Upper Confidence Bounds for the Prediction Error of Kernel Ridge Regression via Gaussian Refitting

Xiaoming Huo; Yijin Ni

REVIEW 3 major objections 2 minor 45 references

Upper Confidence Bounds for the Prediction Error of Kernel Ridge Regression via Gaussian Refitting

T0 review · 3 major / 2 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read Kernel-ridge fits can be certified by a Gaussian-refit upper confidence bound that contracts at the minimax rate and needs no noise moments.

desk verdict Genuinely new mechanism and rate-optimal bound, but the 'finite-sample 1−α UCB' slogan outruns Theorem 4.1, which carries an uncomputable delocalization slack. read the letter →

arxiv 2607.28846 v1 pith:QSINAHEL submitted 2026-07-30 stat.ME

Yijin Ni , Xiaoming Huo This is my paper · ORCID

classification stat.ME MSC 62G0862G09

keywords upperconfidenceboundkernelridgeregressionrealizedpredictionerrorGaussianrefittingorderstatisticcalibrationwildbootstrapcross-validationfloorminimaxrate

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

The paper tries to establish that a single kernel-ridge-regression fit can come with a computable, finite-sample upper confidence bound on the distance between the fitted function and the unknown truth, measured as average squared error at the design points. The proposed bound is built by adding Gaussian noise to the residuals, refitting, and recording how far the fit moves; an order statistic over repeated refits supplies the tail quantile, and a bias term is added before squaring. The authors claim this bound is valid under only symmetric noise—no finite moments, no known variance—provided the target lies in a known reproducing-kernel-Hilbert-space ball, and that it contracts at the minimax rate n^{-2s/(2s+1)}, matching the prediction error itself. They also argue cross-validation intervals cannot do this: any interval computed from held-out losses is stuck at an n^{-1/2} margin floor, so its margin-to-error ratio diverges whenever the fit converges faster than n^{-1/2}.

What carries the argument

The central object is the calibrated Gaussian movement query: draw xi ~ N(0, I_n), refit at y + xi o a, and record m(a) = ||M(y + xi o a) - M(y)||_n, which equals ||H(xi o a)||_n for the linear smoother. The construction rests on a Gaussian comparison inequality: enlarging any coordinate of a cannot lower the upper quantiles of the movement, so a computable envelope a_i >= |w_i| dominates the unknown noise magnitudes. An order statistic of L movements estimates the quantile, and a bias input b bounds ||(H - I) f*||_n; the final bound is (quantile + b)^2. This same mechanism delivers both the finite-sample tail guarantee and the minimax-rate contraction.

What would settle it

A single simulation with the true function inside the RKHS ball, symmetric noise, and the certified worst-case envelope, where empirical coverage falls below the nominal level while the paper's delocalization functionals are small, would refute the conditional validity claim. Equivalently, a construction of two problems whose held-out loss distributions are nearly indistinguishable but whose fit errors differ by order n^{-1/2}, together with any measurable held-out-loss bound with a smaller margin, would falsify the holdout floor.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that replacing sign-flip refitting with Gaussian refitting makes the wild-refit mechanism computable and rate-sharp for kernel ridge regression. Because Gaussian noise is monotone under coordinatewise enlargement of an envelope vector a >= |w|, the quantiles of the refit movement ||H(xi o a)||_n can be used as a conservative proxy for the unobservable noise term ||H w||_n; the bound is bU_alpha = (q_{1-alpha}(a) + b)^2. The paper proves conditional validity with an explicit remainder depending only on delocalization functionals and the number of refits, and it shows the bound attains the minimax rate. By contrast, sign multipliers degenerate to the

Load-bearing premise

The certified bound presupposes the true function lies inside the known reproducing-kernel-Hilbert-space ball with a known kernel bound; if the target is outside that ball—as in the paper's rigid over-smoothed misspecification experiment, where coverage fell to 0.10—the guarantee collapses.

