Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Epistemic Uncertainty in Conformal Scores: A Unified Approach

T0 review · 3 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

Pith's one-line read EPICSCORE wraps any conformal score in a Bayesian predictive CDF, keeping finite-sample marginal coverage and adding asymptotic conditional coverage by widening intervals where data are sparse.

desk verdict A useful Bayesian wrapper for conformal scores with honest marginal coverage, but the asymptotic conditional coverage claim leans on a strong assumption the experiments never verify. read the letter →

arxiv 2502.06995 v2 pith:ZL2RFNN6 submitted 2025-02-10 stat.ML cs.LG

classification stat.MLcs.LG MSC 62G15
keywords conformalpredictionepistemicuncertaintyBayesianpredictivedistributionconditionalcoveragedistribution-freeguaranteesGaussianprocessesMonteCarlodropoutscores
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 proposes EPICSCORE, a wrapper that takes any conformal score $s(x,y)$ — a number saying how unusual a candidate label $y$ is for features $x$ — and replaces it with $s'(x,y) = F(s(x,y)\mid x,\mathcal{D})$, the cumulative probability of that score under a Bayesian model fitted to calibration data. The claim is that this one transformation injects epistemic uncertainty into any conformal method: prediction intervals widen in data-sparse regions, where the Bayesian posterior is spread out, and stay narrow where data are abundant, while the finite-sample marginal coverage guarantee of conformal prediction survives untouched for every Bayesian model. Under a uniform-convergence assumption on the Bayesian model, the paper further proves asymptotic conditional coverage, the strongest achievable form of per-feature validity. If true, this means practitioners can attach epistemic uncertainty to any existing conformal pipeline using Gaussian processes, BART, or Monte Carlo dropout, and still keep a distribution-free guarantee.

What carries the argument

The load-bearing object is the Bayesian posterior predictive CDF, $F(s\mid x,\mathcal{D}) = \int F(s\mid x,\theta)\,f(\theta\mid\mathcal{D})\,d\theta$, computed from a subset of the calibration data and used to define $s'(x,y) = F(s(x,y)\mid x,\mathcal{D})$. It performs a distributional transform: it ranks the original conformal score against the posterior-averaged distribution of scores at $x$, so the conformal threshold $t_{1-\alpha}$ in the transformed space corresponds to the per-input quantile $F^{-1}(t_{1-\alpha}\mid x,\mathcal{D})$ in the original score space. That per-input quantile is the mechanism that turns 'large posterior spread at $x$' into 'widened interval at $x$'; when the posterior concentrates on the true score distribution, the transform reduces to the estimated-CDF score whose asymptotic conditional coverage is already known.

What would settle it

Simulate a regression problem with a strongly skewed, heavy-tailed conditional score distribution and a clearly defined data-sparse region, then fit EPICSCORE with a normal-likelihood Gaussian process, which cannot converge to that true score distribution. If empirical conditional coverage in the sparse region fails to approach $1-\alpha$ as the calibration sample grows, Assumption 1 is violated and the asymptotic conditional coverage claim fails for misspecified Bayesian models; if it does approach $1-\alpha$, the claim survives this particular misspecification.

Watch

Extended reading notes

Core claim

The central claim is that mapping a conformal score through its Bayesian posterior predictive CDF, $s'(x,y)=F(s(x,y)\mid x,\mathcal{D})$, converts epistemic uncertainty into a learned, feature-dependent cutoff: the prediction region becomes $\{y : s(x,y) \le F^{-1}(t_{1-\alpha}\mid x,\mathcal{D})\}$, so where the Bayesian model's predictive distribution is wide, the effective threshold is automatically looser. Theorem 1 proves the transformed score preserves finite-sample marginal coverage for any Bayesian model, because the exchangeability argument behind split conformal prediction is untouched by the transform. Theorem 2 proves that when the predictive CDF converges uniformly in probability to the true conditional distribution of the score (Assumption 1), the regions attain asymptotic conditional coverage, $P(Y\in R_{\mathrm{EPIC}}(X)\mid X=x) \to 1-\alpha$, recovering in the limit the CDF-transformed conformal score of Dheur et al. (2025).

Load-bearing premise

The load-bearing premise is Assumption 1: the Bayesian model's predictive distribution of the conformal score must converge uniformly in probability to the true score distribution as the fitting sample grows — if the model family is wrong and this convergence fails, the asymptotic conditional coverage claim (Theorem 2) collapses, although the always-valid finite-sample marginal coverage (Theorem 1) remains.

Editorial extensions

If this is right

  • Any conformal score — regression residuals, CQR, APS, or density-based scores — can be wrapped with the same recipe, and the marginal coverage guarantee of Theorem 1 holds regardless of which Bayesian model is used.
  • The effective cutoff $F^{-1}(t_{1-\alpha}\mid x,\mathcal{D})$ is a learned function of the features, so the same calibration quantile produces wide intervals where data are scarce and narrow ones where data are abundant.
  • As the calibration sample grows and the Bayesian model converges to the true score distribution, EPICSCORE inherits asymptotic conditional coverage, a property that standard split conformal prediction lacks.
  • In classification, EPICSCORE equals APS computed with predictive probabilities rather than point estimates, and the CIFAR-100 experiments report prediction sets that expand for outlier images, contract for in-distribution images, and improve size-stratified coverage over plain APS.

