pith. sign in

arxiv: 2606.09049 · v2 · pith:52H27TOAnew · submitted 2026-06-08 · 📊 stat.ME · cs.LG· math.ST· stat.ML· stat.TH

Data augmented bootstrap: Unifying confidence interval construction by approximate invariance

Kevin Han Huang This is my paper

Pith reviewed 2026-06-27 15:55 UTC · model grok-4.3

classification 📊 stat.ME cs.LGmath.STstat.MLstat.TH
keywords data augmented bootstrapapproximate invarianceconfidence intervalsconformal predictionbootstrap methodsdata augmentationKolmogorov distance
0
0 comments X

The pith

Data augmented bootstrap constructs confidence intervals from approximately invariant data transformations without requiring exact group symmetries.

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

The paper proposes a framework that builds confidence intervals by applying transformations to data that are only approximately invariant. It shows this recovers both exact methods such as conformal prediction and the classical bootstrap as special cases. Coverage guarantees are derived that become exact in finite samples when invariance is strong and become asymptotic when invariance is weaker. The approach measures invariance via Kolmogorov distance and reduces it to mean and variance matching for statistics with Gaussian universality, which lets data-augmentation steps from machine learning be folded into bootstrap and conformal procedures.

Core claim

The data augmented bootstrap framework constructs confidence intervals from approximately invariant transformations of the data. It recovers conformal prediction, wild bootstrap for MMD U-statistics, SymmPI, and the classical bootstrap as special cases. Theoretical coverage results hold without assuming a group structure and interpolate between finite-sample exactness and asymptotic validity according to the strength of the invariance, with the strength measured in Kolmogorov distance and reducing to conditional mean and variance matching under Gaussian universality.

What carries the argument

Data augmented bootstrap framework that generates confidence intervals from data transformations whose approximate invariance is quantified by Kolmogorov distance.

If this is right

  • DAB recovers methods that rely on exact group symmetries as special cases when invariance is perfect.
  • Coverage guarantees interpolate between finite-sample and asymptotic regimes as the measured invariance strength varies.
  • Data augmentation steps can be inserted into bootstrap, wild bootstrap, and conformal prediction while retaining the interpolated coverage results.
  • The same framework applies to simulated, image, language, and scientific datasets without needing a group structure on the transformations.

Where Pith is reading between the lines

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

  • The interpolation result may allow practitioners to trade off computational cost of stronger augmentations against the tightness of the resulting intervals.
  • Because no group structure is required, the approach could extend to transformations that are only locally invariant, such as small rotations or crops in image data.
  • The reduction to mean-variance matching under Gaussian universality suggests that DAB could be combined with existing asymptotic expansions for other statistics.

Load-bearing premise

The strength of the approximate invariance can be quantified accurately enough by Kolmogorov distance or by mean-variance matching to control the coverage error.

What would settle it

A simulation or real-data experiment in which the Kolmogorov distance between the original and transformed distributions is small yet the resulting intervals fail to achieve the nominal coverage rate, or the distance is large yet coverage still holds at the nominal rate.

Figures

Figures reproduced from arXiv: 2606.09049 by Kevin Han Huang.

