pith. machine review for the scientific record. sign in

arxiv: 2605.01868 · v1 · submitted 2026-05-03 · 💻 cs.LG

Recognition: unknown

Robust Conditional Conformal Prediction via Branched Normalizing Flow

Rui Xu , Xingyuan Chen , Wenxing Huang , Minxuan Huang , Weiyan Chen , Sihong Xie , Hui Xiong
Authors on Pith no claims yet

Pith reviewed 2026-05-10 16:25 UTC · model grok-4.3

classification 💻 cs.LG
keywords conformal predictiondistribution shiftnormalizing flowsconditional coverageWasserstein distanceprediction setsrobustness
0
0 comments X

The pith

Branched normalizing flow transports test inputs and prediction sets to preserve conditional coverage in conformal prediction under distribution shift.

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

The paper establishes a bound on the conditional invalidity of conformal prediction when calibration and test distributions differ, expressed directly in terms of their Wasserstein distance. This bound isolates the effect of mismatched conditional score distributions at individual test points and shows that an invertible transport map can control the resulting coverage error. The authors therefore construct a two-branch normalizing flow that maps a test input into the calibration domain, forms the conformal set there, and maps the set back while retaining the original conditional guarantees. The approach is shown to deliver improved conditional coverage across nine datasets and many confidence levels, addressing the gap left by methods that only match marginal score distributions.

Core claim

We bound the conditional invalidity of CP under distribution shift in terms of the Wasserstein distance between the calibration and test distributions. This result highlights the role of invertible transport in mitigating conditional coverage degradation. Motivated by this insight, we introduce Branched Normalizing Flow (BNF), a two-branch architecture that normalizes a test input to the calibration distribution and transforms the prediction set of the normalized input back to the test distribution while preserving conditional guarantees.

What carries the argument

Branched Normalizing Flow (BNF), a two-branch architecture that normalizes a test input to the calibration distribution, applies conformal prediction there, and maps the resulting set back to the test distribution while preserving conditional guarantees.

If this is right

  • Conditional coverage error scales with Wasserstein distance, so smaller shifts produce tighter guarantees.
  • BNF maintains the original conditional coverage guarantees after transport rather than only marginal coverage.
  • The method yields measurable gains in conditional coverage robustness across nine datasets and a wide range of confidence levels.
  • Marginal coverage remains intact while conditional reliability at individual test points improves.

Where Pith is reading between the lines

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

  • The same transport principle could be applied to other uncertainty methods that currently guarantee only marginal rather than conditional coverage.
  • When the learned flow is sufficiently accurate, the approach may extend to covariate shifts that are not purely location-scale.
  • The Wasserstein bound supplies a practical diagnostic: if the estimated distance is large, the user knows conditional coverage will degrade even with perfect transport.

Load-bearing premise

An accurate invertible transport map between calibration and test distributions can be learned by the branched flow without introducing errors that break the conditional coverage guarantee.

What would settle it

Empirical observation that the conditional coverage error after BNF transport exceeds the Wasserstein-derived bound on a dataset with known shift would falsify the central bound and the preservation claim.

Figures

Figures reproduced from arXiv: 2605.01868 by Hui Xiong, Minxuan Huang, Rui Xu, Sihong Xie, Weiyan Chen, Wenxing Huang, Xingyuan Chen.

