Pith. sign in

REVIEW 4 major objections 8 minor 108 references

Variable Selection Methods for Multivariate, Functional, and Complex Biomedical Data in the AI Age

T0 review · 4 major / 8 minor · reviewed 2026-08-10 · deepseek-v4-flash

Pith's one-line read A fast variable-selection solver for complex, metric-space biomedical responses.

desk verdict The paper's claimed equivalence to FRISO is false, so the metric-space consistency results collapse; the multivariate best-subset extension is plausible but the empirical and theoretical support is not there. read the letter →

arxiv 2501.06868 v1 pith:3JX7GRCP submitted 2025-01-12 stat.ML cs.LGstat.APstat.ME

classification stat.MLcs.LGstat.APstat.ME MSC 62R2062J0790C25
keywords variableselectionbestsubsetmetricspaceresponsesFréchetregressionsaddle-pointoptimizationdualsubgradientmethodfunctionaldatadigitalhealth
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that best-subset variable selection can be made practical for regression problems where the response is multivariate, functional, distributional, or graph-valued in a metric space. It formulates the selection problem as a cardinality-constrained ridge regression, relaxes the binary variables to a saddle-point problem, and solves it with a dual projected subgradient algorithm. The paper claims this solver matches the selection accuracy of the only existing general method for metric-space responses (FRISO) while running orders of magnitude faster, reaching sample sizes in the hundreds of thousands or millions. If true, this would let clinicians and data scientists select interpretable predictors for complex biomedical outcomes such as glucose distribution profiles and brain graphs at cohort scale.

What carries the argument

The engine is the dual saddle-point problem obtained by Fenchel-conjugating the losses and dualizing the ridge penalty, with binary support variables relaxed to the unit interval. The objective is linear in the support variables and concave in the dual variables, so Sion's minimax theorem allows the order of minimization and maximization to be interchanged; a tightness condition from Pilanci et al. guarantees that the relaxed solution is binary. The update loop in Algorithm 1 alternates a sorting step that selects the k variables with largest α⊤XjXj⊤α and a projected subgradient step on the dual variables. The perspective-function reformulation of the ridge penalty is the bridge used to claim equivalence with FRISO, the Fréchet Ridge Selection Operator, an individually penalized ridge regression for global Fréchet regression.

What would settle it

Construct a small counterexample with p = 2, n = 50, correlated predictors, and a distributional response; solve Equation (15) by exhaustive enumeration over the four possible subsets and solve Equation (10) with FRISO's optimizer. If the selected support sets differ for any matched tuning parameters (k, τ), Proposition 4 is false or only approximate. Alternatively, run Algorithm 1 on a problem where the saddle point is not unique and check whether the averaged iterate converges; divergence would falsify the convergence assumption.

Watch

Extended reading notes

Core claim

The central claim is that a single optimization framework handles variable selection across many response types by treating the response space through a Hilbert-space embedding. For metric spaces of negative type, the squared distance embeds isometrically into a Hilbert space, so the global Fréchet regression model becomes a linear model in that transformed space, and the paper argues that the best-subset problem with a ridge penalty is equivalent to the FRISO individually penalized ridge regression problem (Proposition 4). On this equivalence it imports FRISO's consistency result. The computational discovery is that the Boolean relaxation of the saddle-point formulation is tight under a generic uniqueness condition, so a simple projected subgradient iteration, updating the support by sorting and the dual variable by one subgradient step, solves the relaxed problem and recovers the best subset.

Load-bearing premise

The load-bearing premise is Proposition 4: that the cardinality-constrained ridge problem in Equation (15) is exactly equivalent to FRISO's continuously penalized ridge problem in Equation (10); the paper states this equivalence without proof, and both the imported consistency theorem and the metric-space generalization stand on it.

Editorial extensions

If this is right

  • Selection of predictors for distributional and graph-valued responses becomes computationally feasible at cohort scale (n in hundreds of thousands), not just for datasets of a few hundred observations.
  • The same algorithm can fit different loss functions for each coordinate of a multivariate response, enabling mixed models that combine genomic, wearable, and electronic-health-record data.
  • Consistency of the selected variable set follows from FRISO's theory whenever the stated equivalence holds and the response space is of negative type.
  • Support for group and block constraints extends the method to one-hot encoded categorical predictors and additive spline bases.
  • On large multivariate linear problems the paper reports near-perfect support recovery with runtimes under tens of seconds.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the equivalence with FRISO is proved and the constant-step-size subgradient convergence is established, the method would inherit a complete consistency theory and could replace lasso-type screening in personalized-medicine pipelines.
  • A concrete testable extension is to apply the solver to 2-Wasserstein distributional responses in clinical trial data, where current practice often collapses the distribution to mean and variance; the paper's own examples suggest that distributional outcomes can change which predictors are selected.
  • The claimed speedup rests on avoiding matrix inversions, which suggests the same dual subgradient strategy may accelerate other Fenchel-dual sparse regression problems beyond metric-space responses.
Share X Bluesky LinkedIn Reddit HN

Formalized claims in Lean

  1. Claim #1: The central claim is that a single optimization framework handles variable selection across many response types by treating the response space through a Hilbert-space embedding. For metric spaces of negative type, the squared distance embeds isometrically into a Hilbert space, so the global Fréchet regression model becomes a linear model in that transformed space, and the paper argues that the bes

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 8 minor

Summary. The paper proposes a best-subset variable selection framework for regression problems with multivariate, functional, and metric-space-valued responses. The method solves a cardinality-constrained empirical risk minimization with a ridge penalty (Eqs. 1/15), derives a saddle-point dual for convex losses (Theorem 1), relaxes the binary support variables to the unit interval, imports a tightness condition from Pilanci, Wainwright, and El Ghaoui (Theorem 2), and solves the relaxed problem with a projected dual subgradient algorithm (Algorithm 1). Section 2.3 claims, via Proposition 4, that the proposed problem is equivalent to the FRISO individual-ridge procedure of Tucker, Wu, and Müller, and on that basis imports FRISO's selection-consistency theorem (Theorem 5). The remainder applies the method to three diabetes datasets (multivariate biomarkers, bivariate longitudinal CGM outcomes, distributional glucose profiles) and reports simulations for multivariate Euclidean responses and 2-Wasserstein distribution responses, claiming several orders of magnitude speed improvement over competitors.