Reading between the lines

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

  • The 'epistemic uncertainty' EPICSCORE reports is entirely inherited from the Bayesian model's posterior spread: a misspecified model family still produces adaptive intervals, but the widening may occur in the wrong places, and only the marginal guarantee remains intact.
  • The paper's own prior-concentration experiment shows a diffuse prior widens sparse-region intervals while a concentrated prior regularizes them, which suggests a practical tuning recipe — selecting the Bayesian prior to optimize a coverage-sharpness metric such as AISL — that the paper mentions but does not develop.
  • Because the classification variant produces the same score for both standard APS scoring functions, the adaptivity is carried by the predictive distribution $P(y\mid x,\mathcal{D})$, not by the base score, hinting the wrapper applies equally to other discrete structured-prediction spaces.
  • The paper lists distribution shift as future work; a direct test would check whether the widened sparse-region intervals are exactly the ones that preserve coverage when the test distribution shifts toward those regions.
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

3 major / 5 minor

Summary. The paper introduces EPICSCORE, a wrapper that replaces any conformal score s(x,y) by its Bayesian posterior predictive CDF, s'(x,y)=F(s(x,y)|x,D), where D is a subset of the calibration set. The new score is then used in standard split conformal prediction. The authors prove finite-sample marginal coverage (Theorem 1) and claim asymptotic conditional coverage under a uniform-convergence assumption on the posterior predictive CDF (Theorem 2). They instantiate the method with BART, variational Gaussian processes, and MDNs with MC dropout, and report experiments on 13 regression/quantile-regression benchmarks plus a CIFAR-100 classification study.

Significance. The finite-sample marginal-coverage guarantee is genuine and correctly obtained: the Bayesian model is fitted on Dcal,1 while the conformal quantile is computed on the disjoint set Dcal,2, so the exchangeability argument is not circular. The paper also ships code and uses 50 repeated runs, which is a strength. If the asymptotic conditional-coverage claim were fully established, EPICSCORE would be a useful model-agnostic tool for widening conformal regions in data-sparse areas. However, the asymptotic claim currently rests on Assumption 1, which is an oracle correct-specification and uniform-concentration condition that is not verified for, and is unlikely to hold exactly for, the approximate Bayesian models used in the experiments. The contribution is therefore promising but its headline theoretical property is not yet established at the level claimed.

major comments (3)
  1. [Section 3, Assumption 1 and Theorem 2] The asymptotic conditional-coverage claim is conditional on an assumption that is not satisfied by the algorithm as implemented. MC dropout is a variational approximation rather than exact posterior inference, and the Gaussian, mixture, and BART families are not plausible exact models for absolute-residual or quantile-based conformal scores. The proof requires that F(s(X,Y)|X,theta*) be the exact conditional CDF of the score, because the uniformity step P(s''(X,Y)<=u|X)=u is used. For a misspecified family, s'' is not marginally uniform and the proof collapses. Please either state Theorem 2 explicitly as a correct-specification oracle result and reword the abstract and contributions accordingly, or provide checkable sufficient conditions and empirical validation that the condition holds approximately for at least one instantiation.
  2. [Appendix E, proof of Theorem 2] The proof contains a technical gap in the final probability calculation. The equality P(s''(X,Y)<=t''_{1-alpha}+2*epsilon|X)=t''_{1-alpha}+2*epsilon is not valid when t''_{1-alpha}+2*epsilon>1, since the uniform CDF is capped at 1. Moreover, t''_{1-alpha} is a random variable depending on Dcal,2, so the conditional probability should be E[ min(t''_{1-alpha}+2*epsilon,1) | X ] rather than t''_{1-alpha}+2*epsilon. The lower-bound half of the claimed absolute difference is also not derived. As written, the conclusion |P(s'(X,Y)<=t'_{1-alpha}|X)-t''_{1-alpha}|<=2*epsilon+delta is not proven.
  3. [Section 3, Assumption 1] The 'true parameter' theta* is never formally defined. The conformal score s(x,y) itself depends on Dtrain, so the target F(s|x,theta*) is not a fixed object if Dtrain is allowed to grow; the theorem only sends |Dcal| to infinity, but the paper should state that Dtrain is held fixed and should define theta* relative to that fixed score. In addition, Assumption 1 demands sup-norm uniform convergence of a posterior predictive CDF, which is not a standard consequence of Bayesian consistency for the nonparametric families used, and no theorem or reference establishing it for BART, variational GPs, or MC-dropout MDNs is provided.
minor comments (5)
  1. [Abstract and Section 3] The phrase 'distribution-free guarantees' should be qualified: finite-sample marginal coverage is distribution-free, but the asymptotic conditional-coverage result is not distribution-free and depends on Assumption 1.
  2. [Section 4.3 and Table 3] The claim that EPICSCORE 'consistently achieves higher' outlier-to-inlier interval length ratios is overstated; for example, on bike and cycle the EPIC-MDN ratio is below 1 and below several baselines.
  3. [Section 2.2] The equivalent expression R_EPIC(x)={y: s(x,y)<=F^{-1}(t_{1-alpha}|x,D)} assumes invertibility of the predictive CDF; for discrete scores, as in the classification special case, the direct definition via s'(x,y) should be used.
  4. [Appendix D.5] If the AISL metric is used to tune the BART prior hyperparameter beta, the paper should specify which data split is used for computing AISL, to avoid tuning on the test set.
  5. [Throughout] There are scattered typographical errors, such as 'predtion' in Appendix C.1 and 'comparisson' in Appendix D.6, which should be corrected.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the score transformation is fitted on a separate calibration subset, the coverage quantile is computed on a disjoint subset, and the asymptotic conditional-coverage theorem rests on an explicit consistency assumption plus an external lemma, not on a self-citation chain.