Figure 1
Figure 1. Figure 1: Illustration of our theoretical results. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Gaussian, bootstrap and DAB CIs for averages of 2d Gaussians over 500 random trials. In (i), [PITH_FULL_IMAGE:figures/full_fig_p019_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: CIs for 3-electron configuration samples from a FermiNet wavefunction trained for the Lithium atom, under varying n (number of Markov chains) and B. Left. Visualisation of rotational symmetry through the marginal density for 1-electron in the 2d plane. Right. Empirical coverage and length of the CIs for the x-component of the electron dipole moment of the samples. The error bars are over 500 random trials,… view at source ↗
Figure 4
Figure 4. Figure 4: CIs for MMD statistics with RBF kernel on 2d Rademachers over 500 random trials. [PITH_FULL_IMAGE:figures/full_fig_p021_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: CIs for MMD statistics with RBF kernel for testing noiseless MNIST images against those with [PITH_FULL_IMAGE:figures/full_fig_p021_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: A noisy MNIST image with varying σ in (i) – (iv) and with different augmentations in (v) – (vii). 0.8 0.9 1.0 50 100 150 200 250 2.5 5.0 7.5 10.0 Split-CP CP + Ortho CP + Permute (i) Prediction interval for the outcome of a 2d linear model based on linear regression on Gamma vectors n coverage CI length 0.85 0.90 0.95 1.00 10 15 20 25 30 35 40 45 50 30 60 90 120150 180 210 240 270 300 330 360 390 420 450 4… view at source ↗
Figure 7
Figure 7. Figure 7: DAB-variants of conformal prediction over [PITH_FULL_IMAGE:figures/full_fig_p022_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Illustration of the interpolation paths between the two random vectors [PITH_FULL_IMAGE:figures/full_fig_p049_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Simulations with empirical averages of 2d Gaussians over 500 random trials, with varying [PITH_FULL_IMAGE:figures/full_fig_p067_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Simulations with empirical averages of 2d Rademachers over 500 random trials, with varying [PITH_FULL_IMAGE:figures/full_fig_p068_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Simulations with empirical averages of 2d centred Gammas over 500 random trials, with varying [PITH_FULL_IMAGE:figures/full_fig_p068_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Quality of the CIs for the x-component of the electron dipole moment of a FermiNet wave￾function trained for the Lithium atom, over 500 random trials. (i) Empirical v.s. target coverage, where the dashed line y = x indicates the desired coverage level. (ii) CI length v.s. target coverage. (iii) Empirical distribution of CI lengths over the random trials. This follows the same setup as [PITH_FULL_IMAGE:fi… view at source ↗
Figure 13
Figure 13. Figure 13: MMD statistics with RBF kernel on 2d Gaussians over 500 random trials. [PITH_FULL_IMAGE:figures/full_fig_p069_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: MMD statistics with RBF kernel on 2d centred Gammas over 500 random trials. [PITH_FULL_IMAGE:figures/full_fig_p070_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Empirical coverage and CI length of DAB for MMD statistics with RBF kernel under the null [PITH_FULL_IMAGE:figures/full_fig_p070_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Randomly sampled CIFAR-10 images of odd-numbered classes and of even-numbered classes. [PITH_FULL_IMAGE:figures/full_fig_p071_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: A randomly sampled CIFAR-10 image with the augmentations considered in our experiments. [PITH_FULL_IMAGE:figures/full_fig_p071_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Empirical coverage and CI length of DAB for MMD statistics with RBF kernel under the null [PITH_FULL_IMAGE:figures/full_fig_p071_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Empirical power and CI length of DAB for MMD statistics with RBF kernel for testing CIFAR [PITH_FULL_IMAGE:figures/full_fig_p072_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Empirical coverage and CI length of DAB for MMD statistics with RBF kernel under the null [PITH_FULL_IMAGE:figures/full_fig_p072_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Empirical power and CI length of DAB for MMD statistics with RBF kernel for testing back [PITH_FULL_IMAGE:figures/full_fig_p073_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Empirical coverage and CI length of DAB for MMD statistics with RBF kernel under the null of [PITH_FULL_IMAGE:figures/full_fig_p073_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Empirical power and CI length of DAB for MMD statistics with RBF kernel for testing back [PITH_FULL_IMAGE:figures/full_fig_p073_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: DAB-variants of conformal prediction for predicting the outcome of a 2d linear model with [PITH_FULL_IMAGE:figures/full_fig_p074_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Same setup as Figure [PITH_FULL_IMAGE:figures/full_fig_p074_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: DAB-variants of conformal prediction for predicting the outcome of a 2d linear model with [PITH_FULL_IMAGE:figures/full_fig_p075_26.png] view at source ↗
Figure 27
Figure 27. Figure 27: Visualisation of selected molecules from the QM-symm database [ [PITH_FULL_IMAGE:figures/full_fig_p076_27.png] view at source ↗
Figure 28
Figure 28. Figure 28: DAB-variants of conformal prediction for predicting the energy of a randomly chosen C4h [PITH_FULL_IMAGE:figures/full_fig_p077_28.png] view at source ↗
Figure 29
Figure 29. Figure 29: DAB-variants of conformal prediction for predicting the energy of a randomly chosen C4h [PITH_FULL_IMAGE:figures/full_fig_p078_29.png] view at source ↗
Figure 30
Figure 30. Figure 30: Example of a multiple-choice record from the MMLU dataset [ [PITH_FULL_IMAGE:figures/full_fig_p079_30.png] view at source ↗
Figure 31
Figure 31. Figure 31: DAB-variants of conformal prediction for predicting the answer of a multiple-choice question [PITH_FULL_IMAGE:figures/full_fig_p080_31.png] view at source ↗
Figure 32
Figure 32. Figure 32: Same setup as Figure [PITH_FULL_IMAGE:figures/full_fig_p080_32.png] view at source ↗
read the original abstract