Significance. If established, the contribution would be genuinely valuable: a scalable ℓ₀-based solver for variable selection with complex responses, extending the Bertsimas–Pauphilet–Van Parys saddle-point approach to multivariate and metric-valued outcomes, with clinically motivated applications in diabetes research. The saddle-point dual for convex losses is plausible, the use of the Pilanci et al. Boolean-relaxation tightness result is appropriate, and the reported wall-clock times in Tables 4–5 (e.g., about 31 seconds for n = 100,000 with an m = 300 quantile grid) indicate real computational efficiency. However, the central theoretical bridge (Proposition 4) is unproven and appears false on its face (ℓ₀ + ℓ₂ versus an effective ℓ₁-type penalty), so the imported consistency theorem does not apply to the proposed method as stated; Algorithm 1 has no convergence analysis; and the headline speed claims are not backed by any competitor runtimes. These gaps undermine the paper's two central claims, so the manuscript cannot be accepted in its current form.

major comments (4)
  1. [2.3.5 (Proposition 4)] Proposition 4 asserts that problem (15) is equivalent to the FRISO individual-ridge problem (10), but no proof is given and the preceding perspective-function manipulation only rewrites (15); it does not relate (15) to (10). The equivalence is not a routine consequence of the definitions: for scalar squared-error loss, FRISO with constraints λ_j ≥ 0 and Σ_j λ_j ≤ τ induces the effective penalty βᵀ diag(1/λ)β whose minimum over λ is (Σ_j|β_j|)²/τ, an ℓ₁-type penalty, whereas (15) is an ℓ₀-cardinality-plus-ℓ₂ problem; ℓ₁-type and ℓ₀-type objectives are known to have different solution paths under correlated predictors, so the asserted equality fails in general. Because Theorem 5 from [50] is invoked solely through Proposition 4, the selection-consistency conclusion in Section 2.4 is not established for the estimator defined by (15) and Algorithm 1; Proposition 6, which is meant to supply the remaining conditions, is also stated without proof.
  2. [2.1.2 (Algorithm 1)] Algorithm 1 is presented as the paper's main computational contribution, but no convergence analysis is provided: the constant step size δ is unspecified, the projection operator P is not defined with respect to any norm or feasible set, and no theorem or proposition establishes even an asymptotic stationarity property or an O(δ) optimality gap for the averaged output (α̂_T, Ņ). The claim in Section 3.2.2 that "the algorithm obtained the optimal solution in just two seconds" therefore has no support; without a convergence guarantee (or at least a certified tolerance for solving the relaxed problem (3)), the speed and accuracy results in Sections 3–4 cannot be attributed to solving the stated optimization problem.
  3. [Abstract and Sections 4.1–4.2] The abstract claims "several orders of magnitude improvement over competitors," but the experiments do not provide the necessary baselines. Section 4.1 states explicitly that "we do not include a comparison with other approaches in this scenario," so the large-n multivariate timings in Table 4 have no competitor. In the distributional scenario (Section 4.2), the comparison with FRISO reports only that both methods select the true variable with proportion 1.00; Table 5 contains no FRISO runtimes, and the text limits FRISO to "datasets of less n = 300 observations" without quantifying its runtime even at n = 300. The speed advantage over FRISO or any other method is therefore never measured, and the abstract's accuracy claim is also unsupported: on the only comparative scenario the reported result is parity (1.00 versus 1.00), while the real-data sections (3.2–3.4) contain no competitor analyses and report in-sample R² only.
  4. [2.4 (Theorem 5, Remark 5)] Even if Proposition 4 were granted, the consistency statement does not connect to the proposed selector. Theorem 5 concerns the FRISO weights λ̂_j(τ_n), with selection defined by Î(τ) = {j : λ̂_j(τ) > 0}; the proposed method instead selects the k variables with the largest score from the saddle point of (3). The paper never shows that these two selection rules coincide, and Remark 5 concedes that Theorem 5 yields only λ̂_{j'} →_p 0, not exact support recovery. The opening sentence of Section 2.4, "we can select the correct variables as n grows to infinity," is therefore not justified for the estimator that is actually implemented; this is an internal-support gap rather than a disagreement with external consensus.
minor comments (8)
  1. [Throughout] The manuscript contains numerous typographical errors (e.g., "beetween", "contemporany", "ilustrate", "millons", "advanatgues", "uncertainity", "trejectories" in the Figure 2 caption, and "Weigth" and "Percentil" in Figure 6); a thorough proofread is needed.
  2. [Introduction and 2.3.2] The FRISO method is cited as reference [8] in the Introduction and in Section 2.3.2, but [8] is a review on wearable biosensors; the correct reference is the Tucker–Wu–Müller paper, which is unfortunately duplicated as both [50] and [73].
  3. [2.1.1] The tightness result is labeled "Theorem 1" a second time, with a stray "2" following the label; it should be renumbered as Theorem 2.
  4. [2.3.4 (Remark 4)] Remark 4 refers to "Equation 17" where equation (14) is meant.
  5. [3.4.1 versus Table 1] Section 3.4.1 describes a case-control study with 200 participants, while Table 1 reports "over 350 patients" for the same distributional application; the numbers should be reconciled.
  6. [2.4 and 4.1] The Conditions [A]–[D] referenced in Theorem 5 and the "complete results" for Section 4.1 are said to be in an Appendix that is not included in the manuscript.
  7. [4.1] The sample size is given as "n = 1000,000" in one place and "1, 000, 000" in another, while Table 4 reports n up to 100,000; the intended maximum sample size should be stated consistently.
  8. [Table 5] The units of "Average Time" in Table 5 are not specified, and the absence of any FRISO timing column makes the claimed speed comparison unverifiable.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the consistency claim is imported from an external FRISO theorem; the unproven equivalence in Proposition 4 is a correctness gap, not a circular reduction.

