Pith. sign in

REVIEW 2 major objections 4 minor 44 references

Spectrally Deconfounded Gradient Boosting

T0 review · 2 major / 4 minor · reviewed 2026-07-13 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Gradient boosting can be made robust to dense hidden confounding by a spectral loss plus early stopping, not by the spectral transform alone.

desk verdict Solid, usable spectral deconfounding for boosting: clean linear-base theory, mixed-model tuning, and strong scalability over spectral forests. read the letter →

arxiv 2607.09371 v1 pith:TY3MTYEZ submitted 2026-07-10 stat.ML cs.LG

Andrea Nava , Peter Bühlmann , Fabio Sigrist This is my paper
classification stat.MLcs.LG
keywords boostingcausalinferenceconfoundingrandomeffectsmixedmodelsspectraldeconfoundingearlystoppingLAVA
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

Flexible learners like gradient boosting often fit spurious associations created by unobserved confounders, so their fitted functions fail when the confounding structure changes. This paper shows how to bring spectral deconfounding—shrinking the high-variance directions of the design that dense confounders imprint—into nonlinear gradient boosting. The method replaces ordinary squared error with a spectral loss that slows learning along those directions; deconfounding then arises only when that slowdown is paired with regularization, especially intermediate early stopping. A mixed-model reading of the same loss supplies an empirical-Bayes rule for the shrinkage strength and extends the idea to general likelihoods and kernel-based nonlinear confounding. Simulations and a spatially confounded housing example show better recovery of the target function than ordinary boosting, at far lower cost than existing nonlinear spectral baselines.

What carries the argument

The spectral loss LW(f) = (1/2)‖W1/2(Y−f)‖2, with W acting diagonally in the left singular basis of X and down-weighting large singular values; it turns boosting into a recursion whose coordinate-wise rates νw_i open an intermediate stopping window that learns signal before confounding.

What would settle it

In simulations where the latent confounders affect only a few covariates, or where the true signal itself lies in the same top singular directions as the confounders, spectral boosting with early stopping should lose its advantage over ordinary boosting on estimation error of the target function.

Watch

Extended reading notes

Core claim

Under dense hidden confounding, gradient boosting with a spectral (LAVA / mixed-model) loss creates direction-dependent learning rates that suppress confounding-aligned directions; combined with an intermediate early-stopping regime, this recovers the target function asymptotically unbiasedly for linear base learners and, empirically for trees, improves estimation of that function relative to ordinary boosting.

Load-bearing premise

Confounding must be dense enough that its fitted effect concentrates in the leading high-variance directions of the covariates while the true signal does not; if that spectral separation fails, the early-stopping window that deconfounds disappears.

Editorial extensions

If this is right

  • Ordinary L2-boosting remains biased under dense confounding no matter how long it is run; spectral weighting plus intermediate stopping is required for asymptotic unbiasedness.
  • Empirical Bayes on the mixed-model variance ratio can replace fixed spectral rules and remains usable when the design spectrum is polluted by noise unrelated to confounding.
  • The same spectral-loss idea extends, via Laplace approximation and kernel random effects, to non-Gaussian outcomes and to nonlinear confounding beyond the linear factor model.
  • Cross-validation that scores the BLUP of the dense random effect selects stopping times close to an oracle that knows the true target, avoiding the overfit that ordinary validation encourages.
  • Relative to spectrally deconfounded random forests, the boosting implementation is orders of magnitude faster, making spectral deconfounding practical for larger n.

Reading between the lines

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

  • The same spectral-loss-plus-early-stopping logic may transfer to other pathwise learners (forward stagewise, coordinate descent) whose dynamics favor high-variance directions.
  • When signal and confounder share the same high-variance subspace, practitioners may need an auxiliary separation step (instrument, environment shift, or explicit spatial kernel) before spectral boosting can help.
  • Kernel spectral boosting offers a practical diagnostic: if an RBF random effect improves target recovery while a linear one does not, the residual confounding is likely nonlinear in the observed covariates.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The paper develops spectral deconfounding for gradient boosting under dense hidden confounding. It replaces the ordinary squared loss by a spectral (LAVA / mixed-model whitening) loss so that residuals are filtered by F^T F before tree fitting, thereby slowing learning in high-variance directions of X that carry dense confounding. For linear OLS base learners, Theorem 3.1 shows that spectral weights plus an intermediate early-stopping window yield asymptotic unbiasedness for the target signal f0 under spectral separation assumptions; ordinary L2-boosting does not. The spectral loss is derived from a Gaussian random-effects model, which yields an empirical-Bayes rule for the variance components and a BLUP-corrected CV rule for the number of iterations. The framework is extended to non-Gaussian likelihoods via Laplace approximation and to nonlinear confounding via kernel random effects. Simulations and a Boston housing spatial-confounding case study show improved recovery of f0 relative to ordinary boosting and much better scalability than spectrally deconfounded random forests.