Editorial extensions

If this is right

A user can report a finite-sample 95% upper bound on the realized prediction error of a kernel-ridge fit using only fair noise signs; no variance estimate or moment condition is needed.
The bound's width tracks the actual error: under polynomial spectral decay it contracts at n^{-2s/(2s+1)}, the minimax rate, rather than being pinned at n^{-1/2}.
Cross-validation intervals, including nested and quantile-repaired versions, inherit an n^{-1/2} margin floor; whenever smoothness s > 1/2, their margin-to-error ratio diverges polynomially.
A single set of L refits calibrates every confidence level at once, because all levels read off the same ordered movements.
The construction extends empirically to nonlinear constrained fits and real spatial data with full coverage in the reported experiments.

Reading between the lines

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

Editorial inference: If the conditional-symmetry argument transfers to any firmly non-expansive estimator with a computable envelope—as the paper's constrained-fit experiment suggests—the same refit recipe could certify other black-box predictors, not just linear smoothers.
Editorial inference: The n^{-1/2} holdout floor implies that any conformal or split-based uncertainty method targeting the fitted function's realized error, rather than a future observation, will face the same precision barrier; the Gaussian-refit path is one way to route around it.
Editorial inference: A data-dependent spectral envelope, rather than the worst-case one, might close the gap on slowly decaying kernels where the bound stays valid but loose; this is a natural next test.
Editorial inference: Because validity is conditional on noise magnitudes, an adversarial magnitude pattern that concentrates noise on a few coordinates will make the bound vacuous; stress-testing that regime could map the method's practical limits.

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.

Desk Editor's Note

read the letter

The Gaussian-refit construction is the real thing. Replacing Rademacher signs with Gaussian multipliers to get envelope monotonicity via Anderson's inequality is a clever and, as far as I can tell, new mechanism, and the sign-multiplier impossibility result plus the holdout-loss floor are solid contributions on their own. The proof chain is mostly explicit and the authors are honest about the worst-case envelope being loose and about the misspecification regime where coverage collapses to 0.10. The empirical work is careful and the comparisons are fair. This deserves a serious referee.

The soft spot is exactly where the stress-test note points. Theorem 4.1 proves a conditional inequality with an additive slack of C0/φ(z1−α)(δ+δ_N)+C1/√L, and δ, δ_N depend on the unobservable noise magnitudes. So the procedure does not deliver a certified finite-sample 1−α bound at a user-specified α unless you're willing to assume the delocalization functionals are small, which is an unverified condition. The paper's own scope caveat says validity relies on sufficiently delocalized noise, but the abstract's "computable tail bound" and the phrase "calibrated at any confidence level" overstate what is proven. This is a gap in the headline claim, not a hidden fatal flaw — the bound is still useful and likely conservative in practice — but it should be fixed in the wording or mitigated with a data-dependent check, even a crude one.

Smaller issues: the worst-case envelope needs known B and κ, and is quite loose for slow spectral decay; the data-driven envelope used in the experiments is explicitly empirical, so the finite-sample guarantee applies only to the worst-case version; and the code isn't archived in the preprint, which hurts reproducibility. The bias input b for the theoretical bound is fine, and the g>0 quantile inversion point is minor as the reader notes.

For whom: people working on wild refitting, resampling-based uncertainty, or honest inference for kernel methods. I'd send it to review and would engage with it myself; the delocalization-slack issue needs to be addressed, but the core idea is worth publishing and building on.

Referee Report

3 major / 2 minor

Summary. The paper proposes a Gaussian-refit upper confidence bound for the empirical prediction error of kernel ridge regression under fixed design and symmetric noise. The bound is bU_α = (q_{1−α}(a) + b)^2, where q_{1−α}(a) is an order statistic of L refit movements with Gaussian multipliers at an envelope a, and b bounds the smoothing bias. For the theoretical worst-case envelope (18), the paper claims a computable finite-sample upper confidence bound at level 1−α (Theorem 4.1), contraction at the minimax rate (Theorem 4.8), and fundamental limitations of held-out-loss intervals (Propositions 4.10 and 4.12). For a practical data-driven envelope (20), the paper reports full coverage within small multiples of the true error quantile across synthetic and real data, including heavy-tailed noise.