full rationale

The paper's central statistical guarantee is inherited from the external FRISO consistency theorem (Theorem 5, cited to Tucker, Wu and Müller [50], none of whom are authors of this paper). Importing an external result is independent support rather than circularity, and the paper does not fit parameters and then relabel them as predictions. The main load-bearing step is Proposition 4, which asserts that Eq. (15), a cardinality-constrained ℓ0+ℓ2 ridge problem, is equivalent to FRISO's individually λ-penalized ridge problem. That assertion is made without proof and is mathematically non-obvious, since FRISO's penalty induced by the λ-constraint is ℓ1-type rather than ℓ0-type; however, an unproven or even false equivalence is a missing-support/correctness issue, not a circular one. The paper's equations are not identical by construction, and no same-author citation chain forces the conclusion. The unproven convergence of Algorithm 1 is likewise a missing proof. Thus no circular step can be exhibited, and the appropriate verdict is no significant circularity.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The method depends on a sparsity budget k and ridge strength gamma, neither of which is tuned by a data-driven rule. The metric-space theory rests on the negative-type embedding and on the Fréchet regression model, plus the unproven equivalence to FRISO. No new entities are invented.

free parameters (6)
  • k (sparsity budget) = 9 (NHANES), 3 (JDRF), 1/3/5/8 (simulations)
    User-specified cardinality bound in Problem (1); no data-driven selection rule or sensitivity analysis.
  • gamma (ridge strength) = not reported
    Controls ridge regularization in the objective; no tuning procedure or grid values given.
  • delta (subgradient step size) = not reported
    Constant step size in Algorithm 1; no formula, schedule, or convergence condition.
  • Stop criterion for Algorithm 1 = not specified
    The loop ends at an unspecified criterion; affects solution quality and runtime.
  • Kernel bandwidth h for density estimation = not reported
    Used in Eq. (24) to construct distributional glucose responses; not specified or varied.
  • Quantile grid size m = 300 or 500 grid points in applications; 50/150/300 in simulations
    Discretization level for functional/distributional responses; chosen by hand.
assumptions (6)
  • domain assumption Loss functions are convex with additive linear structure.
    Theorem 1 and the dual derivation require convexity; the paper cites convex losses in Table 2.
  • domain assumption Response space is a separable Hilbert space or embeds into one (negative type).
    Section 2.3.3 restricts to negative type spaces via Schoenberg embedding; not all metric spaces satisfy this.
  • domain assumption Global Fréchet regression model of Petersen and Müller is the correct regression model for metric responses.
    Section 2.3.1 adopts this model (Eq. 9) as the base estimator.
  • domain assumption Conditions [A-D] from the appendix hold.
    Theorem 5 cites conditions [A-D] and (U0)-(U2) of [78]; the paper does not verify them in the applications, and the appendix is not included in the preprint.
  • standard math Boolean relaxation tightness condition holds (distinct top-k components).
    Theorem 2 relies on [76, Proposition 1]; in correlated real data this is an assumption.
  • domain assumption Conditional Fréchet mean exists and is unique.
    Section 2.3 assumes uniqueness and existence of the minimizer in Eq. (8).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Variable Selection Methods for Multivariate, Functional, and Complex Biomedical Data in the AI Age." pith.science (2026). https://pith.science/paper/3JX7GRCP

@misc{pith2026250106868,
  author       = {Pith},
  title        = {Pith review of: Variable Selection Methods for Multivariate, Functional, and Complex Biomedical Data in the AI Age},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3JX7GRCP}},
  note         = {Machine review of arXiv:2501.06868}
}
read the original abstract

Many problems within personalized medicine and digital health rely on the analysis of continuous-time functional biomarkers and other complex data structures emerging from high-resolution patient monitoring. In this context, this work proposes new optimization-based variable selection methods for multivariate, functional, and even more general outcomes in metrics spaces based on best-subset selection. Our framework applies to several types of regression models, including linear, quantile, or non parametric additive models, and to a broad range of random responses, such as univariate, multivariate Euclidean data, functional, and even random graphs. Our analysis demonstrates that our proposed methodology outperforms state-of-the-art methods in accuracy and, especially, in speed-achieving several orders of magnitude improvement over competitors across various type of statistical responses as the case of mathematical functions. While our framework is general and is not designed for a specific regression and scientific problem, the article is self-contained and focuses on biomedical applications. In the clinical areas, serves as a valuable resource for professionals in biostatistics, statistics, and artificial intelligence interested in variable selection problem in this new technological AI-era.

Figures

Figures reproduced from arXiv: 2501.06868 by the authors.

Figure 1
Figure 1. Variation in the mean and standard deviation of glucose values for a diabetic individual [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Average glucose trajectories (left) and standard deviation trejectories (right). [PITH_FULL_IMAGE:figures/full_fig_p017_2.png] view at source ↗
Figure 3
Figure 3. Left: Raw CGM time series of two individuals. Center: The corresponding density functions. [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: P-values across the temporal domain of the statistical significance of each variable selected [PITH_FULL_IMAGE:figures/full_fig_p019_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

108 extracted references · 69 canonical work pages

  1. [50]

    D. C. Tucker, Y. Wu, and H.-G. Müller. Variable selection for global fréchet regression.Journal of the American Statistical Association, 0(0):1–15, 2021

  2. [1]

    Xiao Li, Jessilyn Dunn, Denis Salins, Gao Zhou, Wenyu Zhou, Sophia Miryam Schüssler- Fiorenza Rose, Dalia Perelman, Elizabeth Colbert, Ryan Runge, Shannon Rego, Ria Sonecha, Somalee Datta, Tracey McLaughlin, and Michael P. Snyder. Digital health: Tracking physiomes and activity using wearable biosensors reveals useful health-related information.PLoS Biolo...