Significance. If the claims hold, the paper gives a practical and theoretically motivated route to transfer linear spectral deconfounding and mixed-model adjustment to modern nonlinear learners. Strengths include: (i) a clean optimization-path analysis for OLS base learners (Theorem 3.1 with factor-model verification in Appendix B); (ii) an explicit mixed-model / LAVA link (Proposition 2.1) that justifies empirical-Bayes tuning; (iii) a scalable GPBoost implementation with code for reproduction; and (iv) systematic experiments that vary n, p, q, loading sparsity, design misspecification, and nonlinear confounding, plus a real-data spatial example. The main limitation the authors already flag—that exact diagonalization fails for trees—is handled honestly by treating the tree case as empirical. The contribution is incremental relative to linear spectral deconfounding and SDF, but the boosting formulation, EB/BLUP tuning, and non-Gaussian/kernel extensions are useful and well executed for the field.

major comments (2)
  1. [Abstract; §1; Theorem 3.1; end of §3] Theorem 3.1 and Corollary 3.2 establish asymptotic unbiasedness only for OLS base learners under Assumptions 1–3 (spectral concentration of g0 and weight separation). The manuscript is explicit that trees lose exact diagonalization (end of §3), so the central nonlinear claim rests on simulations (§6) and the Boston study (§7). That is acceptable if framed as such, but the abstract and introduction currently present “spectrally deconfounded gradient boosting” as achieving deconfounding for nonlinear models without always distinguishing the theorem’s scope from the empirical tree results. Please tighten those claims so that the load-bearing theory is not over-read as covering trees.
  2. [§3 Theorem 3.1 conditions (i)–(iii); §4.2.2; Figure 5; §7] The intermediate stopping window (mp ν wM → ∞ and mp ν wH → 0) is essential: spectral loss alone does not remove confounding if boosting is run indefinitely. Figure 5 shows that BLUP-corrected CV tracks an oracle that minimizes error to f0, which is strong evidence under the simulation design. For the real-data claim, however, there is no ground-truth f0, so it remains unclear whether the selected stopping time is in the deconfounding regime rather than a pure predictive regime. A short sensitivity analysis (e.g., partial-dependence stability vs. iteration count, or comparison to a held-out spatial reference under different M) would make the practical stopping claim more load-bearing.
minor comments (4)
  1. [§2.4; Eq. (14); Algorithm 2] In §2.4 / Proposition 2.1 the identification nλ2 = σe²/σr² is clear; it would help readers if Algorithm 2 and the LAVA weight formula (14) cross-referenced this identification in one place.
  2. [§6 figures] Figures 3–10 use log(1+x) MSE scales; a brief note in captions that absolute levels of MSEf depend on the Fourier signal scale would aid interpretation across panels.
  3. [Eq. (18); §4.1] The full-space definition of F in (18) is important for trees; a one-sentence reminder that the orthogonal complement is left unshrunk would reduce confusion with the pure col(X) analysis of §3.
  4. [Throughout; §5.3] Minor typos and notation: “LAVA-type” vs “LAVA” is used inconsistently; arXiv id and package versions in the software paragraph are fine but could list the exact GPBoost version used for the reported runs.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: pathwise bias theorem and mixed-model spectral loss are derived from stated assumptions, not fitted tautologies.

full rationale

The load-bearing theoretical claim (Theorem 3.1) is a pathwise bias analysis for OLS base learners under explicit Assumptions 1–3 (reduced model, spectral concentration of g0, weight separation). The boosting recursion is diagonalized in the left singular basis and the intermediate-stopping window is derived from the coordinatewise rates (1−νwi)^m; this does not reduce the target f0 to a fitted constant. Proposition 2.1 equates LAVA whitening with the Gaussian mixed-model operator via Sherman–Morrison–Woodbury—an algebraic identity, not a circular definition of the estimand. Empirical Bayes tunes (σr2,σe2) and BLUP-corrected CV chooses M, but evaluation is MSEf against the ground-truth causal function on held-out data (and a spatial reference model on Boston housing), so the reported gains are not forced by the training criterion. Self-citations (GPBoost, Laplace/quantile extensions) support software and non-Gaussian extensions and are not used as uniqueness theorems that forbid alternatives. Dense-confounding and signal–factor separation are stated as assumptions and probed in simulations; they are weak points of applicability, not circular reductions. Overall the derivation chain is self-contained against external benchmarks.

