pith. sign in

arxiv: 2409.00817 · v5 · submitted 2024-09-01 · 📊 stat.ME · math.ST· stat.ML· stat.TH

Structural adaptation and rate accelerated estimation in bivariate functional data

Omar Kassi , Sunny G.W. Wang This is my paper

Pith reviewed 2026-05-23 20:54 UTC · model grok-4.3

classification 📊 stat.ME math.STstat.MLstat.TH
keywords directional regularityanisotropybivariate functional datachange-of-basissmoothing ratesreplication structurenon-asymptotic bounds
0
0 comments X

The pith

Adapting to directional regularity via change-of-basis yields faster smoothing rates for bivariate functional data.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper defines directional regularity to capture anisotropy in multivariate functional data by incorporating both dimensional smoothness and directional properties. It demonstrates that a data-driven change-of-basis aligned with this regularity produces faster convergence rates for smoothing estimators than standard approaches. The replication structure typical of functional data makes it possible to estimate and identify the required basis matrix, with non-asymptotic error bounds established for the procedure. The method is illustrated on simulated data and applied to a rainfall dataset.

Core claim

Directional regularity is introduced as a definition of anisotropy that augments conventional smoothness notions with directional information. For bivariate functional processes, a change-of-basis that aligns with this regularity produces accelerated rates of convergence for smoothing. An algorithm identifies the change-of-basis matrix from the replication structure of the data, and non-asymptotic bounds quantify the resulting estimation error.

What carries the argument

Directional regularity, a definition of anisotropy that incorporates directional properties in addition to dimensional smoothness, which enables an adaptive change-of-basis for rate improvement.

If this is right

  • Smoothing estimators achieve strictly faster convergence when the change-of-basis is adapted to directional regularity.
  • The algorithm produces consistent estimates of the basis matrix under the replication structure.
  • Non-asymptotic bounds hold uniformly for the adapted smoothing procedure.
  • The same adaptation applies to real bivariate functional datasets such as rainfall measurements.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The approach could be tested on trivariate or higher-dimensional functional data to check whether directional regularity continues to yield rate gains.
  • If the change-of-basis can be estimated from a single long trajectory rather than many replications, the method would apply to non-replicated functional time series.
  • The directional-regularity framework might interact with existing anisotropic kernel or spline methods to produce hybrid estimators.

Load-bearing premise

The replication structure of the functional observations is sufficient to identify and estimate the change-of-basis matrix that matches the directional regularity.

What would settle it

In simulated bivariate functional data generated with known directional regularity but without independent replications, the estimated change-of-basis matrix fails to recover the true directions and the observed smoothing rates remain at the non-adapted level.

Figures

Figures reproduced from arXiv: 2409.00817 by Omar Kassi, Sunny G.W. Wang.

Figure 1
Figure 1. Figure 1: In general, the worst regularity H1 will be obtained on the canonical basis, leading to isotropy (here H1 ă H2q. This can lead to slower rates of convergence. A change-of-basis from pe1, e2q to pu1, u2q enables one to obtain an anisotropic process. Locating the basis (u1, u2) is equivalent to locating the angle between u1 and e1. Illustration of directional regularity e1 e2 u2 u1 α H1 H2 H1 H1 3 [PITH_FUL… view at source ↗
Figure 2
Figure 2. Figure 2: Plot showing the remainder term for different angles [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Results for risk of estimated angles α (with correction). N is the number of curves, M0 the number of observed points along each curve, and σ is the noise level. 6 Discussion In this section we discuss some key aspects of our methodology. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Results for risk of estimated angles α (with correction). N is the number of curves, M0 the number of observed points along each curve, and σ is the noise level. 7.2 Anisotropic detection Our focus so far has been on estimating the direction of the maximizing regularity, which implicitly assumes that anisotropy is present. When the process is intrinsically isotropic, no gains can be made by structural adap… view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of the region in which the thresholding parameter [PITH_FULL_IMAGE:figures/full_fig_p022_5.png] view at source ↗
Figure 1
Figure 1. Figure 1: Results for risk of estimated angles α (with correction). N is the number of curves, M0 the number of observed points along each curve, and σ is the noise level. 4.2 Smoothing for products of fBms The effects of performing a change-of-basis on smoothing surfaces using our methodology in similarly explored for the products of fBms. The simulation setup is the same as Section 5.2, omitting the case of σ = 1,… view at source ↗
Figure 2
Figure 2. Figure 2: Results for risk of estimated angles α (with correction). N is the number of curves, M0 the number of observed points along each curve, and σ is the noise level. 4.3 Anisotropic Detection with higher noise-to-signal ratios Finally, we explore the application of anisotropic detection for the products of fBms. The same setup was used as Section 7.2, with 500 replications performed. The results can be seen in… view at source ↗
read the original abstract