full rationale

I walked the paper's derivation chain. The central object s'(x,y)=F(s(x,y)|x,D) is obtained by fitting a Bayesian predictive CDF on Dcal,1 (Algorithm 1, Step II.2), and the conformal threshold is computed from s' evaluated on a disjoint subset Dcal,2 (Algorithm 1, Step II.3-4). Marginal coverage (Theorem 1) thus follows from standard split-conformal exchangeability (cited to Lei et al. 2018), not from the Bayesian fit — the target coverage is not used as a fitting objective. The asymptotic conditional-coverage claim (Theorem 2) requires Assumption 1, which is stated explicitly as uniform convergence in probability of F(s|x,D) to F(s|x,theta*), with theta* described as the true parameter. The proof shows the empirical quantile of s' is close to the empirical quantile of s''=F(s|x,theta*), then invokes Dheur et al. [2025, Lemma 2] for convergence of the latter quantile to 1-alpha; the uniformity of s'' given X is the standard probability-integral-transform fact following from the assumption. This is a conditional derivation with an identifiable, non-tautological hypothesis; the conclusion is not identical to the assumption, and the Bayesian component is not fitted to coverage. I found no load-bearing self-citation: the authors' own prior work (Cabezas et al. 2024, 2025; Izbicki et al. 2020, 2022) appears in related-work comparisons and as possible extensions, not as justification for Theorem 1 or Theorem 2. The proof's reliance on Dheur et al.'s Lemma 2 is an external result that is not reproduced in the paper, but it is not a self-citation and is not used to forbid alternatives. The main caveat is substantive, not circular: Assumption 1 is strong and is admitted by the authors to be strong, and the MC-dropout/MDN and heteroscedastic BART instantiations are not shown to satisfy it; the paper offers only qualitative empirical robustness. That is a correctness/verification limitation, not a circularity of the derivation. Overall, the paper is self-contained against external benchmarks and its theoretical claim is conditional on a clearly stated assumption, so no circular step reaches the threshold required by the review rules.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The method adds a Bayesian model of the conformal score distribution; the listed hyperparameters are hand-chosen and affect behavior, and the theorems rely on exchangeability plus a uniform-convergence assumption on the Bayesian predictive CDF. No new entities are introduced.

free parameters (3)
  • BART prior depth parameter beta = 0.9 (diffuse) or 0.1 (concentrated) in Appendix D.5; unspecified for main experiments
    Chosen by hand; Appendix D.5 shows it materially changes interval width and AISL, so the method's adaptivity depends on this prior choice.
  • MDN mixture components K = 3
    Fixed by hand; the predictive CDF of scores depends on this capacity choice.
  • GP inducing points = 15, 50, or 150 depending on n
    Set by data size; affects the fidelity of the variational posterior predictive.
assumptions (4)
  • domain assumption Exchangeability of the data (i.i.d. split into training, Dcal1, Dcal2, test)
    Standard conformal prediction assumption; invoked in Theorem 1 and Theorem 2.
  • domain assumption Assumption 1: uniform convergence in probability of the Bayesian predictive CDF F(s|x,D) to the true conditional CDF F(s|x,theta*)
    Load-bearing condition for Theorem 2; requires the Bayesian model to be well-specified and the posterior to concentrate uniformly, which is strong and typically unverifiable.
  • domain assumption The conformal score s has a continuous joint distribution for the upper coverage bound in Theorem 1
    Stated in Theorem 1; used only for the upper bound.
  • standard math Dheur et al. (2025, Lemma 2): the empirical quantile of the true CDF-transform score tends to 1-alpha
    Used in the proof of Theorem 2; treated as an established background result.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Epistemic Uncertainty in Conformal Scores: A Unified Approach." pith.science (2026). https://pith.science/paper/ZL2RFNN6