  3. [2]

    SIAM, 2015

    Michael R Kosorok and Erica EM Moodie.Adaptive Treatment Strategies in Practice: Planning Trials and Analyzing Data for Personalized Medicine. SIAM, 2015

  4. [3]

    Precision medicine.Annual Review of Statistics and Its Application, 6:263–286, 2019

    Michael R Kosorok and Eric B Laber. Precision medicine.Annual Review of Statistics and Its Application, 6:263–286, 2019

  5. [4]

    Chapman and Hall/CRC, 2019

    Anastasios A Tsiatis, Marie Davidian, Shannon T Holloway, and Eric B Laber.Dynamic treatment regimes: Statistical methods for precision medicine. Chapman and Hall/CRC, 2019

  6. [5]

    PhD thesis

    Marcos Matabuena Rodríguez.Contributions on metric spaces with applications in personalized medicine. PhD thesis

  7. [6]

    Statistical analysis of high- dimensional biomedical data: a gentle introduction to analytical goals, common approaches and challenges

    Joachim Rahnenführer, Raffaele De Bin, Andreas Benner, et al. Statistical analysis of high- dimensional biomedical data: a gentle introduction to analytical goals, common approaches and challenges. BMC Medicine, 21(1):182, 2023

  8. [7]

    Glucodensity Functional Profiles Outperform Traditional Continuous Glucose Monitoring Metrics

    Marcos Matabuena, Rahul Ghosal, Javier Enrique Aguilar, Robert Wagner, Carmen Fernández Merino, Juan Sánchez Castro, Vadim Zipunnikov, Jukka-Pekka Onnela, and Francisco Gude. Glucodensity functional profiles outperform traditional continuous glucose monitoring metrics. arXiv preprint arXiv:2410.00912, 2024

Show all 108 references
  1. [8]

    The era of digital health: A review of portable and wearable affinity biosensors.Advanced Functional Materials, 30(29):1906713, 2020

    Jiaobing Tu, Rebeca M Torrente-Rodríguez, Minqiang Wang, and Wei Gao. The era of digital health: A review of portable and wearable affinity biosensors.Advanced Functional Materials, 30(29):1906713, 2020

  2. [9]

    Generalized additive models for location, scale and shape

    Robert A Rigby and D Mikis Stasinopoulos. Generalized additive models for location, scale and shape. Journal of the Royal Statistical Society Series C: Applied Statistics, 54(3):507–554, 2005. 23

  3. [10]

    Springer Science & Business Media, 2011

    Peter Bühlmann and Sara Van De Geer.Statistics for High-Dimensional Data: Methods, Theory and Applications. Springer Science & Business Media, 2011

  4. [11]

    Best subset selection via a modern optimization lens

    Dimitris Bertsimas, Angela King, and Rahul Mazumder. Best subset selection via a modern optimization lens. The Annals of Statistics, pages 813–852, 2016

  5. [12]

    A selective review of group selection in high- dimensional models

    Jian Huang, Patrick Breheny, and Shuangge Ma. A selective review of group selection in high- dimensional models. Statistical Science, 27(4), 2012

  6. [13]

    A general theory of concave regularization for high-dimensional sparse estimation problems.Statistical Science, 27(4):576–593, 2012

    Cun-Hui Zhang and Tong Zhang. A general theory of concave regularization for high-dimensional sparse estimation problems.Statistical Science, 27(4):576–593, 2012

  7. [14]

    High-dimensional statistics with a view toward applications in biology.Annual Review of Statistics and Its Application, 1(1):255–278, 2014

    Peter Bühlmann, Markus Kalisch, and Lukas Meier. High-dimensional statistics with a view toward applications in biology.Annual Review of Statistics and Its Application, 1(1):255–278, 2014

  8. [15]

    Efficient quadratic regularization for expression arrays

    Trevor Hastie and Robert Tibshirani. Efficient quadratic regularization for expression arrays. Biostatistics, 5(3):329–340, 2004

  9. [16]

    A variable selection method for genome-wide association studies

    Qianchuan He and Dan-Yu Lin. A variable selection method for genome-wide association studies. Bioinformatics, 27(1):1–8, 2011

  10. [17]

    Regression shrinkage and selection via the lasso.Journal of the Royal Statistical Society: Series B (Methodological), 58(1):267–288, 1996

    Robert Tibshirani. Regression shrinkage and selection via the lasso.Journal of the Royal Statistical Society: Series B (Methodological), 58(1):267–288, 1996

  11. [18]

    Regularization and variable selection via the elastic net.Journal of the royal statistical society: series B (statistical methodology), 67(2):301–320, 2005

    Hui Zou and Trevor Hastie. Regularization and variable selection via the elastic net.Journal of the royal statistical society: series B (statistical methodology), 67(2):301–320, 2005

  12. [19]

    Variable selection via nonconcave penalized likelihood and its oracle properties

    Jianqing Fan and Runze Li. Variable selection via nonconcave penalized likelihood and its oracle properties. Journal of the American Statistical Association, 96(456):1348–1360, 2001

  13. [20]

    Fast best subset selection: Coordinate descent and local combinatorial optimization algorithms.Operations Research, 68(5):1517–1537, 2020

    Hussein Hazimeh and Rahul Mazumder. Fast best subset selection: Coordinate descent and local combinatorial optimization algorithms.Operations Research, 68(5):1517–1537, 2020

  14. [21]

    Yoav Benjamini and Yosef Hochberg. Controlling the false discovery rate: a practical and pow- erful approach to multiple testing.Journal of the Royal Statistical Society: Series B (Statistical Methodology), 57(1):289–300, 1995

  15. [22]

    False discoveries occur early on the lasso path

    Weijie Su, Małgorzata Bogdan, and Emmanuel Candes. False discoveries occur early on the lasso path. The Annals of Statistics, pages 2133–2150, 2017

  16. [23]

    Slope—adaptive variable selection via convex optimization

    Małgorzata Bogdan, Ewout Van Den Berg, Chiara Sabatti, Weijie Su, and Emmanuel J Candès. Slope—adaptive variable selection via convex optimization. The Annals of Applied Statistics, 9(3):1103, 2015