We propose the data augmented bootstrap (DAB), a framework for constructing confidence intervals from approximately invariant transformations of the data. As special cases, DAB recovers popular methods that rely on exact group symmetries, such as conformal prediction, wild bootstrap for Maximum Mean Discrepancy U-statistics and the recently proposed SymmPI. Meanwhile, DAB also recovers the classical bootstrap method, which exploits the dataset's approximate invariance under uniform sampling of data indices as the dataset size grows. For all DAB methods, we establish theoretical coverage results that interpolate between finite-sample and asymptotic guarantees according to the strength of the invariance, and without assuming a group structure. The approximate invariance is measured in the Kolmogorov distance and, for statistics that satisfy Gaussian universality, reduces to conditional mean and variance matching. This allows us to incorporate data augmentation (DA), a widely used machine learning heuristic based on approximate invariances, into known statistical methods. We empirically test the performance of incorporating DA into bootstrap, wild bootstrap and conformal prediction for simulated settings as well as for image, language and scientific data.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

0 major / 3 minor

Summary. The paper proposes the data augmented bootstrap (DAB) as a framework for confidence interval construction based on approximately invariant data transformations, with invariance quantified via Kolmogorov distance. It recovers conformal prediction, wild bootstrap for MMD U-statistics, SymmPI, and the classical bootstrap as special cases. Theoretical coverage guarantees are claimed to interpolate between finite-sample exactness and asymptotic validity according to the degree of invariance, without requiring a group structure. For statistics obeying Gaussian universality the invariance condition reduces to conditional mean and variance matching, permitting data-augmentation heuristics to be folded into bootstrap and conformal procedures. The claims are supported by empirical experiments on simulated data as well as image, language, and scientific datasets.

Significance. If the interpolation result holds, DAB supplies a single, group-structure-free lens that unifies exact finite-sample methods with asymptotic bootstrap procedures and legitimizes the use of machine-learning-style data augmentation inside classical inferential pipelines. The explicit reduction to mean/variance matching under Gaussian universality and the recovery of several well-known procedures as boundary cases are concrete strengths. The empirical scope across modalities is also useful for assessing practical performance.

minor comments (3)
  1. The abstract states that coverage results 'interpolate between finite-sample and asymptotic guarantees according to the strength of the invariance,' yet provides no explicit statement of the functional dependence of the coverage error on the Kolmogorov distance; a short clarifying sentence or reference to the relevant theorem would help readers locate the precise interpolation statement.
  2. The reduction of Kolmogorov-distance invariance to conditional mean and variance matching is scoped to 'statistics that satisfy Gaussian universality.' The manuscript should state the precise definition or reference used for Gaussian universality and indicate whether this assumption is verified or merely invoked in the empirical sections.
  3. The empirical section reports performance on image, language, and scientific data, but the abstract does not specify the sample sizes, number of Monte Carlo replications, or the exact DAB variants tested; adding these details would improve reproducibility.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of the manuscript, the clear summary of its contributions, and the recommendation for minor revision. No specific major comments were raised.