Assumptions & free parameters 5 free parameters · 6 assumptions · 2 invented entities

Central claims rest on the dense-factor structural model, spectral concentration of bias, algorithmic weight separation, and the mixed-model / Laplace approximations used for tuning and non-Gaussian extensions. Free parameters are the usual boosting knobs plus EB variance components; no new physical entities are invented.

free parameters (5)
  • boosting learning rate ν
    Fixed (e.g. 0.05 in experiments); interacts with iteration count in the intermediate stopping regime of Theorem 3.1.
  • number of boosting iterations M (early stopping)
    Selected by K-fold CV with BLUP-corrected predictions; load-bearing for deconfounding because spectral weights alone do not remove bias if M→∞.
  • random-effect variance components (σ_r², σ_e²) / LAVA λ₂
    Estimated by empirical Bayes maximizing the marginal likelihood; set the spectral weights w_i. Fixed asymptotic rule of Čevid et al. is also compared.
  • tree depth / base-learner complexity
    Hyperparameter of the nonlinear method (depth 3 in qualitative experiments); not covered by the linear-base-learner theory.
  • kernel hyperparameters θ (RBF length scale etc.)
    For nonlinear confounding extension; estimated by EB or median heuristic in experiments.
assumptions (6)
  • domain assumption Dense confounding: loadings satisfy d_min(Γ) ≳ √p so bias is spread and concentrates in leading singular directions of X.
    Stated in §2.3 and used throughout theory and method motivation; verified under factor model in Appendix B.
  • domain assumption Structural SEM X=HΓ+E, Y=f0(X)+Hδ+ε with independence of E, H, ε as stated.
    Equations (1)–(2) and (15)–(16); defines the target f0 versus confounding.
  • domain assumption Spectral separation: confounding g0 lies in top-q left singular subspace; signal projection onto that subspace vanishes (Assumption 2).
    Required for Theorem 3.1; justified via strong-factor PCA / Davis–Kahan arguments in Appendix B.
  • ad hoc to paper Weight separation for LAVA/Trim: w_H→0 on spiked directions, w_M bounded away from 0 on bulk (Assumption 3).
    Algorithmic condition that creates the intermediate stopping window; follows from spike–bulk singular-value behavior under stated random-matrix assumptions.
  • domain assumption Gaussian random-effect model for dense bias (or Laplace approximation thereof for general likelihoods).
    §2.4, §4, §5.1; identifies LAVA with mixed-model whitening and enables EB tuning.
  • standard math Standard linear algebra / random-matrix / factor-model results (SVD, Woodbury, Bai factor recovery, Vershynin operator norms).
    Used in proofs and Appendix B without re-derivation.
invented entities (2)
  • Spectral loss for gradient boosting (FᵀF-filtered residuals as pseudo-response) independent evidence
    purpose: Embed spectral deconfounding into stagewise tree boosting so confounding-aligned directions are learned more slowly.
    Core methodological object; not a physical entity but a new algorithmic construct relative to prior linear spectral work.
  • BLUP-corrected cross-validation for stopping under spectral / mixed-model boosting independent evidence
    purpose: Select iteration count so CV does not reward absorbing confounding into f̂.
    Tuning device derived from the mixed-model view; validated against oracle f0 error in simulations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Spectrally Deconfounded Gradient Boosting." pith.science (2026). https://pith.science/paper/TY3MTYEZ

@misc{pith2026260709371,
  author       = {Pith},
  title        = {Pith review of: Spectrally Deconfounded Gradient Boosting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TY3MTYEZ}},
  note         = {Machine review of arXiv:2607.09371}
}
read the original abstract

Flexible machine-learning methods can be sensitive to hidden confounding: they may learn associations induced by unobserved confounders rather than stable signals. Spectral deconfounding mitigates this problem by shrinking high-variance directions of the covariate matrix that, under dense confounding, carry latent confounder information. Existing work has largely focused on linear models. We develop a nonlinear spectral deconfounding framework for gradient boosting. Our approach replaces the ordinary squared-error loss by a spectral loss, which alters the boosting dynamics by slowing down learning in confounding-aligned directions. We show that deconfounding is not achieved by the spectral loss alone, but by the interaction between spectral shrinkage and regularization, especially in terms of early stopping. Moreover, we provide a mixed-model interpretation that connects LAVA-type shrinkage to random-effects adjustment and yields an empirical-Bayes procedure for tuning the spectral loss. We also extend the method to general likelihoods and nonlinear confounding using Laplace approximations and kernel random effects. Across synthetic and real-world experiments, spectrally deconfounded boosting improves estimation of the target function under hidden confounding and is substantially more scalable than existing nonlinear spectral deconfounding baselines.

