Page - 419 - in Differential Geometrical Theory of Statistics
Image of the Page - 419 -
Text of the Page - 419 -
Entropy2016,18, 425
Let M be themanifold M = {(p11, . . . ,pd1, . . . ,p1N, . . . ,pdN)|(p1i , . . . ,pdi ) ∈ Rd}. The LDDMM
metric on the landmarkmanifoldM is directly related to thekernelKwhenwrittenas a cometric
gp(ξ,η)=∑Ni,j=1ξ iK(pi,pj)ηj. Letting ikdenote the indexof thekthcomponentof the ith landmark,
the cometric is in coordinates gi kjl
p =K(pi,pj)lk. TheChristoffel symbols canbewritten in termsof
derivativesof thecometricgij [31] (recall thatδij= g ikgkj= gjkgki)
Γkij= 1
2 gir (
gklgrs,l−gslgrk,l−grlgks,l )
gsj . (18)
This relation comes from the fact that gjm,k =−gjrgrs,kgsm gives the derivative of themetric.
Thederivativesof thecometric is simplygi kjl
,rq =(δ i
r+δ j
r)∂pqr K(pi,pj)lk. Using(18),derivativesof the
Christoffel symbolscanbecomputed
Γkij,ξ = 1
2 gir,ξ (
gklgrs,l−gslgrk,l−grlgks,l )
gsj+ 1
2 gir (
gklgrs,l−gslgrk,l−grlgks,l )
gsj,ξ
+ 1
2 gir (
gkl,ξg rs
,l+g klgrs,lξ−gsl,ξgrk,l−gslgrk,lξ−grl,ξgks,l−grlgks,lξ )
gsj .
Thisprovides the fulldata fornumerical integrationof theevolutionequationsonFkM.
InFigure 7 (top row),weplotminimizingnormalMPPson the landmarkmanifoldwith two
landmarksandvaryingcovariance in theR2 horizontal andverticaldirection. Theplot shows the
landmarkequivalentof theexperiment inFigure5.Notehowaddingcovariance in thehorizontaland
verticaldirection, respectively,allowstheminimizingnormalMPPtovarymore in thesedirections
because theanisotropically-weightedmetricpenalizeshigh-covariancedirections less.
Figure 7 (bottom row) showsfive curves satisfying theMPPequationswith varyingvertical
V∗FM initialmomentumsimilarly to theplots inFigure6. Again,weseehowtheextradegreesof
freedomallows thepaths to twist, generatingahigher-dimensional family thanclassicalgeodesics
withrespect toC.
(a) (b) (c) (d) (e)
(f) (g) (h) (i) (j)
Figure7. (Toprow)Matchingof twolandmarks (green) to twolandmarks (red)by (a) computinga
minimizingRiemanniangeodesiconthe landmarkmanifold,and(b–e)minimizingMPPswithadded
covariance (arrows) inR2 horizontaldirection(b,c) andvertical (d,e). Theactionof thecorresponding
diffeomorphismsonaregulargrid isvisualizedbythedeformedgridwhich iscoloredbythewarp
strain. Theaddedcovarianceallowsthepaths tohavemoremovement in thehorizontalandvertical
direction,respectively,becausetheanisotropicallyweightedmetricpenalizeshigh-covariancedirections
less. (bottomrow, (f–j))Five landmarktrajectorieswithfixedinitialvelocityandanisotropiccovariance
butvaryingV∗FMvertical initialmomentumξ0. Changingtheverticalmomentum“twists” thepaths.
419
Differential Geometrical Theory of Statistics
- Title
- Differential Geometrical Theory of Statistics
- Authors
- Frédéric Barbaresco
- Frank Nielsen
- Editor
- MDPI
- Location
- Basel
- Date
- 2017
- Language
- English
- License
- CC BY-NC-ND 4.0
- ISBN
- 978-3-03842-425-3
- Size
- 17.0 x 24.4 cm
- Pages
- 476
- Keywords
- Entropy, Coding Theory, Maximum entropy, Information geometry, Computational Information Geometry, Hessian Geometry, Divergence Geometry, Information topology, Cohomology, Shape Space, Statistical physics, Thermodynamics
- Categories
- Naturwissenschaften Physik