Significance. If the stated guarantees held, the paper would be an important contribution: a computable tail bound for the realized error of a kernel-ridge fit under essentially no moment assumptions, with a matching minimax rate and strong empirical tightness. The paper is transparent about its assumptions and limitations, and it ships explicit proofs, reproducible code, and a thorough experimental comparison. The negative results on cross-validation margins (Propositions 4.10 and 4.12) are interesting and appear sound. However, the central validity theorem as stated does not deliver the advertised finite-sample 1−α bound, and the proof has a gap in the Monte-Carlo calibration step. These issues are load-bearing for the headline contribution and need to be addressed before the paper can be accepted.

major comments (3)

[Theorem 4.1 / Appendix A.7, Step 1; Eq. (16)] The proof treats the sample order statistic q_{1−α}(a) as an upper bound on the population quantile Q0 up to O(L^{−1/2}), then uses the inclusion {E(f̂)>bU_α} ⊆ {T>Q0}. This is not justified. The DKW inequality gives |F_L−F|≤ε with high probability, which implies q_{1−α}(a) ≥ Q0−ε, not q_{1−α}(a) ≥ Q0. In fact P(q_{1−α}(a) < Q0) is of constant order, so the set inclusion fails on a non-vanishing probability set. The O(L^{−1/2}) slack in (24) therefore does not follow from the argument given. A correct treatment requires either a DKW-corrected quantile (e.g., a more conservative order statistic with an explicit margin) or a direct averaged comparison of P(T>q) with P(T>Q0). As written, the proof of Theorem 4.1 is incomplete.
[Abstract, §1 Eq. (6), and Theorem 4.1 (24)] The paper claims a computable finite-sample upper confidence bound at level 1−α under symmetric noise. Theorem 4.1, however, proves P(E(f̂)>bU_α | |w|) ≤ α + C0 φ(z_{1−α})^{−1}(δ+δ_N) + C1/√L, where δ, δ_N are defined in (23) from M = diag(|w|)HᵀH diag(|w|) and are not computable from data. For fixed n and L, the slack is unknown, so the user-specified level 1−α is not certified. The discussion after (24) explicitly says the right-hand side tends to α only when δ,ρ→0 and L→∞, which is an asymptotic statement. To claim a finite-sample level, the authors must either (i) provide a conservative calibration using universal bounds on C0, C1 and a more conservative nominal level, or (ii) explicitly relabel the result as an asymptotic or approximate bound. The current abstract and (6) overstate what is proved.
[§4.1, 'Whenever δ→0 and ρ→0...'] The paper's own statement that the bound is useful only in a delocalized regime is honest, but the functionals δ and ρ cannot be verified from data. This reinforces the previous comment: the finite-sample validity claim is conditional on unverifiable structure. The paper should either give a data-dependent certificate for delocalization (which seems difficult) or clearly state that the theoretical guarantee is asymptotic in the delocalization limit and that the finite-sample guarantee is empirical only.

minor comments (2)

[Abstract and §6 vs. Table 4] The abstract states the bound 'maintains full coverage within twice the true 95% error quantile'. This is supported for the synthetic design (Table 2, 1.7–1.9×) but not for the real elevation study (Table 4, 2.4–3.4×). Please qualify the claim to 'within a small constant' or specify that the factor-of-two holds on the synthetic design.
[§3.4, Eq. (20); §6.2] The data-driven envelope uses a pilot penalty λ/20, the only hand-tuned parameter. Its sensitivity is not discussed. A brief robustness check over this factor would strengthen the empirical claims.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular structure: the certified bound's envelope and bias are proven upper bounds, and its quantile is independently simulated; the delocalization slack is an explicit limitation, not a self-referential fit.