Figures

Figures reproduced from arXiv: 2607.09371 by the authors.

Figure 1
Figure 1. Regression setting. Left: Log–log scatter of split-based variable importance; causal features [PITH_FULL_IMAGE:figures/full_fig_p015_1.png] view at source ↗
Figure 2
Figure 2. Classification setting. Left: Log–log scatter of split-based variable importance; causal [PITH_FULL_IMAGE:figures/full_fig_p016_2.png] view at source ↗
Figure 3
Figure 3. Regression setting. Boxplots of the mean squared error for the different methods across [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figures from the paper (24 more)
Figure 4
Figure 4. Figure 4: Regression setting. Boxplots of the runtime for the different methods across simulation [PITH_FULL_IMAGE:figures/full_fig_p017_4.png]
Figure 5
Figure 5. Figure 5: Regression setting. Number of boosting iterations selected by the proposed cross-validation [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]
Figure 6
Figure 6. Figure 6: Regression setting. Boxplots of the estimated random-effect variance [PITH_FULL_IMAGE:figures/full_fig_p018_6.png]
Figure 7
Figure 7. Figure 7: Classification setting. Boxplots of the mean squared error for the different methods across [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: Robustness to violations of the dense-confounding assumption in the regression setting. The [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]
Figure 9
Figure 9. Figure 9: Design-misspecified regression setting. Comparison of empirical-Bayes spectral tuning with [PITH_FULL_IMAGE:figures/full_fig_p021_9.png]
Figure 10
Figure 10. Figure 10: Well-specified regression setting. Comparison of empirical Bayes tuning of the spectral [PITH_FULL_IMAGE:figures/full_fig_p021_10.png]
Figure 11
Figure 11. Figure 11: Nonlinear outcome confounding. Performance as the confounding term becomes increasingly [PITH_FULL_IMAGE:figures/full_fig_p023_11.png]
Figure 12
Figure 12. Figure 12: Nonlinear factor model. Left: spiral manifold [PITH_FULL_IMAGE:figures/full_fig_p023_12.png]
Figure 13
Figure 13. Figure 13: Regression setting with nonlinear factor model and linear outcome confounding. Left: test [PITH_FULL_IMAGE:figures/full_fig_p024_13.png]
Figure 14
Figure 14. Figure 14: Boston housing data. Spatial distribution of median house prices (left) and of the first [PITH_FULL_IMAGE:figures/full_fig_p025_14.png]
Figure 15
Figure 15. Figure 15: Boston housing data. Partial dependence plots for the six features with highest permutation [PITH_FULL_IMAGE:figures/full_fig_p026_15.png]
Figure 16
Figure 16. Figure 16: Boston housing data. Predicted random effects from the spatial boosting reference model [PITH_FULL_IMAGE:figures/full_fig_p027_16.png]
Figure 17
Figure 17. Figure 17: Classification setting. Number of boosting iterations selected by the proposed cross [PITH_FULL_IMAGE:figures/full_fig_p037_17.png]
Figure 18
Figure 18. Figure 18: Classification setting. Boxplots of the estimated random-effect variance [PITH_FULL_IMAGE:figures/full_fig_p038_18.png]
Figure 19
Figure 19. Figure 19: Classification setting. Boxplots of the runtime for the different methods across simulation [PITH_FULL_IMAGE:figures/full_fig_p038_19.png]
Figure 20
Figure 20. Figure 20: Boxplots of the mean squared error for the different methods across simulation scenarios, [PITH_FULL_IMAGE:figures/full_fig_p038_20.png]
Figure 21
Figure 21. Figure 21: Boxplots of the runtime for the different methods across simulation scenarios, displayed on [PITH_FULL_IMAGE:figures/full_fig_p039_21.png]
Figure 22
Figure 22. Figure 22: Number of boosting iterations selected by the proposed cross-validation rule (solid lines) [PITH_FULL_IMAGE:figures/full_fig_p039_22.png]
Figure 23
Figure 23. Figure 23: Boxplots of the estimated random-effect variance [PITH_FULL_IMAGE:figures/full_fig_p039_23.png]
Figure 24
Figure 24. Figure 24: Boxplots of the mean squared error for the different methods across simulation scenarios, [PITH_FULL_IMAGE:figures/full_fig_p040_24.png]
Figure 25
Figure 25. Figure 25: Boxplots of the runtime for the different methods across simulation scenarios, displayed on [PITH_FULL_IMAGE:figures/full_fig_p040_25.png]
Figure 27
Figure 27. Figure 27: Boxplots of the estimated random-effect variance [PITH_FULL_IMAGE:figures/full_fig_p040_27.png]
Figure 26
Figure 26. Figure 26: Number of boosting iterations selected by the proposed cross-validation rule (solid lines) [PITH_FULL_IMAGE:figures/full_fig_p041_26.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