Figure 1
Figure 1. Figure 1: (a) Vanilla prediction set CM(x) has constant size and offers marginal coverage, which is robust if conformal score distributions PV and QV have similar cumulative probabilities at τ . Conditional prediction set CC(x) has input-dependent size and provides conditional guarantees for calibration inputs Xi = x where i = 1, ..., n, but may fail on non-i.i.d. test input Xn+1 = x. The difference between PV and Q… view at source ↗
Figure 2
Figure 2. Figure 2: Invertible generative and normalizing flows. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Marginal coverage, WSC, and WSCG of Augmented BNF+CQR and baselines with [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of Augmented BNF + CQR under distribution shift with CQR in shifted and i.i.d. settings. Since Eq. (26) is optimized empirically, the transformed test data may not perfectly follow the calibration distribution. As a result, Aug￾mented BNF cannot fully eliminate WSCG or exactly achieve the target marginal coverage in [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of conditioning the transformation [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Standard Augmented BNF can produce large prediction sets under MSDG. The Augment-Conditioned [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Distribution shift is induced by label perturbation. Our method consistently maintains comparably [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Characterization of preimage membership under [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Preserving conditional coverage via implicit dependency on test input. The circled numbers indicate [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: WSCG of BNFs constructed with planar and residual flows, compared with Augmented BNF at [PITH_FULL_IMAGE:figures/full_fig_p021_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Illustration of the coupling layer structure and the overall composition of Augmented BNF. [PITH_FULL_IMAGE:figures/full_fig_p022_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Comparison between baselines and the proposed method via a toy example. Augmented BNF [PITH_FULL_IMAGE:figures/full_fig_p024_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Multi-source domain generalization. The test distribution QXY is a random mixture of source distributions, while the calibration distribution PXY is a fixed known distribution. As a result, a distribution shift occurs since PXY ̸= QXY . J.1 Synthetic distribution shifts The Physicochemical Properties of Protein Tertiary Structure (PTS) dataset [9] contains 45,730 instances, with the target variable being … view at source ↗
Figure 14
Figure 14. Figure 14: Empirical data distributions of source domains after applying t-SNE. [PITH_FULL_IMAGE:figures/full_fig_p026_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Generalization performance with dif￾ferent numbers of source domains. To explore the generalization ability of Augmented BNF under varying numbers of source domains, we modified the sampling procedure in Appendix J by changing K ∈ {2, 3, 4, 6, 8, 12}. For each value of K, we generated 10 independent trials using the PTS dataset to account for sampling variability. Augmented BNF combined with CQR was appli… view at source ↗
Figure 16
Figure 16. Figure 16: Impact of data availability on Aug￾mented BNF approximation ability. Generative models may struggle to approximate under￾lying distributions when data are limited, especially in high dimensions [74, 75]. To assess how sample size affects the performance of Augmented BNF, we vary the number of samples in SDk and perform 10 trials for each setting on the PTS dataset. For each trial, we apply Augmented BNF+C… view at source ↗
Figure 17
Figure 17. Figure 17: Marginal coverage achieved by C aug BNF(Xn+1) compared with the proposed marginal lower bound. M.2 Conditional coverage lower bound Next, we establish a conditional coverage lower bound that accounts for the imperfect alignment between calibration and test data induced by Augmented BNF. First, even in the i.i.d. setting, exact conditional coverage is unattainable with finite samples [5, 76, 28]. For insta… view at source ↗
Figure 18
Figure 18. Figure 18: Prediction set size comparison. The standard Augmented BNF can produce large prediction sets, [PITH_FULL_IMAGE:figures/full_fig_p029_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: WSCG of Augmented BNF + CQR and baselines for [PITH_FULL_IMAGE:figures/full_fig_p030_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Comparison of conditioning the transformation of label [PITH_FULL_IMAGE:figures/full_fig_p031_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Distribution shift is induced by label perturbation. Compared with baselines, our method consistently [PITH_FULL_IMAGE:figures/full_fig_p032_21.png] view at source ↗
Figure 11
Figure 11. Figure 11: This necessitates a feature space X of at least two dimensions. Consequently, Augmented BNF inherits this constraint: its architecture presumes that the input feature x is multivariate. 32 [PITH_FULL_IMAGE:figures/full_fig_p032_11.png] view at source ↗
Figure 22
Figure 22. Figure 22: Ablation study on the unbiased setting where a strict data split is conducted. The WSCG metric [PITH_FULL_IMAGE:figures/full_fig_p033_22.png] view at source ↗
read the original abstract

Conformal prediction (CP) constructs prediction sets with marginal coverage guarantees under the assumption that the calibration and test distributions are identical. However, under distribution shift, existing approaches primarily align marginal conformal score distributions, which is sufficient to preserve marginal coverage but does not control the conditional coverage error at individual test inputs. As a consequence, CP can remain unreliable in regions where the conditional score distributions are mismatched. In this work, we bound the conditional invalidity of CP under distribution shift in terms of the Wasserstein distance between the calibration and test distributions. This result highlights the role of invertible transport in mitigating conditional coverage degradation. Motivated by this insight, we introduce Branched Normalizing Flow (BNF), a two-branch architecture that normalizes a test input to the calibration distribution and transforms the prediction set of the normalized input back to the test distribution while preserving conditional guarantees. Empirically, BNF consistently improves conditional coverage robustness on nine datasets across a wide range of confidence levels.