  17. [24]

    Controlling the rate of gwas false discoveries.Genetics, 205(1):61–75, 2017

    Damian Brzyski, Christine B Peterson, Piotr Sobczyk, Emmanuel J Candès, Malgorzata Bogdan, and Chiara Sabatti. Controlling the rate of gwas false discoveries.Genetics, 205(1):61–75, 2017

  18. [25]

    Gene hunting with hidden markov model knockoffs

    Matteo Sesia, Chiara Sabatti, and Emmanuel J Candès. Gene hunting with hidden markov model knockoffs. Biometrika, 106(1):1–18, 2019

  19. [26]

    Multi- resolution localization of causal variants across the genome.Nature Communications, 11(1):1–10, 2020

    Matteo Sesia, Eugene Katsevich, Stephen Bates, Emmanuel Candès, and Chiara Sabatti. Multi- resolution localization of causal variants across the genome.Nature Communications, 11(1):1–10, 2020

  20. [27]

    Least angle regression.The Annals of statistics, 32(2):407–499, 2004

    Bradley Efron, Trevor Hastie, Iain Johnstone, and Robert Tibshirani. Least angle regression.The Annals of statistics, 32(2):407–499, 2004

  21. [28]

    Convexlar: an extension of least angle regression.Journal of Computational and Graphical Statistics, 24(3):603–626, 2015

    Wei Xiao, Yichao Wu, and Hua Zhou. Convexlar: an extension of least angle regression.Journal of Computational and Graphical Statistics, 24(3):603–626, 2015

  22. [29]

    Robust linear model selection based on least angle regression

    Jafar A Khan, Stefan Van Aelst, and Ruben H Zamar. Robust linear model selection based on least angle regression. Journal of the American Statistical Association, 102(480):1289–1299, 2007. 24

  23. [30]

    Sparse regression: Scalable algorithms and empirical performance.Statistical Science, 35(4):555–578, 2020

    Dimitris Bertsimas, Jean Pauphilet, and Bart Van Parys. Sparse regression: Scalable algorithms and empirical performance.Statistical Science, 35(4):555–578, 2020

  24. [31]

    Sparse classification: a scalable discrete optimization perspective

    Dimitris Bertsimas, Jean Pauphilet, and Bart Van Parys. Sparse classification: a scalable discrete optimization perspective. Machine Learning, 110(11):3177–3209, 2021

  25. [32]

    Matabuena

    M. Matabuena. Contributions on metric spaces with applications in personalized medicine. PhD thesis, Universidade de Santiago de Compostela, 2022

  26. [33]

    Glucodensities: A new representation of glucose profiles using distributional data analysis.Statistical Methods in Medical Research, 30(6):1445–1464, 2021

    Marcos Matabuena, Alexander Petersen, Juan C Vidal, and Francisco Gude. Glucodensities: A new representation of glucose profiles using distributional data analysis.Statistical Methods in Medical Research, 30(6):1445–1464, 2021. PMID: 33760665

  27. [34]

    Kernel bicluster- ing algorithm in Hilbert spaces.ArXiv Preprint, 2022

    Marcos Matabuena, JC Vidal, Oscar Hernan Madrid Padilla, and Dino Sejdinovic. Kernel bicluster- ing algorithm in Hilbert spaces.ArXiv Preprint, 2022

  28. [35]

    Distributional data analysis with accelerometer data in a NHANES database with nonparametric survey regression models.ArXiv Preprint, 2021

    Marcos Matabuena and Alex Petersen. Distributional data analysis with accelerometer data in a NHANES database with nonparametric survey regression models.ArXiv Preprint, 2021

  29. [36]

    Scalar on time-by-distribution regression and its application for modelling associations between daily-living physical activity and cognitive functions in alzheimer’s disease

    Rahul Ghosal, Vijay R Varma, Dmitri Volfson, Jacek Urbanek, Jeffrey M Hausdorff, Amber Watts, and Vadim Zipunnikov. Scalar on time-by-distribution regression and its application for modelling associations between daily-living physical activity and cognitive functions in alzhei...

  30. [37]

    Marcos Matabuena, Paulo Félix, Ziad Akram Ali Hammouri, Jorge Mota, and Borja del Pozo Cruz. Physical activity phenotypes and mortality in older adults: a novel distributional data analysis of accelerometry in the nhanes.Aging Clinical and Experimental Research, pages 1–8, 2022

  31. [38]

    Ghosal and M

    R. Ghosal and M. Matabuena. Multivariate scalar on multidimensional distribution regression. 2023

  32. [39]

    Conditional Wasserstein barycenters and interpola- tion/extrapolation of distributions.arXiv preprint arXiv:2107.09218, 2021

    Jianing Fan and Hans-Georg Müller. Conditional Wasserstein barycenters and interpola- tion/extrapolation of distributions.arXiv preprint arXiv:2107.09218, 2021

  33. [40]

    Y. Chen, Z. Lin, and H.-G. Müller. Wasserstein regression.Journal of the American Statistical Association, 116(535):1–14, 2021

  34. [41]

    Petersen, X

    A. Petersen, X. Liu, and A. A. Divani. Wassersteinf-tests and confidence bands for the fréchet regression of density response curves.The Annals of Statistics, 49(1):590–611, 2021

  35. [42]

    Zhou and H.-G

    Y. Zhou and H.-G. Müller. Network regression with graph laplacians.Journal of Machine Learning Research, 23(320):1–41, 2022

  36. [43]

    Dubey and H.-G

    P. Dubey and H.-G. Müller. Modeling time-varying random objects and dynamic networks.Journal of the American Statistical Association, 117(540):2252–2267, 2022

  37. [44]

    J. M. Jeon, Y. K. Lee, E. Mammen, and B. U. Park. Locally polynomial hilbertian additive regression. Bernoulli, 28(3):2034–2066, 2022

  38. [45]

    Kurisu and T

    D. Kurisu and T. Otsu. Model averaging for global fréchet regression. 2023

  39. [46]