44 extracted references · 2 linked inside Pith

  1. [1]

    Journal of Machine Learning Research , volume=

    Spectral deconfounding via perturbed sparse linear models , author=. Journal of Machine Learning Research , volume=

  2. [2]

    Journal of Computational and Graphical Statistics , volume =

    Markus Ulmer and Cyrill Scheidegger and Peter Bühlmann , title =. Journal of Computational and Graphical Statistics , volume =. 2026 , publisher =

    2026

  3. [3]

    The Annals of Statistics , year =

    Chernozhukov, Victor and Hansen, Christian and Liao, Yuan , title =. The Annals of Statistics , year =

  4. [4]

    Journal of the American Statistical Association , volume=

    The blessings of multiple causes , author=. Journal of the American Statistical Association , volume=. 2019 , publisher=

    2019

  5. [5]

    Journal of Machine Learning Research , volume=

    Naive regression requires weaker assumptions than factor models to adjust for multiple cause confounding , author=. Journal of Machine Learning Research , volume=

  6. [6]

    Econometrica , volume=

    Inferential theory for factor models of large dimensions , author=. Econometrica , volume=. 2003 , publisher=

    2003

  7. [7]

    Journal of Machine Learning Research , volume=

    Boosting as a regularized path to a maximum margin classifier , author=. Journal of Machine Learning Research , volume=

  8. [8]

    2019 , eprint=

    A Continuous-Time View of Early Stopping for Least Squares , author=. 2019 , eprint=

    2019

Show all 44 references
  1. [9]

    The Annals of Mathematical Statistics , volume=

    The empirical Bayes approach to statistical decision problems , author=. The Annals of Mathematical Statistics , volume=. 1964 , publisher=

    1964

  2. [10]

    1996 , organization=

    Experiments with a new boosting algorithm , author=. 1996 , organization=

    1996

  3. [11]

    Spectral Deconfounding for High-Dimensional Sparse Additive Models , year =

    Scheidegger, Cyrill and Guo, Zijian and B\". Spectral Deconfounding for High-Dimensional Sparse Additive Models , year =. ACM / IMS J. Data Sci. , month = feb, articleno =

  4. [12]

    Nature Genetics , volume=

    Mixed linear model approach adapted for genome-wide association studies , author=. Nature Genetics , volume=. 2010 , publisher=

    2010

  5. [13]

    PLoS Genetics , volume=

    Population structure in genetic studies: Confounding factors and mixed models , author=. PLoS Genetics , volume=. 2018 , publisher=

    2018

  6. [14]

    arXiv preprint arXiv:2508.06274 , year=

    Latent confounding in high-dimensional nonlinear models , author=. arXiv preprint arXiv:2508.06274 , year=

    arXiv

  7. [15]

    Nature Genetics , volume=

    Principal components analysis corrects for stratification in genome-wide association studies , author=. Nature Genetics , volume=. 2006 , publisher=

    2006

  8. [16]

    Invariance, Causality and Robustness: 2018

    Peter Bühlmann , journal =. Invariance, Causality and Robustness: 2018

    2018

  9. [17]

    Journal of the American Statistical Association , volume=

    Fully exponential Laplace approximations to expectations and variances of nonpositive functions , author=. Journal of the American Statistical Association , volume=. 1989 , publisher=

    1989

  10. [18]

    PloS one , volume=

    Correcting for population structure and kinship using the linear mixed model: theory and extensions , author=. PloS one , volume=. 2013 , publisher=

    2013

  11. [19]