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

3 major / 2 minor

Summary. The paper claims to derive a bound on the conditional invalidity of conformal prediction (CP) under distribution shift in terms of the Wasserstein distance between the joint calibration and test distributions. It introduces Branched Normalizing Flow (BNF), a two-branch architecture that learns an invertible transport to normalize test inputs to the calibration distribution, constructs the conformal set there, and pushes the set forward via the inverse map to the test distribution while preserving conditional coverage. Empirical evaluation on nine datasets is reported to show consistent improvements in conditional coverage robustness across confidence levels.

Significance. If the Wasserstein bound is rigorously established and the BNF approximation errors are controlled so that they do not invalidate the conditional guarantees, the work would provide a principled mechanism for achieving conditional coverage under shift, going beyond marginal score alignment. The empirical gains on multiple datasets indicate potential practical value, and the use of normalizing flows for exact transport is a targeted architectural contribution. However, the absence of derivation details and error propagation analysis in the current form limits the immediate impact.

major comments (3)
  1. [Theoretical analysis (bound on conditional invalidity)] The central theoretical claim bounds conditional invalidity via Wasserstein distance between calibration and test distributions (presumably via optimal transport of score functions). No derivation steps or key lemmas are provided, making it impossible to verify whether the bound correctly controls conditional coverage error at individual test inputs rather than only marginally.
  2. [Branched Normalizing Flow architecture and guarantee preservation] The BNF construction assumes the learned branched flow realizes an exact invertible transport map T that normalizes test points, obtains the conformal set under the calibration measure, and pushes it back via T^{-1} while exactly preserving conditional coverage. Because flow parameters are fit on finite data, the learned map is only approximate; no error term propagates the approximation error of T_hat through the coverage statement or shows that the empirical Wasserstein bound remains valid under the surrogate map. This is load-bearing for the robustness claim.
  3. [Experiments and evaluation] The empirical section reports gains on nine datasets but provides no description of how conditional coverage error is measured at individual test inputs, which baselines are used, or any statistical tests. Without these, the data-to-claim link cannot be verified and the results cannot be interpreted as evidence that BNF delivers the promised robustness.
minor comments (2)
  1. [Method] Clarify the precise definition of the branched flow architecture (e.g., how the two branches share parameters or enforce invertibility) and any regularization used to promote measure preservation.
  2. [Discussion] Add a limitations paragraph discussing when the learned transport may fail to be sufficiently accurate (e.g., high-dimensional inputs or complex shifts) and how this affects the practical utility of the Wasserstein bound.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback, which highlights important areas for strengthening the manuscript. We will revise to include the full theoretical derivation, an error propagation analysis for the approximate transport map, and detailed experimental protocols with statistical tests. These changes directly address the concerns while preserving the core contributions.

read point-by-point responses

  1. Referee: The central theoretical claim bounds conditional invalidity via Wasserstein distance between calibration and test distributions (presumably via optimal transport of score functions). No derivation steps or key lemmas are provided, making it impossible to verify whether the bound correctly controls conditional coverage error at individual test inputs rather than only marginally.

    Authors: We agree that the derivation must be fully explicit. The bound follows from applying the Kantorovich-Rubinstein duality to the difference between conditional coverage probabilities under the calibration and test measures, using the Lipschitz continuity of the conformal score function with respect to the input. A key supporting lemma shows that the Wasserstein distance between the joint distributions induces a pointwise bound on the conditional coverage gap for each fixed test input x via integration against the conditional density. We will insert the complete proof, including all lemmas and the explicit dependence on the Wasserstein distance, as a new appendix section in the revision. revision: yes

  2. Referee: The BNF construction assumes the learned branched flow realizes an exact invertible transport map T that normalizes test points, obtains the conformal set under the calibration measure, and pushes it back via T^{-1} while exactly preserving conditional coverage. Because flow parameters are fit on finite data, the learned map is only approximate; no error term propagates the approximation error of T_hat through the coverage statement or shows that the empirical Wasserstein bound remains valid under the surrogate map. This is load-bearing for the robustness claim.