@misc{pith2026250206995,
  author       = {Pith},
  title        = {Pith review of: Epistemic Uncertainty in Conformal Scores: A Unified Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZL2RFNN6}},
  note         = {Machine review of arXiv:2502.06995}
}
abstract

Conformal prediction methods create prediction bands with distribution-free guarantees but do not explicitly capture epistemic uncertainty, which can lead to overconfident predictions in data-sparse regions. Although recent conformal scores have been developed to address this limitation, they are typically designed for specific tasks, such as regression or quantile regression. Moreover, they rely on particular modeling choices for epistemic uncertainty, restricting their applicability. We introduce $\texttt{EPICSCORE}$, a model-agnostic approach that enhances any conformal score by explicitly integrating epistemic uncertainty. Leveraging Bayesian techniques such as Gaussian Processes, Monte Carlo Dropout, or Bayesian Additive Regression Trees, $\texttt{EPICSCORE}$ adaptively expands predictive intervals in regions with limited data while maintaining compact intervals where data is abundant. As with any conformal method, it preserves finite-sample marginal coverage. Additionally, it also achieves asymptotic conditional coverage. Experiments demonstrate its good performance compared to existing methods. Designed for compatibility with any Bayesian model, but equipped with distribution-free guarantees, $\texttt{EPICSCORE}$ provides a general-purpose framework for uncertainty quantification in prediction problems.

Figures

Figures reproduced from arXiv: 2502.06995 by the authors.

Figure 1
Figure 1. A comparison of predictive intervals from standard [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Prediction sets from Adaptive Prediction Sets (APS) versus the proposed [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. EPICSCORE schematic illustration: Given a fitted base model (first panel), we begin by creating a nonconformity score and evaluating it over the calibration set (second panel). We then model the predictive distribution of the conformal score s(X, Y ) using a specified family of models, integrating the epistemic uncertainty about the data-generating process. The predictive CDF of each original score defines a new con… view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Additional outlier image prediction sets examples. [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]
Figure 5
Figure 5. Figure 5: Additional inlier prediction sets examples. [PITH_FULL_IMAGE:figures/full_fig_p017_5.png]
Figure 6
Figure 6. Figure 6: Left: Prediction set sizes for the top 150 most outlying observations. Right: Prediction set sizes for the top 150 [PITH_FULL_IMAGE:figures/full_fig_p018_6.png]
Figure 7
Figure 7. Figure 7: Comparison of prediction regions produced by [PITH_FULL_IMAGE:figures/full_fig_p024_7.png]
Figure 8
Figure 8. Figure 8: Variation of the Average Interval Score Loss (AISL) as a function of the β hyperparameter in the BART model used within the EPICSCORE framework. Lower AISL values indicate more favorable balance, suggesting that tuning β can significantly affect performance by controll…
Figure 9
Figure 9. Figure 9: Computational cost of EPICSCORE as a function of sample size n (log scale), using the three Bayesian predictive models: BART, MDN and Variational GP. While all three models show comparable running times at moderate scales(e.g., n = 10000), MDNs exhibit significantly be…
Figure 10
Figure 10. Figure 10: Comparison of prediction regions generated by the HPD (Highest Predictive Density) split [Izbicki et al., 2022] [PITH_FULL_IMAGE:figures/full_fig_p027_10.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Optimal Conformal Prediction under Epistemic Uncertainty

    stat.ML 2025-05 conditional novelty 6.0 of 10

    Bernoulli Prediction Sets provably produce smallest conformal prediction sets with conditional coverage when the model outputs valid credal sets, generalizing APS.

Reference graph

Works this paper leans on

77 extracted references · 44 canonical work pages · cited by 1 Pith paper

  1. [1]

    Tennessee’s student teacher achievement ratio (star) project

    Charles M Achilles, Helen Pate Bain, Fred Bellott, Jayne Boyd-Zaharias, Jeremy Finn, John Folger, John Johnston, and Elizabeth Word. Tennessee’s student teacher achievement ratio (star) project. Harvard Dataverse, 1: 0 2008, 2008

  2. [2]

    A gentle introduction to conformal prediction and distribution-free uncertainty quantification

    Anastasios N Angelopoulos and Stephen Bates. A gentle introduction to conformal prediction and distribution-free uncertainty quantification. arXiv preprint arXiv:2107.07511, 2021

  3. [3]

    Bernardo and A.F.M

    J.M. Bernardo and A.F.M. Smith. Bayesian Theory. Wiley Series in Probability and Statistics. Wiley, 2009. ISBN 9780470317716. URL https://books.google.com.br/books?id=11nSgIcd7xQC

  4. [4]

    Mixture density networks

    Christopher M Bishop. Mixture density networks. 1994

  5. [5]

    Mondrian conformal regressors

    Henrik Bostr \"o m and Ulf Johansson. Mondrian conformal regressors. In Conformal and Probabilistic Prediction and Applications, pages 114--133. PMLR, 2020

  6. [6]