    Version 0.11 , volume=

    Package ‘ranger’ , author=. Version 0.11 , volume=

  12. [20]

    Journal of Machine Learning Research , volume=

    Gaussian process boosting , author=. Journal of Machine Learning Research , volume=

  13. [21]

    Nature , volume=

    Genes mirror geography within Europe , author=. Nature , volume=. 2008 , publisher=

    2008

  14. [22]

    The Journal of Economic Education , volume=

    A simple proof of the FWL theorem , author=. The Journal of Economic Education , volume=. 2008 , publisher=

    2008

  15. [23]

    SIAM Review , volume =

    Updating the inverse of a matrix , author =. SIAM Review , volume =

  16. [24]

    Matrix Algebra From a Statistician's Perspective , author =

  17. [25]

    Large sample analysis of the median heuristic , journal =

    Garreau, Damien and Jitkrittum, Wittawat and Kanagawa, Motonobu , year =. Large sample analysis of the median heuristic , journal =

  18. [26]

    2021 , url =

    Fabio Sigrist and Tim Gyger and Pascal Kündig , note =. 2021 , url =

    2021

  19. [27]

    Journal of Agricultural, Biological and Environmental Statistics , publisher=

    Katzfuss, Matthias and Guinness, Joseph and Gong, Wenlong and Zilber, Daniel , year=. Journal of Agricultural, Biological and Environmental Statistics , publisher=

  20. [28]

    2005 , author =

    A Unifying View of Sparse Approximate Gaussian Process Regression , journal =. 2005 , author =

    2005

  21. [29]

    Boosting with the

    B. Boosting with the. Journal of the American Statistical Association , volume=. 2003 , publisher=

    2003

  22. [30]

    Biometrika , volume=

    A useful variant of the Davis--Kahan theorem for statisticians , author=. Biometrika , volume=. 2015 , publisher=

    2015

  23. [31]

    University of California, Irvine , volume=

    High-dimensional probability , author=. University of California, Irvine , volume=

  24. [32]

    Annals of Statistics , volume=

    Doubly debiased lasso: High-dimensional inference under hidden confounding , author=. Annals of Statistics , volume=

  25. [33]

    Neural Information Processing Systems Datasets and Benchmarks Track , year=

    Why do tree-based models still outperform deep learning on tabular data? , author=. Neural Information Processing Systems Datasets and Benchmarks Track , year=

  26. [34]

    2026 , url =

    Laplace Approximations for Mixed-Effects and Gaussian Process Quantile Regression , author=. 2026 , url =. 2605.21197 , archivePrefix=

    2026 arXiv

  27. [35]

    2024 , url =

    mlbench: Machine Learning Benchmark Problems , author =. 2024 , url =

    2024

  28. [36]

    1998 , organization =

    UCI Repository of Machine Learning Databases , author =. 1998 , organization =

    1998

  29. [37]

    Forward stagewise regression and the monotone lasso , volume=

    Hastie, Trevor and Taylor, Jonathan and Tibshirani, Robert and Walther, Guenther , year=. Forward stagewise regression and the monotone lasso , volume=. Electronic Journal of Statistics , publisher=

  30. [38]

    Least angle regression , volume=

    Efron, Bradley and Hastie, Trevor and Johnstone, Iain and Tibshirani, Robert , year=. Least angle regression , volume=. The Annals of Statistics , publisher=

  31. [39]

    Latent Gaussian Model Boosting , year=

    Sigrist, Fabio , journal=. Latent Gaussian Model Boosting , year=

  32. [40]

    Annals of Statistics , pages=

    Greedy function approximation: a gradient boosting machine , author=. Annals of Statistics , pages=. 2001 , publisher=

    2001

  33. [41]

    The Annals of Statistics , volume=

    Additive logistic regression: a statistical view of boosting (with discussion and a rejoinder by the authors) , author=. The Annals of Statistics , volume=. 2000 , publisher=

    2000

  34. [42]

    The Annals of Statistics , pages=

    Factor modeling for high-dimensional time series: inference for the number of factors , author=. The Annals of Statistics , pages=. 2012 , publisher=

    2012

  35. [43]

    Journal of Econometrics , volume=

    High dimensional covariance matrix estimation using a factor model , author=. Journal of Econometrics , volume=. 2008 , publisher=

    2008

  36. [44]

    The Annals of Statistics , number =

    Leo Breiman , title =. The Annals of Statistics , number =

Pith tools

Reviewed July 13, 2026 · model on record in the stance chip above.