    Authors: This is a valid point. The current manuscript treats the learned flow as an approximation to the ideal transport without quantifying the residual. In revision we will add a theorem that decomposes the total conditional coverage error into the Wasserstein term plus an additive error controlled by the total variation distance between the learned pushforward and the true calibration measure. We further show that this additional term vanishes as the flow training loss (negative log-likelihood on the calibration data) approaches zero, and we include a brief empirical check of flow reconstruction error on held-out calibration points. revision: yes

  3. Referee: The empirical section reports gains on nine datasets but provides no description of how conditional coverage error is measured at individual test inputs, which baselines are used, or any statistical tests. Without these, the data-to-claim link cannot be verified and the results cannot be interpreted as evidence that BNF delivers the promised robustness.

    Authors: We will expand the experimental section with a dedicated subsection. Conditional coverage error is computed by binning test inputs according to a kernel density estimate of the calibration distribution and reporting the average absolute deviation from the target coverage level within each bin; results are averaged over 10 random train-calibration-test splits. Baselines comprise standard split conformal prediction, marginal score alignment via quantile regression, and two recent conditional conformal methods. We will report mean and standard deviation across splits together with paired Wilcoxon signed-rank tests (p-values) to establish statistical significance of the observed improvements. revision: yes

Circularity Check

0 steps flagged

No circularity: bound derived from external Wasserstein theory; BNF approximates transport without reducing guarantee to fitted input

full rationale

The paper first states a bound on conditional invalidity expressed in terms of the Wasserstein distance between calibration and test distributions. This bound is presented as a general result from optimal transport and does not depend on the BNF parameters or any fitted quantity defined inside the method. The BNF is then introduced as a practical architecture to realize an approximate invertible map that pushes the conformal set forward while aiming to preserve the conditional property. No equation equates the coverage guarantee to a self-defined score or renames a fitted residual as a prediction. No self-citation is used to justify a uniqueness claim or to smuggle an ansatz. The derivation therefore remains self-contained against external benchmarks (Wasserstein geometry and standard normalizing-flow invertibility) rather than collapsing to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the existence of a learnable invertible transport between calibration and test distributions and on the validity of the Wasserstein-based bound on conditional invalidity. Because only the abstract is available, the precise assumptions and any fitted parameters inside the flow are not specified.

axioms (1)
  • domain assumption An invertible transport map between calibration and test distributions exists and can be approximated by a branched normalizing flow without destroying conditional coverage properties.
    The method explicitly relies on this map to normalize inputs and map sets back while preserving guarantees.
invented entities (1)
  • Branched Normalizing Flow (BNF) no independent evidence
    purpose: Two-branch architecture that performs normalization of test inputs to calibration space and inverse transformation of prediction sets back to test space.
    BNF is introduced in the paper as the core technical contribution; no independent evidence outside the empirical results is provided.

pith-pipeline@v0.9.0 · 5480 in / 1482 out tokens · 46531 ms · 2026-05-10T16:25:31.284728+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

81 extracted references · 12 canonical work pages · 3 internal anchors

  1. [1]

    Springer, 2005

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

    2005

  2. [2]

    A tutorial on conformal prediction, 2007

    Glenn Shafer and Vladimir V ovk. A tutorial on conformal prediction, 2007

    2007

  3. [3]

    Tibshirani, and Larry Wasserman

    Jing Lei, Max G’Sell, Alessandro Rinaldo, Ryan J. Tibshirani, and Larry Wasserman. Distribution-free predictive inference for regression, 2017

    2017

  4. [4]

    Regression conformal prediction with nearest neighbours.Journal of Artificial Intelligence Research, 40:815–840, 2011

    Harris Papadopoulos, Vladimir V ovk, and Alex Gammerman. Regression conformal prediction with nearest neighbours.Journal of Artificial Intelligence Research, 40:815–840, 2011

    2011

  5. [5]

    Conditional validity of inductive conformal predictors

    Vladimir V ovk. Conditional validity of inductive conformal predictors. InAsian conference on machine learning, pages 475–490. PMLR, 2012

    2012

  6. [6]

    Out-of-distribution generalization via risk ex- trapolation (rex)