    Lof: identifying density-based local outliers

    Markus M Breunig, Hans-Peter Kriegel, Raymond T Ng, and J \"o rg Sander. Lof: identifying density-based local outliers. In Proceedings of the 2000 ACM SIGMOD international conference on Management of data, pages 93--104, 2000

  7. [7]

    Distribution-free calibration of statistical confidence sets

    Luben Cabezas, Guilherme P Soares, Thiago R Ramos, Rafael B Stern, and Rafael Izbicki. Distribution-free calibration of statistical confidence sets. arXiv preprint arXiv:2411.19368, 2024

  8. [8]

    Regression trees for fast and adaptive prediction intervals

    Luben MC Cabezas, Mateus P Otto, Rafael Izbicki, and Rafael B Stern. Regression trees for fast and adaptive prediction intervals. Information Sciences, 686: 0 121369, 2025

Show all 77 references
  1. [9]

    Distributional conformal prediction

    Victor Chernozhukov, Kaspar W \"u thrich, and Yinchu Zhu. Distributional conformal prediction. Proceedings of the National Academy of Sciences, 118 0 (48): 0 e2107794118, 2021

  2. [10]

    BART : Bayesian additive regression trees

    Hugh A Chipman, Edward I George, and Robert E McCulloch. BART : Bayesian additive regression trees. Annals of Applied Statistics, 6 0 (1): 0 266--298, 2012

  3. [11]

    Uncertainty-aware online extrinsic calibration: A conformal prediction approach

    Mathieu Cocheteux, Julien Moreau, and Franck Davoine. Uncertainty-aware online extrinsic calibration: A conformal prediction approach. arXiv preprint arXiv:2501.06878, 2025

  4. [12]

    Cerdeira, F

    Paulo Cortez, A. Cerdeira, F. Almeida, T. Matos, and J. Reis. Wine Quality . UCI Machine Learning Repository, 2009. DOI : https://doi.org/10.24432/C56S3T

  5. [13]

    Amnioml: amniotic fluid segmentation and volume prediction with uncertainty quantification

    Daniel Csillag, Lucas Monteiro Paes, Thiago Ramos, Jo \ a o Vitor Romano, Rodrigo Schuller, Roberto B Seixas, Roberto I Oliveira, and Paulo Orenstein. Amnioml: amniotic fluid segmentation and volume prediction with uncertainty quantification. In Proceedings of the AAAI Confere...

  6. [14]

    Conditional density estimation tools in python and r with applications to photometric redshifts and likelihood-free cosmological inference

    Niccol \`o Dalmasso, Taylor Pospisil, Ann B Lee, Rafael Izbicki, Peter E Freeman, and Alex I Malz. Conditional density estimation tools in python and r with applications to photometric redshifts and likelihood-free cosmological inference. Astronomy and Computing, 30: 0 100362, 2020

  7. [15]

    Distribution-free conformal joint prediction regions for neural marked temporal point processes

    Victor Dheur, Tanguy Bosser, Rafael Izbicki, and Souhaib Ben Taieb. Distribution-free conformal joint prediction regions for neural marked temporal point processes. Machine Learning, 113 0 (9): 0 7055--7102, 2024

  8. [16]

    Multi-output conformal regression: A unified comparative study with new conformity scores

    Victor Dheur, Matteo Fontana, Yorick Estievenart, Naomi Desobry, and Souhaib Ben Taieb. Multi-output conformal regression: A unified comparative study with new conformity scores. arXiv preprint arXiv:2501.10533, 2025

  9. [17]

    Catboost: gradient boosting with categorical features support

    Anna Veronika Dorogush, Vasily Ershov, and Andrey Gulin. Catboost: gradient boosting with categorical features support. arXiv preprint arXiv:1810.11363, 2018

  10. [18]

    Uci machine learning repository, 2017

    Dheeru Dua, Casey Graff, et al. Uci machine learning repository, 2017. URL http://archive. ics. uci. edu/ml, 7 0 (1): 0 62, 2017

  11. [19]

    Improving conditional coverage via orthogonal quantile regression

    Shai Feldman, Stephen Bates, and Yaniv Romano. Improving conditional coverage via orthogonal quantile regression. Advances in neural information processing systems, 34: 0 2060--2071, 2021

  12. [20]

    Conformal bayesian computation

    Edwin Fong and Chris C Holmes. Conformal bayesian computation. Advances in Neural Information Processing Systems, 34: 0 18268--18279, 2021

  13. [21]

    Conformal prediction: a unified review of theory and new challenges

    Matteo Fontana, Gianluca Zeni, and Simone Vantini. Conformal prediction: a unified review of theory and new challenges. Bernoulli, 29 0 (1): 0 1--23, 2023

  14. [22]

    The limits of distribution-free conditional predictive inference

    Rina Foygel Barber, Emmanuel J Candes, Aaditya Ramdas, and Ryan J Tibshirani. The limits of distribution-free conditional predictive inference. Information and Inference: A Journal of the IMA, 10 0 (2): 0 455--482, 2021

