be degenerate, and such an excellent geometric structure is rarely defined on the entire space. To fix this trouble, in the present paper, we propose a novel generalization of the dually flat structure for a certain class of singular models from the viewpoint of Lagrange and Legendre singularity theory - we introduce a quasi-Hessian manifold endowed with a possibly degenerate metric and a particular symmetric cubic tensor, which exceeds the concept of statistical manifolds and is adapted to the theory of (weak) contrast functions. In particular, we establish Amari-Nagaoka's extended Pythagorean theorem and projection theorem in this general setup, and consequently, most of applications of these theorems are suitably justified even for such singular cases. This work is motivated by various interests with different backgrounds from Frobenius structure in mathematical physics to Deep Learning in data science.

math.DG↗