full rationale

The derivation is self-contained and non-circular. The target E(fhat) never enters the construction of bU. The theoretical bound uses the worst-case envelope (18), proven pointwise via the reproducing property and Cauchy-Schwarz: |w_i| <= |ew_i| + (||fhat||_H + B) sqrt(k(x_i,x_i)), and the bias input b = (1/2)B sqrt(lambda), proven from ||(H-I)f*||_n^2 <= lambda B^2/4. The refit movement m(a) = ||H(xi o a)||_n is simulated from Gaussian multipliers, and q_{1-alpha}(a) is an order statistic of these independent draws, so it is an independently generated threshold rather than a quantity fitted to the prediction error. The domination chain rests on the external Anderson inequality and a Berry-Esseen comparison proved in the paper. The residual terms delta, delta_N, and L^{-1/2} in Theorem 4.1 are explicit slack, and the paper itself cautions that validity is conditional on delocalized noise and that the data-driven envelope (20) is only empirically valid; these are honesty limitations or correctness concerns, not circular reductions. No load-bearing self-citation appears: Wainwright (2025) is used as a baseline that the paper explicitly shows degenerates for kernel ridge regression, not as justification of the central claim.

Assumptions & free parameters 1 free parameters · 10 assumptions · 0 invented entities

The certified bound depends on standard probability inequalities and on domain assumptions stated in the paper: a known RKHS ball for f*, symmetric conditional noise, spectral decay/bounded leverage for the rate, and sub-Gaussian tails only for contraction speed. No new entities are introduced, and the only hand-set constant in the practical pipeline is the pilot penalty factor λ/20.

free parameters (1)

pilot penalty factor = λ/20 (one twentieth of the ridge penalty)
Used in the data-driven envelope (20) to form the bias vector b̂ from an undersmoothed pilot fit (Appendix B.2). It is chosen by hand; empirical coverage claims may depend on it, but no theoretical guarantee does.

assumptions (10)

standard math Anderson's inequality for centered Gaussian measures
Used in Lemma 3.4: enlarging coordinates of the multiplier noise scale can only raise upper quantiles of the movement norm.
standard math Berry–Esseen bounds for quadratic forms (fourth-moment theorem)
Used in Theorem 4.1 to control the gap between the Gaussian and Rademacher movement distributions.
standard math Dvoretzky–Kiefer–Wolfowitz inequality for order-statistic calibration
Used to convert the Monte Carlo order statistic into a quantile estimate with O(1/√L) error.
standard math Hanson–Wright inequality
Used in Theorem 4.8 to lower-bound ||Hw||²_n around its conditional mean.
domain assumption Assumption 2.1: f* in RKHS ball with known B and kernel bound κ²
Feeds the worst-case envelope and the bias input b=½B√λ; without it no certified bound.
domain assumption Assumption 2.2: noise signs iid Rademacher, independent of magnitudes
The sole distributional assumption; licenses the sign-perturbation comparison in Theorem 4.1.
domain assumption Assumption 4.4: eigenvalue decay µ_j ≍ j^{-2s}, λ ≍ n^{-2s/(2s+1)}, bounded leverage
Underpins the minimax rate and the boundedness of the movement's mean.
domain assumption Assumption 4.5: sub-Gaussian noise tails for rate (not for validity)
Controls how quickly the bound contracts; heavy-tailed noise keeps validity but loses the rate.
domain assumption Assumption 4.7: noise energy through the smoother is bounded below
Needed for the ratio bound Û/E(f̂)=O_P(1).
domain assumption Fixed design and penalty λ fixed before seeing y
The smoother H is a constant matrix; same-data tuning of λ is outside the guarantees (Remark 4.3).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Upper Confidence Bounds for the Prediction Error of Kernel Ridge Regression via Gaussian Refitting." pith.science (2026). https://pith.science/paper/QSINAHEL