    Chen and H.-G

    H. Chen and H.-G. Müller. Sliced wasserstein regression.arXiv preprint arXiv:2306.10601, 2023

  40. [47]

    R. Lyons. Second errata to distance covariance in metric spaces. The Annals of Probability, 49(5):2668–2670, 2021

  41. [48]

    Dubey and H.-G

    P. Dubey and H.-G. Müller. Fréchet analysis of variance for random objects.Biometrika, 106(4):803– 821, 2019

  42. [49]

    Fout and B

    A. Fout and B. K. Fosdick. Fréchet covariance and manova tests for random objects in multiple metric spaces. arXiv preprint arXiv:2306.12066, 2023. 25

  43. [51]

    Matabuena, C

    M. Matabuena, C. Díaz-Louzao, R. Ghosal, and F. Gude. Personalized imputation in metric spaces via conformal prediction: Applications in predicting diabetes development with continuous glucose monitoring information. 2024

  44. [53]

    Bhattacharjee and H.-G

    S. Bhattacharjee and H.-G. Müller. Geodesic mixed effects models for repeatedly ob- served/longitudinal random objects.arXiv preprint arXiv:2307.05726, 2023

  45. [54]

    Conformal uncertainty quantification using kernel depth measures in separable hilbert spaces

    Marcos Matabuena, Rahul Ghosal, Pavlo Mozharovskyi, Oscar Hernan Madrid Padilla, and Jukka- Pekka Onnela. Conformal uncertainty quantification using kernel depth measures in separable hilbert spaces. arXiv preprint arXiv:2405.13970, 2024

  46. [55]

    Uncertainty quantification in metric spaces.arXiv preprint arXiv:2405.05110, 2024

    Gábor Lugosi and Marcos Matabuena. Uncertainty quantification in metric spaces.arXiv preprint arXiv:2405.05110, 2024

  47. [56]

    Zhang, B

    Q. Zhang, B. Li, and L. Xue. Nonlinear sufficient dimension reduction for distribution-on-distribution regression. arXiv preprint arXiv:2207.04613, 2022

  48. [57]

    Bhattacharjee and H.-G

    S. Bhattacharjee and H.-G. Müller. Single index fréchet regression.arXiv preprint arXiv:2108.05437, 2021

  49. [58]

    Ghosal, M

    A. Ghosal, M. Matabuena, W. Meiring, and A. Petersen. Predicting distributional profiles of physical activity in the nhanes database using a partially linear single-index fréchet regression model. arXiv preprint arXiv:2302.07692, 2023

  50. [59]

    R. Qiu, Z. Yu, and Z. Lin. Semi-supervised fréchet regression.arXiv preprint arXiv:2403.10514, 2024

  51. [60]

    C. Schötz. The Fréchet Mean and Statistics in Non-Euclidean Spaces. PhD thesis, PhD thesis, 2021

  52. [61]

    S. Hanneke. Universally consistent online learning with arbitrarily dependent responses. In International Conference on Algorithmic Learning Theory, pages 488–497. PMLR, 2022

  53. [62]

    Bulté and H

    M. Bulté and H. Sørensen. Medoid splits for efficient random forests in metric spaces.arXiv preprint arXiv:2306.17031, 2023

  54. [63]

    Bhattacharjee, B

    S. Bhattacharjee, B. Li, and L. Xue. Nonlinear global fréchet regression for random objects via weak conditional expectation.arXiv preprint arXiv:2310.07817, 2023

  55. [64]

    Current state of commercial wearable technology in physical activity monitoring 2015–2017.International journal of exercise science, 11(7):503, 2018

    Jennifer A Bunn, James W Navalta, Charles J Fountaine, and Joel D Reece. Current state of commercial wearable technology in physical activity monitoring 2015–2017.International journal of exercise science, 11(7):503, 2018

  56. [65]

    Mobile devices and wearable technology for measuring patient outcomes after surgery: a systematic review

    Stephen R Knight, Nathan Ng, Athanasios Tsanas, Kenneth Mclean, Claudia Pagliari, and Ewen M Harrison. Mobile devices and wearable technology for measuring patient outcomes after surgery: a systematic review. NPJ digital medicine, 4(1):1–14, 2021

  57. [66]

    Wearablesandthemedicalrevolution

    JessilynDunn, RyanRunge, andMichaelSnyder. Wearablesandthemedicalrevolution. Personalized medicine, 15(5):429–448, 2018

  58. [67]

    Integrative omics for health and disease.Nature Reviews Genetics, 19(5):299–310, 2018

    Konrad J Karczewski and Michael P Snyder. Integrative omics for health and disease.Nature Reviews Genetics, 19(5):299–310, 2018

  59. [68]

    Network biology bridges the gaps between quantitative genetics and multi-omics to map complex diseases.Current Opinion in Chemical Biology, 66:102101, 2022

    Si Wu, Dijun Chen, and Michael P Snyder. Network biology bridges the gaps between quantitative genetics and multi-omics to map complex diseases.Current Opinion in Chemical Biology, 66:102101, 2022. 26

  60. [69]

    Glucodensities: a new representation of glucose profiles using distributional data analysis.Statistical Methods in Medical Research, 30(6):1445–1464, 2021

    Marcos Matabuena, Alexander Petersen, Juan C Vidal, and Francisco Gude. Glucodensities: a new representation of glucose profiles using distributional data analysis.Statistical Methods in Medical Research, 30(6):1445–1464, 2021

  61. [70]

    Fréchetanalysisofvarianceforrandomobjects

    ParomitaDubeyandHans-GeorgMüller. Fréchetanalysisofvarianceforrandomobjects. Biometrika, 106(4):803–821, 10 2019

  62. [71]

    Functional models for time-varying random objects

    Paromita Dubey and Hans-Georg Müller. Functional models for time-varying random objects. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 82(2):275–327, 2020

  63. [72]

    Image-to-image regression with distribution-free uncertainty quantification and applications in imaging

    Anastasios N Angelopoulos, Amit Pal Kohli, Stephen Bates, Michael Jordan, Jitendra Malik, Thayer Alshaabi, Srigokul Upadhyayula, and Yaniv Romano. Image-to-image regression with distribution-free uncertainty quantification and applications in imaging. InInternational Conferenc...