    David Krueger, Ethan Caballero, Joern-Henrik Jacobsen, Amy Zhang, Jonathan Binas, Dinghuai Zhang, Remi Le Priol, and Aaron Courville. Out-of-distribution generalization via risk ex- trapolation (rex). InInternational conference on machine learning, pages 5815–5826. PMLR, 2021

    2021

  7. [7]

    Conformal prediction beyond exchangeability.The Annals of Statistics, 51(2):816–845, 2023

    Rina Foygel Barber, Emmanuel J Candes, Aaditya Ramdas, and Ryan J Tibshirani. Conformal prediction beyond exchangeability.The Annals of Statistics, 51(2):816–845, 2023

    2023

  8. [8]

    Wasserstein-regularized conformal prediction under general distribution shift

    Rui Xu, Chao Chen, Yue Sun, Parvathinathan Venkitasubramaniam, and Sihong Xie. Wasserstein-regularized conformal prediction under general distribution shift. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  9. [9]

    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

    work page doi:10.24432/c5qw3h 2013

  10. [10]

    Bike Sharing

    Hadi Fanaee-T. Bike Sharing. UCI Machine Learning Repository, 2013. DOI: https://doi.org/10.24432/C5W894

    work page doi:10.24432/c5w894 2013

  11. [11]

    Zhiyong Cui, Kristian Henrickson, Ruimin Ke, and Yinhai Wang. Traffic graph convolutional recurrent neural network: A deep learning framework for network-scale traffic learning and forecasting.IEEE Transactions on Intelligent Transportation Systems, 2019

    2019

  12. [12]

    Mimic-iv, a freely accessible electronic health record dataset.Scientific data, 10(1):1, 2023

    Alistair EW Johnson, Lucas Bulgarelli, Lu Shen, Alvin Gayles, Ayad Shammout, Steven Horng, Tom J Pollard, Sicheng Hao, Benjamin Moody, Brian Gow, et al. Mimic-iv, a freely accessible electronic health record dataset.Scientific data, 10(1):1, 2023

    2023

  13. [13]

    The eicu collaborative research database, a freely available multi-center database for critical care research.Scientific data, 5(1):1–13, 2018

    Tom J Pollard, Alistair EW Johnson, Jesse D Raffa, Leo A Celi, Roger G Mark, and Omar Badawi. The eicu collaborative research database, a freely available multi-center database for critical care research.Scientific data, 5(1):1–13, 2018

    2018

  14. [14]

    Cola-gnn: Cross-location attention based graph neural networks for long-term ili prediction

    Songgaojun Deng, Shusen Wang, Huzefa Rangwala, Lijing Wang, and Yue Ning. Cola-gnn: Cross-location attention based graph neural networks for long-term ili prediction. InProceedings of the 29th ACM international conference on information & knowledge management, pages 245–254, 2020

    2020

  15. [15]

    Inductive confidence machines for regression

    Harris Papadopoulos, Kostas Proedrou, V olodya V ovk, and Alex Gammerman. Inductive confidence machines for regression. InMachine learning: ECML 2002: 13th European conference on machine learning Helsinki, Finland, August 19–23, 2002 proceedings 13, pages 345–356. Springer, 2002

    2002

  16. [16]

    Conformal risk control.arXiv preprint arXiv:2208.02814,

    Anastasios N Angelopoulos, Stephen Bates, Adam Fisch, Lihua Lei, and Tal Schuster. Confor- mal risk control.arXiv preprint arXiv:2208.02814, 2022

    work page arXiv 2022

  17. [17]

    Robust validation: Confident predictions even when distributions shift.Journal of the American Statistical Association, pages 1–66, 2024

    Maxime Cauchois, Suyash Gupta, Alnur Ali, and John C Duchi. Robust validation: Confident predictions even when distributions shift.Journal of the American Statistical Association, pages 1–66, 2024. 10

    2024

  18. [18]

    Coverage-guaranteed prediction sets for out-of-distribution data

    Xin Zou and Weiwei Liu. Coverage-guaranteed prediction sets for out-of-distribution data. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 17263–17270, 2024

    2024

  19. [19]

    Federated conformal predictors for distributed uncertainty quantification

    Charles Lu, Yaodong Yu, Sai Praneeth Karimireddy, Michael Jordan, and Ramesh Raskar. Federated conformal predictors for distributed uncertainty quantification. InInternational Conference on Machine Learning, pages 22942–22964. PMLR, 2023