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper defines the DAB framework directly from an externally measured approximate invariance (Kolmogorov distance on transformations of the data), without any equation or definition that reduces the coverage result to a fitted quantity defined from the same data. Special cases (conformal prediction, wild bootstrap, classical bootstrap) are recovered by specializing the invariance strength, which is an input rather than an output of the procedure. The interpolation between finite-sample and asymptotic guarantees is stated to follow from the strength of this external invariance measure, and the reduction to conditional mean/variance matching is explicitly scoped to statistics obeying Gaussian universality. No self-citation is load-bearing for the central claim, and no ansatz or uniqueness theorem is smuggled in. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the existence of a Kolmogorov-distance measure of approximate invariance that can be computed or bounded for data-augmentation transformations, and on the domain assumption that relevant statistics satisfy Gaussian universality so that the distance reduces to mean-variance matching. No free parameters or invented entities are stated in the abstract.

axioms (2)
  • domain assumption Relevant statistics satisfy Gaussian universality, allowing approximate invariance measured by Kolmogorov distance to reduce to conditional mean and variance matching.
    Invoked in the abstract to justify incorporating data augmentation into bootstrap and conformal methods.
  • domain assumption Approximate invariance can be quantified without requiring an underlying group structure.
    Stated explicitly as part of the theoretical coverage results that do not assume a group structure.

pith-pipeline@v0.9.1-grok · 5713 in / 1522 out tokens · 30961 ms · 2026-06-27T15:55:08.005026+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

98 extracted references · 5 canonical work pages · 3 internal anchors

  1. [1]

    Advances in Neural Information Processing Systems , volume=

    A wild bootstrap for degenerate kernel tests , author=. Advances in Neural Information Processing Systems , volume=

  2. [2]

    Advances in Neural Information Processing Systems , pages =

    Asymptotics of the Bootstrap via Stability with Applications to Inference with Model Selection , author=. Advances in Neural Information Processing Systems , pages =

  3. [3]

    Sankhya A , pages=

    Permutation tests using arbitrary permutation distributions , author=. Sankhya A , pages=. 2023 , publisher=

    2023

  4. [4]

    NeurIPS 2023 AI for Science Workshop , year=

    Hypothesis tests for distributional group symmetry with applications to particle physics , author=. NeurIPS 2023 AI for Science Workshop , year=

    2023

  5. [5]

    A survey on image data augmentation for deep learning , author=. J. Big Data , volume=. 2019 , publisher=

    2019

  6. [6]

    Journal of big Data , volume=

    Text data augmentation for deep learning , author=. Journal of big Data , volume=. 2021 , publisher=

    2021

  7. [7]

    Gaussian and Non-Gaussian Universality of Data Augmentation , author=. Ann. Statist. (forthcoming) , year=

  8. [8]

    Dependent wild bootstrap for degenerate U-and V-statistics , author=. J. Multivariate Anal. , volume=. 2013 , publisher=

    2013

  9. [9]

    Asian conference on machine learning , pages=

    Conditional validity of inductive conformal predictors , author=. Asian conference on machine learning , pages=. 2012 , organization=

    2012

  10. [10]

    Distribution-free prediction bands for non-parametric regression , author=. J. Roy. Statist. Soc. Ser. B , volume=. 2014 , publisher=

    2014

  11. [11]

    The limits of distribution-free conditional predictive inference , author=. Inf. Inference , volume=. 2021 , publisher=

    2021

  12. [12]

    2005 , publisher=

    Algorithmic learning in a random world , author=. 2005 , publisher=

    2005

  13. [13]

    A high-dimensional convergence theorem for

    Huang, Kevin H and Liu, Xing and Duncan, Andrew and Gandy, Axel , booktitle=. A high-dimensional convergence theorem for. 2023 , organization=

    2023

  14. [14]