  15. [23]

    A unified framework for constructing, tuning and assessing photometric redshift density estimates in a selection bias setting

    Peter E Freeman, Rafael Izbicki, and Ann B Lee. A unified framework for constructing, tuning and assessing photometric redshift density estimates in a selection bias setting. Monthly Notices of the Royal Astronomical Society, 468 0 (4): 0 4556--4565, 2017

  16. [24]

    Personalizedus: Interpretable breast cancer risk assessment with local coverage uncertainty quantification

    Alek Fr \"o hlich, Thiago Ramos, Gustavo Cabello, Isabela Buzatto, Rafael Izbicki, and Daniel Tiezzi. Personalizedus: Interpretable breast cancer risk assessment with local coverage uncertainty quantification. arXiv:2408.15458, 2024

  17. [25]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning

    Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning . In International Conference on Machine Learning, Proceedings of Machine Learning Research, pages 1050--1059. PMLR, 2016

  18. [26]

    Gpytorch: Blackbox matrix-matrix gaussian process inference with gpu acceleration

    Jacob Gardner, Geoff Pleiss, Kilian Q Weinberger, David Bindel, and Andrew G Wilson. Gpytorch: Blackbox matrix-matrix gaussian process inference with gpu acceleration. Advances in neural information processing systems, 31, 2018

  19. [27]

    Strictly proper scoring rules, prediction, and estimation

    Tilmann Gneiting and Adrian E Raftery. Strictly proper scoring rules, prediction, and estimation. Journal of the American statistical Association, 102 0 (477): 0 359--378, 2007

  20. [28]

    Localized conformal prediction: A generalized inference framework for conformal prediction

    Leying Guan. Localized conformal prediction: A generalized inference framework for conformal prediction. Biometrika, 110 0 (1): 0 33--50, 2023

  21. [29]

    Superconductivty Data

    Kam Hamidieh. Superconductivty Data . UCI Machine Learning Repository, 2018. DOI : https://doi.org/10.24432/C53P47

  22. [30]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770--778, 2016

  23. [31]

    A survey on uncertainty quantification methods for deep learning

    Wenchong He, Zhe Jiang, Tingsong Xiao, Zelin Xu, and Yukun Li. A survey on uncertainty quantification methods for deep learning. arXiv preprint arXiv:2302.13425, 2023

  24. [32]

    Potential of integrated field spectroscopy and spatial analysis for enhanced assessment of soil contamination: A prospective review

    Ana Horta, B Malone, U Stockmann, B Minasny, TFA Bishop, AB McBratney, R Pallasser, and L Pozza. Potential of integrated field spectroscopy and spatial analysis for enhanced assessment of soil contamination: A prospective review. Geoderma, 241: 0 180--209, 2015

  25. [33]

    Aleatoric and epistemic uncertainty in machine learning: An introduction to concepts and methods

    Eyke H \"u llermeier and Willem Waegeman. Aleatoric and epistemic uncertainty in machine learning: An introduction to concepts and methods. Machine Learning, 110: 0 457--506, 2021

  26. [34]

    Batch normalization: accelerating deep network training by reducing internal covariate shift

    Sergey Ioffe and Christian Szegedy. Batch normalization: accelerating deep network training by reducing internal covariate shift. In Proceedings of the 32nd International Conference on International Conference on Machine Learning - Volume 37, ICML'15, page 448–456. JMLR.org, 2015

  27. [35]

    Machine Learning Beyond Point Predictions: Uncertainty Quantification

    Rafael Izbicki. Machine Learning Beyond Point Predictions: Uncertainty Quantification. 1st edition, 2025. ISBN 978-65-01-20272-3

  28. [36]

    Converting high-dimensional regression to high-dimensional conditional density estimation

    Rafael Izbicki and Ann B Lee. Converting high-dimensional regression to high-dimensional conditional density estimation. Electronic Journal of Statistics, 11: 0 2800--2831, 2017

  29. [37]

    Flexible distribution-free conditional predictive bands using density estimators

    Rafael Izbicki, Gilson Shimizu, and Rafael Stern. Flexible distribution-free conditional predictive bands using density estimators. In International Conference on Artificial Intelligence and Statistics, pages 3068--3077. PMLR, 2020

  30. [38]

    Cd-split and hpd-split: Efficient conformal regions in high dimensions

    Rafael Izbicki, Gilson Shimizu, and Rafael B Stern. Cd-split and hpd-split: Efficient conformal regions in high dimensions. Journal of Machine Learning Research, 23 0 (87): 0 1--32, 2022

  31. [39]

    Conformal approach to gaussian process surrogate evaluation with coverage guarantees

    Edgar Jaber, Vincent Blot, Nicolas Brunel, Vincent Chabridon, Emmanuel Remy, Bertrand Iooss, Didier Lucor, Mathilde Mougeot, and Alessandro Leite. Conformal approach to gaussian process surrogate evaluation with coverage guarantees. arXiv preprint arXiv:2401.07733, 2024