    2023

  20. [20]

    Distributed conformal prediction via message passing, 2025

    Haifeng Wen, Hong Xing, and Osvaldo Simeone. Distributed conformal prediction via message passing, 2025

    2025

  21. [21]

    Statistical aspects of wasserstein distances.Annual review of statistics and its application, 6(1):405–431, 2019

    Victor M Panaretos and Yoav Zemel. Statistical aspects of wasserstein distances.Annual review of statistics and its application, 6(1):405–431, 2019

    2019

  22. [22]

    Fundamentals of stein’s method

    Nathan Ross. Fundamentals of stein’s method. 2011

    2011

  23. [23]

    Normalizing flows: An introduction and review of current methods.IEEE transactions on pattern analysis and machine intelligence, 43(11):3964–3979, 2020

    Ivan Kobyzev, Simon JD Prince, and Marcus A Brubaker. Normalizing flows: An introduction and review of current methods.IEEE transactions on pattern analysis and machine intelligence, 43(11):3964–3979, 2020

    2020

  24. [24]

    Normalizing flows for probabilistic modeling and inference.Journal of Machine Learning Research, 22(57):1–64, 2021

    George Papamakarios, Eric Nalisnick, Danilo Jimenez Rezende, Shakir Mohamed, and Balaji Lakshminarayanan. Normalizing flows for probabilistic modeling and inference.Journal of Machine Learning Research, 22(57):1–64, 2021

    2021

  25. [25]

    Augmented normalizing flows: Bridging the gap between generative flows and latent variable models, 2020

    Chin-Wei Huang, Laurent Dinh, and Aaron Courville. Augmented normalizing flows: Bridging the gap between generative flows and latent variable models, 2020

    2020

  26. [26]

    Density estimation using Real NVP

    Laurent Dinh, Jascha Sohl-Dickstein, and Samy Bengio. Density estimation using real nvp. arXiv preprint arXiv:1605.08803, 2016

    work page internal anchor Pith review arXiv 2016

  27. [27]

    Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization

    Shiori Sagawa, Pang Wei Koh, Tatsunori B Hashimoto, and Percy Liang. Distributionally robust neural networks for group shifts: On the importance of regularization for worst-case generalization.arXiv preprint arXiv:1911.08731, 2019

    work page internal anchor Pith review arXiv 1911

  28. [28]

    The limits of distribution-free conditional predictive inference.Information and Inference: A Journal of the IMA, 10(2):455–482, 2021

    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(2):455–482, 2021

    2021

  29. [29]

    Conformalized quantile regression

    Yaniv Romano, Evan Patterson, and Emmanuel Candes. Conformalized quantile regression. Advances in neural information processing systems, 32, 2019

    2019

  30. [30]

    normflows: A pytorch package for normalizing flows.Journal of Open Source Software, 8(86):5361, 2023

    Vincent Stimper, David Liu, Andrew Campbell, Vincent Berenz, Lukas Ryll, Bernhard Schölkopf, and José Miguel Hernández-Lobato. normflows: A pytorch package for normalizing flows.Journal of Open Source Software, 8(86):5361, 2023

    2023

  31. [31]

    Sinkhorn distances: Lightspeed computation of optimal transport.Advances in neural information processing systems, 26, 2013

    Marco Cuturi. Sinkhorn distances: Lightspeed computation of optimal transport.Advances in neural information processing systems, 26, 2013

    2013

  32. [32]

    The sinkhorn–knopp algorithm: convergence and applications.SIAM Journal on Matrix Analysis and Applications, 30(1):261–275, 2008

    Philip A Knight. The sinkhorn–knopp algorithm: convergence and applications.SIAM Journal on Matrix Analysis and Applications, 30(1):261–275, 2008

    2008

  33. [33]

    Interpolating between optimal transport and mmd using sinkhorn divergences

    Jean Feydy, Thibault Séjourné, François-Xavier Vialard, Shun-ichi Amari, Alain Trouve, and Gabriel Peyré. Interpolating between optimal transport and mmd using sinkhorn divergences. InThe 22nd International Conference on Artificial Intelligence and Statistics, pages 2681–2690, 2019