    Proceedings of Thirty Eighth Conference on Learning Theory , pages=

    Universality of High-Dimensional Logistic Regression and a Novel CGMT under Dependence with Applications to Data Augmentation , author=. Proceedings of Thirty Eighth Conference on Learning Theory , pages=. 2025 , organization=

    2025

  15. [15]

    Gaussian universality for approximately polynomial functions of high-dimensional data , author=

  16. [16]

    Dobriban, Edgar and Yu, Mengxin , title =. J. Roy. Statist. Soc. Ser. B , volume =. 2025 , month =

    2025

  17. [17]

    2011 , publisher=

    Chen, Louis HY and Goldstein, Larry and Shao, Qi-Man , volume=. 2011 , publisher=

    2011

  18. [18]

    IEEE Trans

    Applications of the Lindeberg principle in communications and statistical learning , author=. IEEE Trans. Inf. Theory , volume=. 2011 , publisher=

    2011

  19. [19]

    Conference on Learning Theory , pages=

    Universality of empirical risk minimization , author=. Conference on Learning Theory , pages=. 2022 , organization=

    2022

  20. [20]

    Predictive inference with the jackknife+ , author=. Ann. Statist. , volume=. 2021 , publisher=

    2021

  21. [21]

    A group-theoretic framework for data augmentation , author=. J. Mach. Learn. Res , volume=

  22. [22]

    arXiv preprint arXiv:2502.20579 , year=

    Characterizing the Training-Conditional Coverage of Full Conformal Inference in High Dimensions , author=. arXiv preprint arXiv:2502.20579 , year=

    work page arXiv

  23. [23]

    Conformal prediction beyond exchangeability , author=. Ann. Statist. , volume=. 2023 , publisher=

    2023

  24. [24]

    Breakthroughs in statistics: Methodology and distribution , pages=

    Bootstrap methods: another look at the jackknife , author=. Breakthroughs in statistics: Methodology and distribution , pages=. 1992 , publisher=

    1992

  25. [25]

    1992 , publisher=

    The bootstrap and Edgeworth expansion , author=. 1992 , publisher=

    1992

  26. [26]

    Random quadratic forms and the bootstrap for U-statistics , author=. J. Multivariate Anal. , volume=. 1994 , publisher=

    1994

  27. [27]

    Conditional predictive inference for stable algorithms , author=. Ann. Statist. , volume=. 2023 , publisher=

    2023

  28. [28]

    Advances in Neural Information Processing Systems , volume=

    Universality laws for gaussian mixtures in generalized linear models , author=. Advances in Neural Information Processing Systems , volume=

  29. [29]

    The generalization error of max-margin linear classifiers: Benign overfitting and high dimensional asymptotics in the overparametrized regime , author=. Ann. Statist. , volume=. 2025 , publisher=

    2025

  30. [30]

    IEEE Trans

    Universality laws for high-dimensional learning with random features , author=. IEEE Trans. Inf. Theory , volume=. 2022 , publisher=

    2022

  31. [31]

    , author=

    A tutorial on conformal prediction. , author=. J. Mach. Learn. Res , volume=

  32. [32]

    Conformal prediction: A gentle introduction , author=. Found. Trends Mach. Learn. , volume=. 2023 , publisher=

    2023

  33. [33]

    Bernoulli , volume=

    Conformal prediction: a unified review of theory and new challenges , author=. Bernoulli , volume=. 2023 , publisher=

    2023

  34. [34]

    2005 , publisher=

    Permutation, parametric and bootstrap tests of hypotheses , author=. 2005 , publisher=

    2005

  35. [35]

    MMD aggregated two-sample test , author=. J. Mach. Learn. Res , volume=

  36. [36]

    46th Annual IEEE Symposium on Foundations of Computer Science (FOCS'05) , pages=

    Noise stability of functions with low influences: invariance and optimality , author=. 46th Annual IEEE Symposium on Foundations of Computer Science (FOCS'05) , pages=. 2005 , organization=

    2005

  37. [37]

    A generalization of the

    Chatterjee, Sourav , journal=. A generalization of the. 2006 , publisher=

    2006

  38. [38]