  32. [40]

    House sales in king county, usa

    Kaggle . House sales in king county, usa. Online, 2016. URL https://www.kaggle.com/harlfoxem/housesalesprediction/metadata. Accessed: August, 2019

  33. [41]

    Adam: A method for stochastic optimization

    Diederik P Kingma. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014

  34. [42]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009

  35. [43]

    On weight initialization in deep neural networks

    Siddharth Krishna Kumar. On weight initialization in deep neural networks. arXiv preprint arXiv:1704.08863, 2017

  36. [44]

    Distribution-free prediction bands for non-parametric regression

    Jing Lei and Larry Wasserman. Distribution-free prediction bands for non-parametric regression. Journal of the Royal Statistical Society Series B: Statistical Methodology, 76 0 (1): 0 71--96, 2014

  37. [45]

    Distribution-free predictive inference for regression

    Jing Lei, Max G'Sell, Alessandro Rinaldo, Ryan J Tibshirani, and Larry Wasserman. Distribution-free predictive inference for regression. Journal of the American Statistical Association, 113 0 (523): 0 1094--1111, 2018

  38. [46]

    Awesome conformal prediction, 2024

    Valery Manokhin. Awesome conformal prediction, 2024. URL https://github.com/valeman/awesome-conformal-prediction

  39. [47]

    Bike sharing demand - rmsle: 0.3194

    Raj Mehra. Bike sharing demand - rmsle: 0.3194. https://www.kaggle.com/code/rajmehra03/bike-sharing-demand-rmsle-0-3194/input?select=train.csv, 2023. Accessed: [Insert Date Here]

  40. [48]

    A literature review of fault diagnosis based on ensemble learning

    Zhibao Mian, Xiaofei Deng, Xiaohui Dong, Yuzhu Tian, Tianya Cao, Kairan Chen, and Tareq Al Jaber. A literature review of fault diagnosis based on ensemble learning. Engineering Applications of Artificial Intelligence, 127: 0 107357, 2024

  41. [49]

    Optimisation of large wave farms using a multi-strategy evolutionary framework

    Mehdi Neshat, Bradley Alexander, Nataliia Y Sergiienko, and Markus Wagner. Optimisation of large wave farms using a multi-strategy evolutionary framework. In Proceedings of the 2020 genetic and evolutionary computation conference, pages 1150--1158, 2020

  42. [50]

    Inductive confidence machines for regression

    Harris Papadopoulos, Kostas Proedrou, Volodya Vovk, and Alex Gammerman. Inductive confidence machines for regression. In Machine learning: ECML 2002: 13th European conference on machine learning Helsinki, Finland, August 19--23, 2002 proceedings 13, pages 345--356. Springer, 2002

  43. [51]

    Gaussian process interpolation with conformal prediction: methods and comparative analysis

    Aur \'e lien Pion and Emmanuel Vazquez. Gaussian process interpolation with conformal prediction: methods and comparative analysis. In LOD 2024, 10th International Conference on Machine Learning, Optimization, and Data Science, 2024

  44. [52]

    Conditionally valid probabilistic conformal prediction

    Vincent Plassier, Alexander Fishkov, Maxim Panov, and Eric Moulines. Conditionally valid probabilistic conformal prediction. stat, 1050: 0 1, 2024

  45. [53]

    Heteroscedastic bart via multiplicative regression trees

    Matthew T Pratola, Hugh A Chipman, Edward I George, and Robert E McCulloch. Heteroscedastic bart via multiplicative regression trees. Journal of Computational and Graphical Statistics, 29 0 (2): 0 405--417, 2020

  46. [54]

    Bayesian additive regression trees for probabilistic programming

    Miriana Quiroga, Pablo G Garay, Juan M Alonso, Juan Martin Loyola, and Osvaldo A Martin. Bayesian additive regression trees for probabilistic programming. arXiv preprint arXiv:2206.03619, 2022

  47. [55]

    Physicochemical Properties of Protein Tertiary Structure

    Prashant Rana. Physicochemical Properties of Protein Tertiary Structure . UCI Machine Learning Repository, 2013. DOI : https://doi.org/10.24432/C5QW3H

  48. [56]

    Conformalized quantile regression

    Yaniv Romano, Evan Patterson, and Emmanuel Cand\` e s. Conformalized quantile regression . In Advances in Neural Information Processing Systems, volume 32, pages 3543--3553. Curran Associates, Inc., 2019

  49. [57]

    Classification with valid and adaptive coverage

    Yaniv Romano, Matteo Sesia, and Emmanuel Candes. Classification with valid and adaptive coverage. Advances in Neural Information Processing Systems, 33: 0 3581--3591, 2020

  50. [58]

    Integrating uncertainty awareness into conformalized quantile regression

    Raphael Rossellini, Rina Foygel Barber, and Rebecca Willett. Integrating uncertainty awareness into conformalized quantile regression. In International Conference on Artificial Intelligence and Statistics, pages 1540--1548. PMLR, 2024