We introduce directional regularity, a new definition of anisotropy for multivariate functional data. Instead of taking the conventional view, which determines anisotropy as a notion of smoothness along a dimension, directional regularity additionally views anisotropy through the lens of directions. We show that faster rates of convergence for smoothing can be obtained through a change-of-basis by adapting to the anisotropy of a bivariate process. An algorithm for the estimation and identification of the change-of-basis matrix is constructed, made possible due to the replication structure of functional data. Non-asymptotic bounds are provided for our algorithm, supplemented by numerical evidence from an extensive simulation study. Finally, a real-world rainfall measurement dataset is analyzed with our methods.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The paper introduces directional regularity as a new definition of anisotropy for bivariate functional data that incorporates directional information beyond conventional smoothness along dimensions. It claims that a change-of-basis adaptation to this anisotropy yields faster smoothing rates, constructs an algorithm to estimate and identify the change-of-basis matrix using the replication structure of functional data, provides non-asymptotic bounds for the procedure, and supports the claims with simulations and analysis of a rainfall dataset.

Significance. If the identification of the change-of-basis matrix and the resulting rate improvements are rigorously established, the work could meaningfully advance methods for anisotropic bivariate functional data by leveraging replication for adaptation, with potential applications in spatial or environmental statistics. The non-asymptotic bounds and extensive simulation study are positive features that would strengthen the contribution if the core identification result holds.

major comments (1)
  1. [Abstract] Abstract: the central claim that replication enables estimation and identification of the change-of-basis matrix (yielding strictly faster rates) is load-bearing, yet the provided description gives no indication that directional regularity produces a unique, recoverable matrix (e.g., via eigenstructure or directional variation of the covariance operator). A theorem establishing uniqueness and consistency of the estimator relative to the rate gain is required.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading and constructive feedback. We address the single major comment below, noting that the manuscript already contains the requested theoretical results on uniqueness and consistency, which we will highlight more explicitly in the abstract.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that replication enables estimation and identification of the change-of-basis matrix (yielding strictly faster rates) is load-bearing, yet the provided description gives no indication that directional regularity produces a unique, recoverable matrix (e.g., via eigenstructure or directional variation of the covariance operator). A theorem establishing uniqueness and consistency of the estimator relative to the rate gain is required.

    Authors: We agree that the abstract should more clearly signal the theoretical support for identification and rate gains. Theorem 3.2 establishes that directional regularity yields a unique change-of-basis matrix recoverable from the eigenstructure of the directional covariance operators. Theorem 4.1 then proves consistency of the replication-based estimator together with the accelerated convergence rates. We will revise the abstract to reference these results and the role of replication in identification. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained

full rationale

The abstract and description introduce directional regularity as a new definition and construct an estimation algorithm for the change-of-basis matrix enabled by replication structure. No quoted equations or steps reduce by construction to fitted parameters, self-citations, or ansatzes. The claim of faster smoothing rates via adaptation stands as an independent result without load-bearing self-referential reductions. This is the normal non-circular outcome for a methods paper presenting a new definition and algorithm.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no specific free parameters, axioms, or invented entities are detailed.

