+ - 0:00:00
Notes for current slide
Notes for next slide

On the geometric interpretation of MFPCA

and the usage of the Gram matrix

Steven Golovkine · Edward Gunning · Andrew J. Simpkin · Norma Bargary

54es Journée de Statistique de la SFDS

July 5th, 2023

1 / 16

Multivariate functional data

observation
2 / 16

Some notations

  • Observation space:

H=L2(T1)××L2(TP)P terms.

  • Inner product in H:

f,g=p=1PTpf(p)(tp)g(p)(tp)dtp.

  • For N realizations of a process X, we note

    • the mean function μ,

    • the covariance operator Γ, with covariance kernel C,

    • the Gram (inner-product) matrix M.

  • Each feature of each observation is sampled on a regular grid of Mp points.

3 / 16

Cloud of individuals


cloud_obs    cloud_obs_proj
4 / 16

Cloud of individuals

  • Let πn,n{1,,N} be a weight on each observation such that nπn=1.

  • Distance between observations

d2(Mf,Mg)=fg,fg,f,gH.

  • Inertia of the cloud CN using d

n=1Nπnd2(Mn,Gμ)=12n=1Nm=1Nπnπmd2(Mn,Mm)=p=1PTpVarX(p)(tp)dtp.

5 / 16

Cloud of individuals

  • Let πn,n{1,,N} be a weight on each observation such that nπn=1.

  • Distance between observations

d2(Mf,Mg)=fg,fg,f,gH.

  • Inertia of the cloud CN using d

n=1Nπnd2(Mn,Gμ)=12n=1Nm=1Nπnπmd2(Mn,Mm)=p=1PTpVarX(p)(tp)dtp.

  • Another distance between observations

dΓ2(Mf,Mg)=fg,Γ(fg),f,gH.

  • Inertia of the cloud CN using dΓ

n=1NπndΓ2(Mn,Gμ)=12n=1Nm=1NπnπmdΓ2(Mn,Mm)=p=1PTp|||Cp(tp,)|||2dtp.

5 / 16

Cloud of features

cloud_features    cloud_features_proj
6 / 16

Cloud of features

  • Distance between features

d2(Mf,Mg)=n=1NπnXnμ,fg2,f,gH.

  • Inertia of the cloud CP

n=1Nπnd2(Mn,Gμ)=12n=1Nm=1NπnπmdΓ2(Mn,Mm)=p=1PTp|||Cp(tp,)|||2dtp.

  • Correlation coefficient

cosθfg=n=1NπnXnμ,fXnμ,g(n=1NπnXnμ,f2)1/2(n=1NπnXnμ,g2)1/2=f,Γgf,Γfg,Γg.

7 / 16

Duality diagram

diagram
Duality diagram (extended from De la Cruz and Holmes (2011)).
8 / 16

MFPCA

  • Consider the matrix M of size N×N with entries

Mij=πiπjXiμ,Xjμ,i,j1,N.

  • Eigenvalues of Γ and M are related by

λk=lk,k=1,2,

  • Eigenvectors of Γ and M are related by

ϕk(t)=1Nlkn=1Nvnk{Xn(t)μ(t)},k=1,2,

  • Scores are given by

cnk=Nlkvnk,n=1,,N,k=1,2,

9 / 16

Computational complexity

  • Assume Ma=p=1PMpa and K=p=1PKp.

  • Using the diagonalization of the covariance operator (Happ and Greven (2018))

O(NM2+M3+Np=1PMpKpUnivariate covariancedecomposition+NK2+K3Univariate scoresdecomposition+Kp=1PMpKp+NK2Multivariate eigencomponentsand scores estimation).

  • Using the diagonalization of the inner product matrix

O(N2M1+N3Gram matrixdecomposition+KPN+KNMultivariate eigencomponentsand scores estimation).

  • Note that, here, the smoothing part is not considered into the computational complexity.
10 / 16

Simulation of multivariate functional data

comput_time_image_1
11 / 16

Simulation of multivariate functional data

reconst_error_image_1
12 / 16

Simulation of images data

comput_time_image
13 / 16

Simulation of images data

reconst_error_image
14 / 16

Takeaway ideas

  • We gave a geometric interpretation of the duality between rows and columns of a functional data matrix.

  • We provided relationships between the eigenelements of the covariance operator and the ones of the Gram matrix.

  • When to use the covariance operator?

    • Only one-dimensional curves.

    • For sparse to relatively dense functional data.

  • When to use the Gram matrix?

    • For two-dimensional (or higher dimensional) functional data (images).

    • For ultra-dense functional data.

  • The paper is available on arXiv: arXiv:2306.12949

Thank you for your attention!

15 / 16

References

De la Cruz, O. and S. Holmes (2011). “The Duality Diagram in Data Analysis: Examples of Modern Applications”. In: The annals of applied statistics 5.4, pp. 2266–2277. ISSN: 1932-6157.

Happ, C. and S. Greven (2018). “Multivariate Functional Principal Component Analysis for Data Observed on Different (Dimensional) Domains”. In: Journal of the American Statistical Association 113.522, pp. 649-659. DOI: 10.1080/01621459.2016.1273115.

16 / 16

Multivariate functional data

observation
2 / 16
Paused

Help

Keyboard shortcuts

, , Pg Up, k Go to previous slide
, , Pg Dn, Space, j Go to next slide
Home Go to first slide
End Go to last slide
Number + Return Go to specific slide
b / m / f Toggle blackout / mirrored / fullscreen mode
c Clone slideshow
p Toggle presenter mode
t Restart the presentation timer
?, h Toggle this help
Esc Back to slideshow