    2019

  34. [34]

    Conformal prediction under covariate shift.Advances in neural information processing systems, 32, 2019

    Ryan J Tibshirani, Rina Foygel Barber, Emmanuel Candes, and Aaditya Ramdas. Conformal prediction under covariate shift.Advances in neural information processing systems, 32, 2019

    2019

  35. [35]

    Adversarially robust conformal prediction

    Asaf Gendler, Tsui-Wei Weng, Luca Daniel, and Yaniv Romano. Adversarially robust conformal prediction. InInternational Conference on Learning Representations, 2021

    2021

  36. [36]

    Adaptive graph convolutional recurrent network for traffic forecasting.Advances in neural information processing systems, 33:17804–17815, 2020

    Lei Bai, Lina Yao, Can Li, Xianzhi Wang, and Can Wang. Adaptive graph convolutional recurrent network for traffic forecasting.Advances in neural information processing systems, 33:17804–17815, 2020. 11

    2020

  37. [37]

    Knowing what you know: valid and validated confidence sets in multiclass and multilabel prediction.Journal of machine learning research, 22(81):1–42, 2021

    Maxime Cauchois, Suyash Gupta, and John C Duchi. Knowing what you know: valid and validated confidence sets in multiclass and multilabel prediction.Journal of machine learning research, 22(81):1–42, 2021

    2021

  38. [38]

    Principal component analysis.Wiley interdisciplinary reviews: computational statistics, 2(4):433–459, 2010

    Hervé Abdi and Lynne J Williams. Principal component analysis.Wiley interdisciplinary reviews: computational statistics, 2(4):433–459, 2010

    2010

  39. [39]

    Visualizing data using t-sne.Journal of machine learning research, 9(11), 2008

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

    2008

  40. [40]

    UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction

    Leland McInnes, John Healy, and James Melville. Umap: Uniform manifold approximation and projection for dimension reduction.arXiv preprint arXiv:1802.03426, 2018

    work page internal anchor Pith review arXiv 2018

  41. [41]

    Adaptive conformal classification with noisy labels

    Matteo Sesia, YX Wang, and Xin Tong. Adaptive conformal classification with noisy labels. arXiv preprint arXiv:2309.05092, 2023

    work page arXiv 2023

  42. [42]

    Conformal prediction is robust to label noise.arXiv preprint arXiv:2209.14295, 2, 2022

    Bat-Sheva Einbinder, Stephen Bates, Anastasios N Angelopoulos, Asaf Gendler, and Yaniv Romano. Conformal prediction is robust to label noise.arXiv preprint arXiv:2209.14295, 2, 2022

    work page arXiv 2022

  43. [43]

    The speed of mean glivenko-cantelli convergence.The Annals of Mathematical Statistics, 40(1):40–50, 1969

    Richard Mansfield Dudley. The speed of mean glivenko-cantelli convergence.The Annals of Mathematical Statistics, 40(1):40–50, 1969

    1969

  44. [44]

    Sharp asymptotic and finite-sample rates of convergence of empirical measures in wasserstein distance

    Jonathan Weed and Francis Bach. Sharp asymptotic and finite-sample rates of convergence of empirical measures in wasserstein distance. 2019

    2019

  45. [45]

    Springer Science & Business Media, 2007

    Irene Fonseca and Giovanni Leoni.Modern methods in the calculus of variations: Lˆ p spaces. Springer Science & Business Media, 2007

    2007

  46. [46]

    Variational inference with normalizing flows, 2016

    Danilo Jimenez Rezende and Shakir Mohamed. Variational inference with normalizing flows, 2016

    2016

  47. [47]

    Deep residual learning for image recognition, 2015

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition, 2015

    2015

  48. [48]

    Improved variational inference with inverse autoregressive flow.Advances in neural information processing systems, 29, 2016

    Durk P Kingma, Tim Salimans, Rafal Jozefowicz, Xi Chen, Ilya Sutskever, and Max Welling. Improved variational inference with inverse autoregressive flow.Advances in neural information processing systems, 29, 2016

    2016

  49. [49]

    Mondrian conformal predictive distri- butions

    Henrik Boström, Ulf Johansson, and Tuwe Löfström. Mondrian conformal predictive distri- butions. InConformal and Probabilistic Prediction and Applications, pages 24–38. PMLR, 2021

