Methods of Information Geometry by Shun-Ichi Amari, Hiroshi Nagaoka

By Shun-Ichi Amari, Hiroshi Nagaoka

Details geometry presents the mathematical sciences with a brand new framework of study. It has emerged from the research of the common differential geometric constitution on manifolds of likelihood distributions, which is composed of a Riemannian metric outlined via the Fisher info and a one-parameter family members of affine connections referred to as the $\alpha$-connections. The duality among the $\alpha$-connection and the $(-\alpha)$-connection including the metric play a necessary position during this geometry. this sort of duality, having emerged from manifolds of chance distributions, is ubiquitous, showing in various difficulties which would haven't any specific relation to chance conception. in the course of the duality, it really is attainable to research a variety of basic difficulties in a unified viewpoint.

The first 1/2 this booklet is dedicated to a complete creation to the mathematical origin of knowledge geometry, together with preliminaries from differential geometry, the geometry of manifolds or likelihood distributions, and the overall idea of twin affine connections. the second one 1/2 the textual content offers an outline of huge components of purposes, corresponding to records, linear platforms, details idea, quantum mechanics, convex research, neural networks, and affine differential geometry.

Show description

By Shun-Ichi Amari, Hiroshi Nagaoka

Details geometry presents the mathematical sciences with a brand new framework of study. It has emerged from the research of the common differential geometric constitution on manifolds of likelihood distributions, which is composed of a Riemannian metric outlined via the Fisher info and a one-parameter family members of affine connections referred to as the $\alpha$-connections. The duality among the $\alpha$-connection and the $(-\alpha)$-connection including the metric play a necessary position during this geometry. this sort of duality, having emerged from manifolds of chance distributions, is ubiquitous, showing in various difficulties which would haven't any specific relation to chance conception. in the course of the duality, it really is attainable to research a variety of basic difficulties in a unified viewpoint.

The first 1/2 this booklet is dedicated to a complete creation to the mathematical origin of knowledge geometry, together with preliminaries from differential geometry, the geometry of manifolds or likelihood distributions, and the overall idea of twin affine connections. the second one 1/2 the textual content offers an outline of huge components of purposes, corresponding to records, linear platforms, details idea, quantum mechanics, convex research, neural networks, and affine differential geometry.

Show description

Read or Download Methods of Information Geometry PDF

Best geometry and topology books

Real Methods in Complex and CR Geometry: Lectures given at the C.I.M.E. Summer School held in Martina Franca, Italy, June 30 - July 6, 2002

The geometry of actual submanifolds in complicated manifolds and the research in their mappings belong to the main complex streams of up to date arithmetic. during this zone converge the concepts of assorted and complicated mathematical fields resembling P. D. E. 's, boundary worth difficulties, triggered equations, analytic discs in symplectic areas, complicated dynamics.

Designing fair curves and surfaces: shape quality in geometric modeling and computer-aided design

This state of the art examine of the innovations used for designing curves and surfaces for computer-aided layout purposes makes a speciality of the primary that reasonable shapes are consistently freed from unessential gains and are easy in layout. The authors outline equity mathematically, reveal how newly constructed curve and floor schemes warrantly equity, and support the consumer in deciding on and elimination form aberrations in a floor version with no destroying the valuable form features of the version.

Extra resources for Methods of Information Geometry

Sample text

M = P (Q(F )\{0}), image into P (F ) of the isotropic cone minus its origin of F , is m-dimensional and called the compactification of V = Er,s . M is identical to the homogeneous space PO(F )/Sim(V ),41 quotient group of P O(F ) = O(r + 1, s + 1)/Z2 by the group SimV of similarities of V . 40 Directly for C , there are four basis elements 1, e , e , and e e with e2 = 1, e2 = 11 1 2 1 2 1 2 −1, e1 e2 = −e2 e1 , (e1 e2 )2 = 1, (e1 e2 )e1 = −e2 , (e1 e2 )e2 = −e1 . If we map 1 → 10 01 0 −1 1 0 , e1 → , e2 → , and then e1 e2 → , we get an algebra 01 10 1 0 0 −1 isomorphism between C11 and m(2, R).

Let {e1 , . . , er , er+1 , . . , er+s } be the standard orthonormal basis of V and {e0 , en+1 } be a basis of H such that for any x in H, x = x 0 e0 + x n+1 en+1 , (x|x) = (x 0 )2 − (x n+1 )2 . The equation of the cone Q(F ) is the following n+1 r x = (x 0 , x 1 , . . , x n+1 ) ∈ Q(F ) if and only if i=0 The Euclidean sphere of radius the following equation: (x l )2 = 0. (x i )2 − l=r+1 √ 2 associated with the basis {e0 , . . , en+1 } of F has r n+1 (x i )2 + i=0 (x l )2 = 2. l=r+1 x belongs to the intersection of Q(F ) and of the sphere if and only if r n+1 (x i )2 = i=0 (x j )2 = 1, j =r+1 that is, if and only if x belongs to the product of the unit sphere S r of the standard Euclidean space Er+1 , with the basis {e0 , .

By definition, (q) = (−1) n(n−1) 2 n ai (mod(K × )2 ) i=1 is the “discriminant” of q. The construction of a Clifford algebra associated with a quadratic regular space (E, q) is based on the fundamental idea of taking the square root of a quadratic form, more precisely of writing q(x) as the square of a linear form ϕ on E such that for any x ∈ E, q(x) = (ϕ(x))2 . 2 Clifford Mappings Let A be any associative algebra with a unit element 1A . 1 Definition A Clifford mapping f from (E, q) into A is a linear mapping f such that for any x ∈ E, (f (x))2 = q(x)1A .

Download PDF sample

Rated 4.82 of 5 – based on 13 votes