Seite - 416 - in Differential Geometrical Theory of Statistics
Bild der Seite - 416 -
Text der Seite - 416 -
Entropy2016,18, 425
Note that the introductionof theRiemannianmetricgR implies thatWij arenowdependenton
themanifoldcoordinatesxi.
6.NumericalExperiments
Weaimat visualizingmost probable paths for thedrivingprocess andprojections of curves
satisfying the MPP Equation (13) in two cases: On 2D surfaces embedded inR3 and on finite
dimensional landmarkmanifoldsthatarise fromequippingasubsetof thediffeomorphismgroupwith
aright-invariantmetricandlettingtheactiondescendto the landmarksbya leftaction. Thesurface
examples are implemented in Python using the Theano [30] framework for symbolic operations,
automatic differentiation, andnumerical evaluation. The landmark equations are detailed below
and implemented inNumpyusingNumpy’s standardODEintegrators. Thecode for running the
experiments isavailableathttp://bitbucket.com/stefansommer/mpps/.
6.1. EmbeddedSurfaces
WevisualizenormalMPPsandprojectionsofcurvessatisfyingtheMPPEquation(13)onsurfaces
embeddedinR3 inthreecases: ThesphereS2,onanellipsoid,andonahyperbolicsurface. Thesurfaces
are chosen in order to have bothpositive andnegative curvature, and to have varyingdegree of
symmetry. In all cases, an open subset of the surfaces are represented in a single chart by amap
F : R2 → R3. For the sphere and ellipsoid, this gives a representation of the surface, except for
thesouthpole. ThemetricandChristoffel symbolsarecalculatedusingthesymbolicdifferentiation
featuresofTheano. The integrationareperformedbyasimpleEuler integrator.
Figures4–6showfamiliesofcurvessatisfyingtheMPPequations in threecases: (1)Withfixed
startingpointx0∈Mandinitialvelocity x˙0∈TMbutvaryinganisotropyrepresentedbychanging
frameu in thefiberabove x0; (2)minimizingnormalMPPswithfixedstartingpoint andendpoint
x0,x1∈Mbutchanging frameuabovex0; (3)fixedstartingpointx0∈Mandframeubutvarying
V∗FMverticalpartof the initialmomentum ξ0∈T∗FM. Thefirst andsecondcases thusshowthe
effectofvaryinganisotropy,while the thirdcase illustrates theeffectof the“twist” that thed2 degrees
in theverticalmomentumallows.Note thedisplayedanti-developedcurves inR2 that forclassicalC
geodesicswouldalwaysbestraight lines.
(a) (b) (c)
Figure4.CurvessatisfyingtheMPPequations (toprow)andcorrespondinganti-development (bottom
row) on three surfaces embedded inR3: (a) An ellipsoid; (b) a sphere; (c) a hyperbolic surface.
The familyofcurves isgeneratedbyrotatingbyπ/2radians theanisotropiccovariancerepresented in
the initial frameu0 anddisplayedin thegrayellipse.
416
Differential Geometrical Theory of Statistics
- Titel
- Differential Geometrical Theory of Statistics
- Autoren
- Frédéric Barbaresco
- Frank Nielsen
- Herausgeber
- MDPI
- Ort
- Basel
- Datum
- 2017
- Sprache
- englisch
- Lizenz
- CC BY-NC-ND 4.0
- ISBN
- 978-3-03842-425-3
- Abmessungen
- 17.0 x 24.4 cm
- Seiten
- 476
- Schlagwörter
- Entropy, Coding Theory, Maximum entropy, Information geometry, Computational Information Geometry, Hessian Geometry, Divergence Geometry, Information topology, Cohomology, Shape Space, Statistical physics, Thermodynamics
- Kategorien
- Naturwissenschaften Physik