Information geometry and its applications (Q908098): Difference between revisions

The book is an introduction to information geometry and its applications. We learn from the preface of this book that the term ``information geometry'' refers to a way of exploring the world of information by means of geometry. This field emerged from the study of the invariant geometrical structures inherent in statistical inference. Technically speaking, one has a Riemannian metric together with dually coupled affine connections in a manifold of probability distributions. The fields of applications of information geometry include, besides statistical inference, mathematics, physics, neurosciences and several areas of information sciences, like machine learning processing and optimization. The book is divided into 4 parts. In part I begins with the definition of a divergence between two points of a manifold. This is measure of the degree of separation between the two points. This leads to the notion of dually coupled asymmetry, and from here to the dual world. The manifold is then equipped with a dually flat structure equipped with a Riemannian metric. A generalized Pythagorean theorem in a dually flat information manifold is given. The dually flat structure is a special case of a Riemannian geometry equipped with non-flat dual affine connections, studied in Part II. Part II is a gentle introduction to dual differential geometry. On Part III the author addresses the field of statistical inference and he presents various ideas and results on this subject. Part IV is an overview of applications of information geometry in several fields, including machine learning and signal processing. The author starts with a question of pattern recognition, namely, obtaining a representative of a cluster of patterns, called the center of the cluster. Another question addressed is that of dividing a non-labeled aggregate of patterns into a set of clusters. In the case where the patterns are generated randomly subject to a probability distribution depending on each class, stochastic versions of the above problems is addressed. Information geometry deals with these problems in terms of divergence. The book is original and interesting.