  51. [59]

    Natural gradients in practice: Non-conjugate variational inference in gaussian process models

    Hugh Salimbeni, Stefanos Eleftheriadis, and James Hensman. Natural gradients in practice: Non-conjugate variational inference in gaussian process models. In International Conference on Artificial Intelligence and Statistics, pages 689--697. PMLR, 2018

  52. [60]

    Theory of statistics

    Mark J Schervish. Theory of statistics. Springer Science & Business Media, 2012

  53. [61]

    S. J. Schmidt , A. I. Malz , J. Y. H. Soo , I. A. Almosallam , M. Brescia , S. Cavuoti , J. Cohen-Tanugi , A. J. Connolly , J. DeRose , P. E. Freeman , M. L. Graham , K. G. Iyer , M. J. Jarvis , J. B. Kalmbach , E. Kovacs , A. B. Lee , G. Longo , C. B. Morrison , J. A. Newman ...

  54. [62]

    A tutorial on gaussian process regression: Modelling, exploring, and exploiting functions

    Eric Schulz, Maarten Speekenbrink, and Andreas Krause. A tutorial on gaussian process regression: Modelling, exploring, and exploiting functions. Journal of mathematical psychology, 85: 0 1--16, 2018

  55. [63]

    A comparison of some conformal quantile regression methods

    Matteo Sesia and Emmanuel J Cand \`e s. A comparison of some conformal quantile regression methods. Stat, 9 0 (1): 0 e261, 2020

  56. [64]

    A tutorial on conformal prediction

    Glenn Shafer and Vladimir Vovk. A tutorial on conformal prediction. Journal of Machine Learning Research, 9 0 (3), 2008

  57. [65]

    Dropout: a simple way to prevent neural networks from overfitting

    Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: a simple way to prevent neural networks from overfitting. The journal of machine learning research, 15 0 (1): 0 1929--1958, 2014

  58. [66]

    Bayesian uncertainty estimation for batch normalized deep networks

    Mattias Teye, Hossein Azizpour, and Kevin Smith. Bayesian uncertainty estimation for batch normalized deep networks. In International Conference on Machine Learning, pages 4907--4916. PMLR, 2018

  59. [67]

    Combined Cycle Power Plant

    Pnar Tfekci and Heysem Kaya. Combined Cycle Power Plant . UCI Machine Learning Repository, 2014. DOI : https://doi.org/10.24432/C5002N

  60. [68]

    A review of predictive uncertainty estimation with machine learning

    Hristos Tyralis and Georgia Papacharalampous. A review of predictive uncertainty estimation with machine learning. Artificial Intelligence Review, 57 0 (4): 0 94, 2024

  61. [69]

    Valle, R

    D. Valle, R. Leite, Rafael Izbicki, C. Silva, and L. Haneda. Local uncertainty maps for land-use/land-cover classification without remote sensing and modeling work using a class-conditional conformal approach. International Journal of Applied Earth Observation and Geoinformati...

  62. [70]

    Quantifying uncertainty in land-use land-cover classification using conformal statistics

    Denis Valle, Rafael Izbicki, and Rodrigo Vieira Leite. Quantifying uncertainty in land-use land-cover classification using conformal statistics. Remote Sensing of Environment, 295: 0 113682, 2023

  63. [71]

    Local uncertainty maps for land-use/land-cover classification without remote sensing and modeling work using a class-conditional conformal approach

    Denis Valle, Rodrigo Leite, Rafael Izbicki, Carlos Silva, and Leo Haneda. Local uncertainty maps for land-use/land-cover classification without remote sensing and modeling work using a class-conditional conformal approach. International Journal of Applied Earth Observation and...

  64. [72]

    Visualizing data using t-sne

    Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of machine learning research, 9 0 (11), 2008

  65. [73]

    Algorithmic learning in a random world, volume 29

    Vladimir Vovk, Alexander Gammerman, and Glenn Shafer. Algorithmic learning in a random world, volume 29. Springer, 2005

  66. [74]

    From aleatoric to epistemic: Exploring uncertainty quantification techniques in artificial intelligence

    Tianyang Wang, Yunze Wang, Jun Zhou, Benji Peng, Xinyuan Song, Charles Zhang, Xintian Sun, Qian Niu, Junyu Liu, Silin Chen, et al. From aleatoric to epistemic: Exploring uncertainty quantification techniques in artificial intelligence. arXiv preprint arXiv:2501.03282, 2025

  67. [75]

    Frasian inference

    Larry Wasserman. Frasian inference. 2011

  68. [76]

    Gaussian processes for machine learning, volume 2

    Christopher KI Williams and Carl Edward Rasmussen. Gaussian processes for machine learning, volume 2. MIT press Cambridge, MA, 2006

  69. [77]

    Concrete Compressive Strength

    I-Cheng Yeh. Concrete Compressive Strength . UCI Machine Learning Repository, 1998. DOI : https://doi.org/10.24432/C5PK67

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.