@misc{pith2026260728846,
  author       = {Pith},
  title        = {Pith review of: Upper Confidence Bounds for the Prediction Error of Kernel Ridge Regression via Gaussian Refitting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QSINAHEL}},
  note         = {Machine review of arXiv:2607.28846}
}

abstract

Assessing a single model fit requires a computable upper confidence bound for the gap between the fit and the unknown truth, as mean estimates ignore realization variance. Standard cross-validation margins are bottlenecked at order $n^{-1/2}$ by noise fluctuations, even when the true error shrinks faster. While wild refitting cancels this noise level, existing Rademacher sign methods degenerate for kernel ridge regression and rely on unobservable quantities. We propose a Gaussian refit for kernel ridge regression. By Anderson's inequality, the fit movement is monotone in the noise sizes, yielding a computable tail bound. Assuming only symmetric noise, the bound requires no moment assumptions and is calibrated at any confidence level via order statistics. Theoretically, using a worst-case envelope, the bound contracts at the minimax rate $O_P(n^{-2s/(2s+1)})$, correctly matching the prediction error. Empirically, using a practical data-driven envelope, the bound maintains full coverage within twice the true $95\%$ error quantile. By contrast, cross-validation exceeds this quantile by factors up to $51$, and by hundreds under infinite-variance noise. The procedure extends empirically to nonlinear constrained estimators and real spatial data.

Figures

Figures reproduced from arXiv: 2607.28846 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 2 linked inside Pith

[1]

Abbasi-Yadkori, D

Y. Abbasi-Yadkori, D. P \'a l, and C. Szepesv \'a ri. Improved algorithms for linear stochastic bandits. In Adv. Neural Inf. Process. Syst., volume 24, 2011

2011
[2]

T. W. Anderson. The integral of a symmetric unimodal function over a symmetric convex set and some probability inequalities. Proc. Amer. Math. Soc., 6 0 (2): 0 170--176, 1955

1955
[3]

Austern and W

M. Austern and W. Zhou. Asymptotics of cross-validation. Ann. Inst. Henri Poincar \'e Probab. Stat. , 61 0 (4), 2025

2025
[4]

Y. Baraud. Confidence balls in G aussian regression. Ann. Statist., 32 0 (2): 0 528--551, 2004

2004
[5]

R. F. Barber, E. J. Cand \`e s, A. Ramdas, and R. J. Tibshirani. Predictive inference with the jackknife+. Ann. Statist., 49 0 (1): 0 486--507, 2021

2021
[6]

Bates, T

S. Bates, T. Hastie, and R. Tibshirani. Cross-validation: what does it estimate and how well does it do it? J. Amer. Statist. Assoc., 119 0 (546): 0 1434--1445, 2024

2024
[7]

Bayle, A

P. Bayle, A. Bayle, L. Janson, and L. Mackey. Cross-validation confidence intervals for test error. In Adv. Neural Inf. Process. Syst., volume 33, 2020

2020
[8]

P. C. Bellec and C.-H. Zhang. Second-order S tein: SURE for SURE and other applications in high-dimensional inference. Ann. Statist., 49 0 (4): 0 1864--1903, 2021

1903

Show all 45 references

[9]

Bengio and Y

Y. Bengio and Y. Grandvalet. No unbiased estimator of the variance of K -fold cross-validation. J. Mach. Learn. Res., 5: 0 1089--1105, 2004

2004
[10]

Beran and L

R. Beran and L. D \"u mbgen. Modulation of estimators and confidence sets. Ann. Statist., 26 0 (5): 0 1826--1856, 1998

1998
[11]

T. T. Cai and M. G. Low. Adaptive confidence balls. Ann. Statist., 34 0 (1): 0 202--228, 2006

2006
[12]

Caponnetto and E

A. Caponnetto and E. De Vito. Optimal rates for the regularized least-squares algorithm. Found. Comput. Math., 7 0 (3): 0 331--368, 2007

2007
[13]

S. R. Chowdhury and A. Gopalan. On kernelized multi-armed bandits. In Proc. 34th Int. Conf. Machine Learning, volume 70 of Proceedings of Machine Learning Research, pages 844--853, 2017

2017
[14]