  64. [73]

    Tucker, Yichao Wu, and Hans-Georg Müller

    Danielle C. Tucker, Yichao Wu, and Hans-Georg Müller. Variable selection for global Fréchet regression. Journal of the American Statistical Association, 0(0):1–15, 2021

  65. [74]

    Extended comparisons of best subset selection, forward stepwise selection, and the lasso.arXiv preprint arXiv:1707.08692, 2017

    Trevor Hastie, Robert Tibshirani, and Ryan J Tibshirani. Extended comparisons of best subset selection, forward stepwise selection, and the lasso.arXiv preprint arXiv:1707.08692, 2017

  66. [75]

    What is a fenchel conjugate.Notices of the AMS, 59(1):44–46, 2012

    H Bauschke and Yves Lucet. What is a fenchel conjugate.Notices of the AMS, 59(1):44–46, 2012

  67. [76]

    Wainwright, and Laurent El Ghaoui

    Mert Pilanci, Martin J. Wainwright, and Laurent El Ghaoui. Sparse learning via boolean relaxations. Mathematical Programming, 151(1):63–87, 2015

  68. [77]

    On general minimax theorems.Pacific Journal of Mathematics, 8(1):171–176, 1958

    Maurice Sion. On general minimax theorems.Pacific Journal of Mathematics, 8(1):171–176, 1958

  69. [78]

    Fréchet regression for random objects with euclidean predictors

    Alexander Petersen and Hans-Georg Müller. Fréchet regression for random objects with euclidean predictors. The Annals of Statistics, 47(2):691–719, 2019

  70. [79]

    Chapman and Hall/CRC, 2021

    James Stephen Marron and Ian L Dryden.Object oriented data analysis. Chapman and Hall/CRC, 2021

  71. [80]

    Overview of object oriented data analysis.Biometrical Journal, 56(5):732–753, 2014

    J Steve Marron and Andrés M Alonso. Overview of object oriented data analysis.Biometrical Journal, 56(5):732–753, 2014

  72. [81]

    Data analysis on nonstandard spaces

    Stephan F Huckemann and Benjamin Eltzner. Data analysis on nonstandard spaces. Wiley Interdisciplinary Reviews: Computational Statistics, 13(3):e1526, 2021

  73. [82]

    On certain metric spaces arising from euclidean spaces by a change of metric and their imbedding in hilbert space.Annals of mathematics, pages 787–793, 1937

    Isaac J Schoenberg. On certain metric spaces arising from euclidean spaces by a change of metric and their imbedding in hilbert space.Annals of mathematics, pages 787–793, 1937

  74. [83]

    Metric spaces and positive definite functions.Transactions of the American Mathematical Society, 44(3):522–536, 1938

    Isaac J Schoenberg. Metric spaces and positive definite functions.Transactions of the American Mathematical Society, 44(3):522–536, 1938

  75. [84]

    Strong negative type in spheres.Pacific Journal of Mathematics, 307(2):383–390, 2020

    Russell Lyons. Strong negative type in spheres.Pacific Journal of Mathematics, 307(2):383–390, 2020

  76. [85]

    A new perspective on low-rank optimization

    Dimitris Bertsimas, Ryan Cory-Wright, and Jean Pauphilet. A new perspective on low-rank optimization. Mathematical Programming, 202(1):47–92, 2023

  77. [86]

    An update on the etiology and epidemiology of diabetes mellitus

    Ernest Adeghate, Peter Schattner, and Earl Dunn. An update on the etiology and epidemiology of diabetes mellitus. Annals of the New York academy of sciences, 1084(1):1–29, 2006

  78. [87]

    Aging, diabetes, and the public health system in the united states.American journal of public health, 102(8):1482–1497, 2012

    Carl J Caspersen, G Darlene Thomas, Letia A Boseman, Gloria LA Beckles, and Ann L Albright. Aging, diabetes, and the public health system in the united states.American journal of public health, 102(8):1482–1497, 2012

  79. [88]

    Global trends in diabetes complications: a review of current evidence.Diabetologia, 62:3–16, 2019

    Jessica L Harding, Meda E Pavkov, Dianna J Magliano, Jonathan E Shaw, and Edward W Gregg. Global trends in diabetes complications: a review of current evidence.Diabetologia, 62:3–16, 2019. 27

  80. [89]

    Marcos Matabuena, Juan C Vidal, Rahul Ghosal, and Jukka-Pekka Onnela. Deep learning framework with uncertainty quantification for survey data: Assessing and predicting diabetes mellitus risk in the american population.arXiv preprint arXiv:2403.19752, 2024

  81. [90]

    Diet and exercise in the prevention and treatment of type 2 diabetes mellitus.Nature Reviews Endocrinology, 16(10):545–555, 2020

    Faidon Magkos, Mads F Hjorth, and Arne Astrup. Diet and exercise in the prevention and treatment of type 2 diabetes mellitus.Nature Reviews Endocrinology, 16(10):545–555, 2020

  82. [91]

    Steffes, Edward Gregg, Frederick L

    Elizabeth Selvin, Michael W. Steffes, Edward Gregg, Frederick L. Brancati, and Josef Coresh. Performance of a1c for the classification and prediction of diabetes.Diabetes care, 34(1):84–89, Jan

  83. [92]

    Short-term variability in measures of glycemia and implications for the classification of diabetes.Archives of internal medicine, 167(14):1545–1551, 2007

    Elizabeth Selvin, Ciprian M Crainiceanu, Frederick L Brancati, and Josef Coresh. Short-term variability in measures of glycemia and implications for the classification of diabetes.Archives of internal medicine, 167(14):1545–1551, 2007

  84. [93]

    An overview of univariate and multivariate karhunen loève expansions in statistics.Journal of the Indian Society for Probability and Statistics, 23(2):285–326, 2022