    Gaussian universality of perceptrons with random labels , author=. Phys. Rev. E , volume=. 2024 , publisher=

    2024

  39. [39]

    Universality of regularized regression estimators in high dimensions , author=. Ann. Statist. , volume=. 2023 , publisher=

    2023

  40. [40]

    Advances in Neural Information Processing Systems , volume=

    Conformalized quantile regression , author=. Advances in Neural Information Processing Systems , volume=

  41. [41]

    Distribution-free predictive inference for regression , author=. J. Amer. Statist. Assoc. , volume=. 2018 , publisher=

    2018

  42. [42]

    Least ambiguous set-valued classifiers with bounded error levels , author=. J. Amer. Statist. Assoc. , volume=. 2019 , publisher=

    2019

  43. [43]

    The journal of machine learning research , volume=

    A kernel two-sample test , author=. The journal of machine learning research , volume=. 2012 , publisher=

    2012

  44. [44]

    The two-sample problem for poisson processes: Adaptive tests with a nonasymptotic wild bootstrap approach , author=

  45. [45]

    Weighted bootstrapping of U-statistics , author=. J. Statist. Plann. Inference , volume=. 1994 , publisher=

    1994

  46. [46]

    Distributional and L^q norm inequalities for polynomials over convex bodies in

    Carbery, Anthony and Wright, James , journal=. Distributional and L^q norm inequalities for polynomials over convex bodies in. 2001 , publisher=

    2001

  47. [47]

    The existence of regular conditional probabilities: necessary and sufficient conditions , author=. Ann. Probab. , pages=. 1985 , publisher=

    1985

  48. [48]

    Theory Probab

    An estimate for concentration functions , author=. Theory Probab. Appl. , volume=. 1961 , publisher=

    1961

  49. [49]

    Comparison and anti-concentration bounds for maxima of Gaussian random vectors , author=. Probab. Theory Related Fields , volume=. 2015 , publisher=

    2015

  50. [50]

    Bernoulli (forthcoming) , year=

    Sharp Anti-Concentration Inequalities for Extremum Statistics via Copulas , author=. Bernoulli (forthcoming) , year=

  51. [51]

    Matrix concentration inequalities via the method of exchangeable pairs , author=

  52. [52]

    2018 , publisher=

    High-dimensional probability: An introduction with applications in data science , author=. 2018 , publisher=

    2018

  53. [53]

    Stochastic Process

    Moment bounds for dependent sequences in smooth Banach spaces , author=. Stochastic Process. Appl. , volume=. 2015 , publisher=

    2015

  54. [54]

    Probability and moment inequalities for additive functionals of geometrically ergodic Markov chains , author=. J. Theoret. Probab. , volume=. 2024 , publisher=

    2024

  55. [55]

    arXiv preprint arXiv:2409.05202 (2024)

    A survey on mixup augmentations and beyond , author=. arXiv preprint arXiv:2409.05202 , year=

    work page arXiv

  56. [56]

    Deep learning for population size history inference: Design, comparison and combination with approximate Bayesian computation , author=. Mol. Ecol. Resour. , volume=. 2021 , publisher=

    2021

  57. [57]

    Deep learning suggests that gene expression is encoded in all parts of a co-evolving interacting gene regulatory structure , author=. Nat. Commun. , volume=. 2020 , publisher=

    2020

  58. [58]

    International Conference on Learning Representations , year=

    mixup: Beyond Empirical Risk Minimization , author=. International Conference on Learning Representations , year=

  59. [59]

    Proceedings of the 58th annual meeting of the association for computational linguistics , pages=

    Mixtext: Linguistically-informed interpolation of hidden space for semi-supervised text classification , author=. Proceedings of the 58th annual meeting of the association for computational linguistics , pages=

  60. [60]

    ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Mixspeech: Data augmentation for low-resource automatic speech recognition , author=. ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2021 , organization=

    2021

  61. [61]

    Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021 , pages=

    Ssmix: Saliency-based span mixup for text classification , author=. Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021 , pages=