pith-pipeline@v0.9.0 · 5639 in / 863 out tokens · 22763 ms · 2026-05-23T20:54:34.283912+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    Ammous, S., Dedecker, J., and Duval, C. (2024). Adaptive directional estimator of the density in R^d for independent and mixing sequences. J. Multivariate Anal. , 203:105332

    work page 2024

  2. [2]

    M., Secchi, P., Domanin, M., and Nobile, F

    Azzimonti, L., Sangalli, L. M., Secchi, P., Domanin, M., and Nobile, F. (2015). Blood flow velocity field estimation via spatial regression with PDE penalization. J. Amer. Statist. Assoc. , 110(511):1057--1071

    work page 2015

  3. [3]

    Belloni, A., Chernozhukov, V., Chetverikov, D., and Kato, K. (2015). Some new asymptotic theory for least squares series: pointwise and uniform results. J. Econometrics , 186(2):345--366

    work page 2015

  4. [4]

    S., Carey, M., Ramsay, J

    Bernardi, M. S., Carey, M., Ramsay, J. O., and Sangalli, L. M. (2018). Modeling spatial anisotropy via regression with partial differential regularization. J. Multivariate Anal. , 167:15--30

    work page 2018

  5. [5]

    and Porcu, E

    Coeurjolly, J.-F. and Porcu, E. (2018). Fast and exact simulation of complex-valued stationary G aussian processes through embedding circulant matrix. J. Comput. Graph. Statist. , 27(2):278--290

    work page 2018

  6. [6]

    and Hall, P

    Davies, S. and Hall, P. (1999). Fractal analysis of surface roughness by using spatial data. J. R. Stat. Soc. Ser. B Stat. Methodol. , 61(1):3--37. With discussion and a reply by the authors

    work page 1999

  7. [7]

    Dieker, T. (2004). Simulation of fractional brownian motion

    work page 2004

  8. [8]

    and Guerre, E

    Fan, Y. and Guerre, E. (2016). Multivariate local polynomial estimators: Uniform boundary properties and asymptotic linear representation. In Essays in Honor of Aman Ullah , volume 36, pages 489--537. Emerald Group Publishing Limited

    work page 2016

  9. [9]

    Germain, C., Da Costa , J., Lavialle, O., and Baylou, P. (2003). Multiscale estimation of vector field anisotropy application to texture characterization. Signal Processing , 83(7):1487--1503

    work page 2003

  10. [10]

    Golovkine, S., Klutchnikoff, N., and Patilea, V. (2022). Learning the smoothness of noisy curves with application to online curve estimation. Electron. J. Stat. , 16(1):1485--1560

    work page 2022

  11. [11]

    Golovkine, S., Klutchnikoff, N., and Patilea, V. (2023). Adaptive optimal estimation of irregular mean and covariance functions. arxiv:2108.06507v2

    work page arXiv 2023

  12. [12]

    Herbin, E. (2006). From N parameter fractional B rownian motions to N parameter multifractional B rownian motions. Rocky Mountain J. Math. , 36(4):1249--1284

    work page 2006

  13. [13]

    and Kokoszka, P

    Horv\'ath, L. and Kokoszka, P. (2012). Inference for functional data with applications . Springer Series in Statistics. Springer, New York

    work page 2012

  14. [14]

    Jain, A., Hong, L., and Bolle, R. (1997). On-line fingerprint verification. IEEE Trans. Pattern Anal. Mach. Intell. , 19(4):302--314

    work page 1997

  15. [15]

    Jiang, X. (2005). On orientation and anisotropy estimation for online fingerprint authentication. IEEE Trans. Signal Process. , 53(10):4038--4049

    work page 2005

  16. [16]