    Ranadeep Daw, Matthew Simpson, Christopher K Wikle, Scott H Holan, and Jonathan R Bradley. An overview of univariate and multivariate karhunen loève expansions in statistics.Journal of the Indian Society for Probability and Statistics, 23(2):285–326, 2022

  85. [94]

    Variable selection for high dimensional multivariate outcomes

    Tamar Sofer, Lee Dicker, and Xihong Lin. Variable selection for high dimensional multivariate outcomes. Statistica Sinica, 24(4):1633, 2014

  86. [95]

    National health and nutrition examination survey: sample design, 2007-2010.Vital and Health Statistics

    Lester R Curtin, Leyla K Mohadjer, Sylvia M Dohrmann, Deanna Kruszon-Moran, Lisa B Mirel, Margaret D Carroll, Rosemarie Hirsch, Vicki L Burt, and Clifford L Johnson. National health and nutrition examination survey: sample design, 2007-2010.Vital and Health Statistics. Series ...

  87. [96]

    Dynamic human environmental exposome revealed by longitudinal personal monitoring.Cell, 175(1):277–291, 2018

    Chao Jiang, Xin Wang, Xiyan Li, Jingga Inlora, Ting Wang, Qing Liu, and Michael Snyder. Dynamic human environmental exposome revealed by longitudinal personal monitoring.Cell, 175(1):277–291, 2018

  88. [97]

    Digital health: tracking physiomes and activity using wearable biosensors reveals useful health-related information

    Xiao Li, Jessilyn Dunn, Denis Salins, Gao Zhou, Wenyu Zhou, Sophia Miryam Schüssler- Fiorenza Rose, Dalia Perelman, Elizabeth Colbert, Ryan Runge, Shannon Rego, et al. Digital health: tracking physiomes and activity using wearable biosensors reveals useful health-related infor...

  89. [98]

    Longitudinal personal DNA methylome dynamics in a human with a chronic condition

    Rui Chen, Lin Xia, Kailing Tu, Meixue Duan, Kimberly Kukurba, Jennifer Li-Pook-Than, Dan Xie, and Michael Snyder. Longitudinal personal DNA methylome dynamics in a human with a chronic condition. Nature medicine, 24(12):1930–1939, Dec 2018

  90. [99]

    Longitudinal multi-omics of host–microbe dynamics in prediabetes.Nature, 569(7758):663–671, 2019

    Wenyu Zhou, M Reza Sailani, Kévin Contrepois, Yanjiao Zhou, Sara Ahadi, Shana R Leopold, Martin J Zhang, Varsha Rao, Monika Avina, Tejaswini Mishra, et al. Longitudinal multi-omics of host–microbe dynamics in prediabetes.Nature, 569(7758):663–671, 2019

  91. [100]

    The function-on-scalar lasso with applications to longitudinal gwas.Electronic Journal of Statistics, 11(1):1351–1389, 2017

    Rina Foygel Barber, Matthew Reimherr, and Thomas Schill. The function-on-scalar lasso with applications to longitudinal gwas.Electronic Journal of Statistics, 11(1):1351–1389, 2017

  92. [101]

    Simultaneous variable selection and smoothing for high- dimensional function-on-scalar regression.Electronic Journal of Statistics, 12(2):4602–4639, 2018

    Alice Parodi and Matthew Reimherr. Simultaneous variable selection and smoothing for high- dimensional function-on-scalar regression.Electronic Journal of Statistics, 12(2):4602–4639, 2018

  93. [102]

    Optimal function-on-scalar regression over complex domains.Electronic Journal of Statistics, 17(1):156–197, 2023

    Matthew Reimherr, Bharath Sriperumbudur, and Hyun Bin Kang. Optimal function-on-scalar regression over complex domains.Electronic Journal of Statistics, 17(1):156–197, 2023

  94. [103]

    High-dimensional adaptive function-on-scalar regression

    Zhaohu Fan and Matthew Reimherr. High-dimensional adaptive function-on-scalar regression. Econometrics and Statistics, 1:167–183, 2017

  95. [104]

    Adaptive function-on-scalar regression with a smoothing elastic net

    Ardalan Mirshani and Matthew Reimherr. Adaptive function-on-scalar regression with a smoothing elastic net. Journal of Multivariate Analysis, 185:104765, 2021

  96. [105]

    PMID: 18779236

    Continuous glucose monitoring and intensive treatment of type 1 diabetes.New England Journal of Medicine, 359(14):1464–1476, 2008. PMID: 18779236. 28

  97. [106]

    Distributional data analysis via quantile functions and its application to modelling digital biomarkers of gait in Alzheimer’s disease.Biostatistics, 2021

    Rahul Ghosal, Vijay R Varma, Dmitri Volfson, Inbar Hillel, Jacek Urbanek, Jeffrey M Hausdorff, Amber Watts, and Vadim Zipunnikov. Distributional data analysis via quantile functions and its application to modelling digital biomarkers of gait in Alzheimer’s disease.Biostatistics, 2021

  98. [107]

    Functional data analysis for density functions by transformation to a Hilbert space.The Annals of Statistics, 44(1):183–218, 2016

    Alexander Petersen and Hans-Georg Müller. Functional data analysis for density functions by transformation to a Hilbert space.The Annals of Statistics, 44(1):183–218, 2016

  99. [108]

    Weinstock, Stephanie N

    Ruth S. Weinstock, Stephanie N. DuBose, Richard M. Bergenstal, Naomi S. Chaytor, Christina Peterson, Beth A. Olson, Medha N. Munshi, Alysa J.S. Perrin, Kellee M. Miller, Roy W. Beck, David R. Liljenquist, Grazia Aleppo, John B. Buse, Davida Kruger, Anuj Bhargava, Robin S. Gola...

  100. [109]

    Alexander Petersen, Xi Liu, and Afshin A. Divani. WassersteinF-tests and confidence bands for the Fréchet regression of density response curves.The Annals of Statistics, 49(1):590 – 611, 2021. 29

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.