    2021

  62. [62]

    Advances in Neural Information Processing Systems , volume=

    Contrast and mix: Temporal contrastive video domain adaptation with background mixing , author=. Advances in Neural Information Processing Systems , volume=

  63. [63]

    Advances in Neural Information Processing Systems , volume=

    Conformal prediction under covariate shift , author=. Advances in Neural Information Processing Systems , volume=

  64. [64]

    On the bootstrap of U and V statistics , author=. Ann. Statist. , pages=. 1992 , publisher=

    1992

  65. [65]

    Mercer’s theorem on general domains: On the interaction between measures, kernels, and RKHSs , author=. Constr. Approx. , volume=. 2012 , publisher=

    2012

  66. [66]

    Advances in Neural Information Processing Systems , volume=

    Debiased machine learning without sample-splitting for stable estimators , author=. Advances in Neural Information Processing Systems , volume=

  67. [67]

    International Conference on Artificial Intelligence and Statistics , pages=

    Posterior uncertainty quantification in neural networks using data augmentation , author=. International Conference on Artificial Intelligence and Statistics , pages=. 2024 , organization=

    2024

  68. [68]

    Ab initio calculation of real solids via neural network ansatz , author=. Nat. Commun. , volume=. 2022 , publisher=

    2022

  69. [69]

    Proceedings of the 42nd International Conference on Machine Learning , pages =

    Diagonal Symmetrization of Neural Network Solvers for the Many-Electron Schrödinger Equation , author=. Proceedings of the 42nd International Conference on Machine Learning , pages =. 2025 , volume =

    2025

  70. [70]

    Ab initio solution of the many-electron Schr

    Pfau, David and Spencer, James S and Matthews, Alexander GDG and Foulkes, W Matthew C , journal=. Ab initio solution of the many-electron Schr. 2020 , publisher=

    2020

  71. [71]

    Science , volume=

    Solving the quantum many-body problem with artificial neural networks , author=. Science , volume=. 2017 , publisher=

    2017

  72. [72]

    Deep-neural-network solution of the electronic Schr

    Hermann, Jan and Sch. Deep-neural-network solution of the electronic Schr. Nat. Chem. , volume=. 2020 , publisher=

    2020

  73. [73]

    QM-sym, a symmetrized quantum chemistry database of 135 kilo molecules , author=. Sci. Data , volume=. 2019 , publisher=

    2019

  74. [74]

    Robust and scalable uncertainty estimation with conformal prediction for machine-learned interatomic potentials , author=. Mach. Learn.: Sci. Technol. , volume=. 2022 , publisher=

    2022

  75. [75]

    ICML 2023 Neural Conversational AI Workshop , year =

    Conformal prediction with large language models for multi-choice question answering , author=. ICML 2023 Neural Conversational AI Workshop , year =

    2023

  76. [76]

    International Conference on Learning Representations , year=

    Measuring Massive Multitask Language Understanding , author=. International Conference on Learning Representations , year=

  77. [77]

    Proceedings of the Twelfth Symposium on Conformal and Probabilistic Prediction with Applications , pages =

    A Review of Nonconformity Measures for Conformal Prediction in Regression , author =. Proceedings of the Twelfth Symposium on Conformal and Probabilistic Prediction with Applications , pages =. 2023 , editor =

    2023

  78. [78]

    Advances in Neural Information Processing Systems , volume=

    Classification with valid and adaptive coverage , author=. Advances in Neural Information Processing Systems , volume=

  79. [79]

    Proceedings of the 41st International Conference on Machine Learning , pages =

    Conformal Validity Guarantees Exist for Any Data Distribution (and How to Find Them) , author =. Proceedings of the 41st International Conference on Machine Learning , pages =. 2024 , volume =

    2024

  80. [80]

    The Thirteenth International Conference on Learning Representations , year=

    Wasserstein-Regularized Conformal Prediction under General Distribution Shift , author=. The Thirteenth International Conference on Learning Representations , year=

Showing first 80 references.