B. C. Cs \'a ji and K. B. Kis. Distribution-free uncertainty quantification for kernel methods by gradient perturbations. Machine Learning, 108: 0 1677--1699, 2019

2019
[15]

B. C. Cs \'a ji, M. C. Campi, and E. Weyer. Sign-perturbed sums: a new system identification approach for constructing exact non-asymptotic confidence regions in linear regression models. IEEE Trans. Signal Process., 63 0 (1): 0 169--181, 2015

2015
[16]

Davidson and E

R. Davidson and E. Flachaire. The wild bootstrap, tamed at last. Journal of Econometrics, 146 0 (1): 0 162--169, 2008

2008
[17]

D \"o bler and G

C. D \"o bler and G. Peccati. The fourth moment theorem on the P oisson space. Ann. Probab., 45 0 (3): 0 1804--1849, 2017

2017
[18]

Fiedler, C

C. Fiedler, C. W. Scherer, and S. Trimpe. Practical and rigorous uncertainty bounds for G aussian process regression. In Proc. AAAI Conf. Artificial Intelligence, volume 35, pages 7439--7447, 2021

2021
[19]

S. Geisser. The predictive sample reuse method with applications. J. Amer. Statist. Assoc., 70 0 (350): 0 320--328, 1975

1975
[20]

Hu and D

H. Hu and D. Simchi-Levi. Perturbing the derivative: Wild refitting for model-free evaluation of machine learning models under B regman losses. arXiv preprint arXiv:2509.02476, 2025 a

2025
[21]

Hu and D

H. Hu and D. Simchi-Levi. Perturbing the derivative: Doubly wild refitting for model-free evaluation of opaque machine learning predictors. arXiv preprint arXiv:2511.18789, 2025 b

2025
[22]

J. P. Imhof. Computing the distribution of quadratic forms in normal variables. Biometrika, 48 0 (3/4): 0 419--426, 1961

1961
[23]

Juditsky and S

A. Juditsky and S. Lambert-Lacroix. Nonparametric confidence set estimation. Math. Methods Statist., 12 0 (4): 0 410--428, 2003

2003
[24]

A. Lahr, J. K \"o hler, A. Scampicchio, and M. N. Zeilinger. Optimal kernel regression bounds under energy-bounded noise. arXiv:2505.22235, 2025

2025
[25]

Laurent and P

B. Laurent and P. Massart. Adaptive estimation of a quadratic functional by model selection. Ann. Statist., 28 0 (5): 0 1302--1338, 2000

2000
[26]

J. Lei. Cross-validation with confidence. J. Amer. Statist. Assoc., 115 0 (532): 0 1978--1997, 2020

1978
[27]

J. Lei, M. G'Sell, A. Rinaldo, R. J. Tibshirani, and L. Wasserman. Distribution-free predictive inference for regression. J. Amer. Statist. Assoc., 113 0 (523): 0 1094--1111, 2018

2018
[28]

K.-C. Li. Honest confidence regions for nonparametric regression. Ann. Statist., 17 0 (3): 0 1001--1008, 1989

1989
[29]

J. Lin, A. Rudi, L. Rosasco, and V. Cevher. Optimal rates for spectral algorithms with least-squares regression over H ilbert spaces. Appl. Comput. Harmon. Anal., 48 0 (3): 0 868--890, 2020

2020
[30]

R. Y. Liu. Bootstrap procedures under some non-i.i.d. models. Ann. Statist., 16 0 (4): 0 1696--1708, 1988

1988
[31]

E. T. Maddalena, P. Scharnhorst, and C. N. Jones. Deterministic error bounds for kernel-based learning techniques under bounded noise. Automatica, 134: 0 109896, 2021

2021
[32]

E. Mammen. When Does Bootstrap Work? Asymptotic Results and Simulations, volume 77 of Lecture Notes in Statistics. Springer-Verlag, New York, 1992

1992
[33]