    Kassi, O., Klutchnikoff, N., and Patilea, V. (2023). Learning the regularity of multivariate functional data

    work page 2023

  17. [17]

    and Reimherr, M

    Kokoszka, P. and Reimherr, M. (2017). Introduction to functional data analysis . Texts in Statistical Science Series. CRC Press, Boca Raton, FL

    work page 2017

  18. [18]

    Lepski, O. (2015). Adaptive estimation over anisotropic functional classes via oracle approach. Ann. Statist. , 43(3):1178--1242

    work page 2015

  19. [19]

    Lepski, O. V. and Rebelles, G. (2020). Structural adaptation in the density model. Math. Stat. Learn. , 3(3-4):345--386

    work page 2020

  20. [20]

    Maissoro, H., Patilea, V., and Vimond, M. (2024). Adaptive estimation for weakly dependent functional times series

    work page 2024

  21. [21]

    Matheron, G. (1973). The intrinsic random functions and their applications. Advances in Appl. Probability , 5:439--468

    work page 1973

  22. [22]

    Ramsay, J. O. and Silverman, B. W. (2005). Functional data analysis . Springer Series in Statistics. Springer, New York, second edition

    work page 2005

  23. [23]

    Ramsay, T. (2002). Spline smoothing over difficult regions. J. R. Stat. Soc. Ser. B Stat. Methodol. , 64(2):307--319

    work page 2002

  24. [24]

    and Yor, M

    Revuz, D. and Yor, M. (1999). Continuous martingales and B rownian motion , volume 293 of Grundlehren der mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences] . Springer-Verlag, Berlin, third edition

    work page 1999

  25. [25]

    Richard, F. J. P. (2016). Tests of isotropy for rough textures of trended images. Statist. Sinica , 26(3):1279--1304

    work page 2016

  26. [26]

    and Tsybakov, A

    Samarov, A. and Tsybakov, A. (2004). Nonparametric independent component analysis. Bernoulli , 10(4):565--582

    work page 2004

  27. [27]

    M., Ramsay, J

    Sangalli, L. M., Ramsay, J. O., and Ramsay, T. O. (2013). Spatial spline regression models. J. R. Stat. Soc. Ser. B. Stat. Methodol. , 75(4):681--703

    work page 2013

  28. [28]

    and Hsing, T

    Shen, J. and Hsing, T. (2020). Hurst function estimation . Ann. Statist. , 48(2):838 -- 862

    work page 2020

  29. [29]

    Stein, M. L. (2002). Fast and exact simulation of fractional B rownian surfaces. J. Comput. Graph. Statist. , 11(3):587--599

    work page 2002

  30. [30]

    Tsybakov, A. B. (2009). Introduction to nonparametric estimation . Springer Series in Statistics. Springer, New York. Revised and extended from the 2004 French original, Translated by Vladimir Zaiats

    work page 2009

  31. [31]

    G., Patilea, V., and Klutchnikoff, N

    Wang, S. G., Patilea, V., and Klutchnikoff, N. (2024+). Adaptive functional principal components analysis. J. R. Stat. Soc. Ser. B. Stat. Methodol

    work page 2024

  32. [32]

    Wood, A. T. A. and Chan, G. (1994). Simulation of stationary G aussian processes in [0,1]^d . J. Comput. Graph. Statist. , 3(4):409--432

    work page 1994

  33. [33]

    N., Bravington, M

    Wood, S. N., Bravington, M. V., and Hedley, S. L. (2008). Soap film smoothing. J. R. Stat. Soc. Ser. B Stat. Methodol. , 70(5):931--955

    work page 2008

  34. [34]

    Boucheron, S., Lugosi, G., and Massart, P. (2013). Concentration Inequalities: A Nonasymptotic Theory of Independence . Oxford University Press

    work page 2013

  35. [35]

    and Yor, M

    Revuz, D. and Yor, M. (2013). Continuous martingales and Brownian motion , volume 293. Springer Science & Business Media

    work page 2013