E. Mammen. Bootstrap and wild bootstrap for high dimensional linear models. Ann. Statist., 21 0 (1): 0 255--285, 1993

1993
[34]

Nourdin, G

I. Nourdin, G. Peccati, and G. Reinert. Invariance principles for homogeneous sums: universality of G aussian W iener chaos. Ann. Probab., 38 0 (5): 0 1947--1985, 2010

1947
[35]

Robins and A

J. Robins and A. van der Vaart. Adaptive nonparametric confidence sets. Ann. Statist., 34 0 (1): 0 229--253, 2006

2006
[36]

Rudelson and R

M. Rudelson and R. Vershynin. Hanson-- W right inequality and sub- G aussian concentration. Electron. Commun. Probab., 18: 0 1--9, 2013

2013
[37]

Singh and S

R. Singh and S. Vijaykumar. Kernel ridge regression inference. arXiv:2302.06578, 2023

2023 arXiv
[38]

C. M. Stein. Estimation of the mean of a multivariate normal distribution. Ann. Statist., 9 0 (6): 0 1135--1151, 1981

1981
[39]

C. J. Stone. Optimal global rates of convergence for nonparametric regression. Ann. Statist., 10 0 (4): 0 1040--1053, 1982

1982
[40]

M. Stone. Cross-validatory choice and assessment of statistical predictions. J. R. Statist. Soc. B, 36 0 (2): 0 111--147, 1974

1974
[41]

V. Vovk, A. Gammerman, and G. Shafer. Algorithmic Learning in a Random World. Springer, 2005

2005
[42]

S. Wager. Cross-validation, risk estimation, and model selection: comment on a paper by R osset and T ibshirani. J. Amer. Statist. Assoc., 115 0 (529): 0 157--160, 2020

2020
[43]

M. J. Wainwright. Wild refitting for black box prediction. arXiv preprint arXiv:2506.21460, 2025

2025 arXiv
[44]

C. F. J. Wu. Jackknife, bootstrap and other resampling methods in regression analysis. Ann. Statist., 14 0 (4): 0 1261--1295, 1986

1986
[45]

Yang and A

Y. Yang and A. Barron. Information-theoretic determination of minimax rates of convergence. Ann. Statist., 27 0 (5): 0 1564--1599, 1999

1999

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.

[1] [1]

Abbasi-Yadkori, D

Y. Abbasi-Yadkori, D. P \'a l, and C. Szepesv \'a ri. Improved algorithms for linear stochastic bandits. In Adv. Neural Inf. Process. Syst., volume 24, 2011

2011

[2] [2]

T. W. Anderson. The integral of a symmetric unimodal function over a symmetric convex set and some probability inequalities. Proc. Amer. Math. Soc., 6 0 (2): 0 170--176, 1955

1955

[3] [3]

Austern and W

M. Austern and W. Zhou. Asymptotics of cross-validation. Ann. Inst. Henri Poincar \'e Probab. Stat. , 61 0 (4), 2025

2025

[4] [4]

Y. Baraud. Confidence balls in G aussian regression. Ann. Statist., 32 0 (2): 0 528--551, 2004

2004

[5] [5]

R. F. Barber, E. J. Cand \`e s, A. Ramdas, and R. J. Tibshirani. Predictive inference with the jackknife+. Ann. Statist., 49 0 (1): 0 486--507, 2021

2021

[6] [6]

Bates, T

S. Bates, T. Hastie, and R. Tibshirani. Cross-validation: what does it estimate and how well does it do it? J. Amer. Statist. Assoc., 119 0 (546): 0 1434--1445, 2024

2024

[7] [7]

Bayle, A

P. Bayle, A. Bayle, L. Janson, and L. Mackey. Cross-validation confidence intervals for test error. In Adv. Neural Inf. Process. Syst., volume 33, 2020

2020

[8] [8]

P. C. Bellec and C.-H. Zhang. Second-order S tein: SURE for SURE and other applications in high-dimensional inference. Ann. Statist., 49 0 (4): 0 1864--1903, 2021

1903