    2021

  50. [50]

    Locally valid and discriminative prediction intervals for deep learning models.Advances in Neural Information Processing Systems, 34:8378–8391, 2021

    Zhen Lin, Shubhendu Trivedi, and Jimeng Sun. Locally valid and discriminative prediction intervals for deep learning models.Advances in Neural Information Processing Systems, 34:8378–8391, 2021

    2021

  51. [51]

    Localized conformal prediction: A generalized inference framework for confor- mal prediction.Biometrika, 110(1):33–50, 2023

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

    2023

  52. [52]

    arXiv preprint arXiv:2305.12616 , year=

    Isaac Gibbs, John J Cherian, and Emmanuel J Candès. Conformal prediction with conditional guarantees.arXiv preprint arXiv:2305.12616, 2023

    work page arXiv 2023

  53. [53]

    An information theoretic perspective on conformal prediction.Advances in Neural Information Processing Systems, 37:101000–101041, 2024

    Alvaro Correia, Fabio Valerio Massoli, Christos Louizos, and Arash Behboodi. An information theoretic perspective on conformal prediction.Advances in Neural Information Processing Systems, 37:101000–101041, 2024

    2024

  54. [54]

    Learning optimal conformal classifiers, 2022

    David Stutz, Ali Taylan Cemgil, Arnaud Doucet, et al. Learning optimal conformal classifiers. arXiv preprint arXiv:2110.09192, 2021

    work page arXiv 2021

  55. [55]

    On volume minimization in conformal regression, 2025

    Batiste Le Bars and Pierre Humbert. On volume minimization in conformal regression, 2025

    2025

  56. [56]

    Improving conditional coverage via orthog- onal quantile regression.Advances in neural information processing systems, 34:2060–2071, 2021

    Shai Feldman, Stephen Bates, and Yaniv Romano. Improving conditional coverage via orthog- onal quantile regression.Advances in neural information processing systems, 34:2060–2071, 2021. 12

    2060

  57. [57]

    Normalizing flows for conformal regression.arXiv preprint arXiv:2406.03346, 2024

    Nicolo Colombo. Normalizing flows for conformal regression.arXiv preprint arXiv:2406.03346, 2024

    work page arXiv 2024

  58. [58]

    Contra: Conformal prediction region via normalizing flow transformation

    Zhenhan Fang, Aixin Tan, and Jian Huang. Contra: Conformal prediction region via normalizing flow transformation. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  59. [59]

    Multivariate conformal prediction using optimal transport, 2025

    Michal Klein, Louis Bethune, Eugene Ndiaye, and Marco Cuturi. Multivariate conformal prediction using optimal transport, 2025

    2025

  60. [60]

    Optimal transport-based conformal predic- tion, 2025

    Gauthier Thurin, Kimia Nadjahi, and Claire Boyer. Optimal transport-based conformal predic- tion, 2025

    2025

  61. [61]

    Estimating conditional quantiles with the help of the pinball loss

    Ingo Steinwart and Andreas Christmann. Estimating conditional quantiles with the help of the pinball loss. 2011

    2011

  62. [62]

    Johnson, and Hannes Leeb

    Danijel Kivaranovic, Kory D. Johnson, and Hannes Leeb. Adaptive, distribution-free prediction intervals for deep networks, 2020

    2020

  63. [63]

    Matteo Sesia and Emmanuel J. Candès. A comparison of some conformal quantile regression methods.Stat, 9(1), January 2020

    2020

  64. [64]

    PhD thesis, Université Paris-Saclay, 2020

    Jean Feydy.Analyse de données géométriques, au delà des convolutions. PhD thesis, Université Paris-Saclay, 2020

    2020

  65. [65]

    Domain adaptation for time series forecasting via attention sharing

    Xiaoyong Jin, Youngsuk Park, Danielle Maddix, Hao Wang, and Yuyang Wang. Domain adaptation for time series forecasting via attention sharing. InInternational Conference on Machine Learning, pages 10280–10297. PMLR, 2022

    2022

  66. [66]

    Reaction-diffusion graph ordinary differential equation networks: Traffic-law-informed speed prediction under mismatched data

    Yue Sun, Chao Chen, Yuesheng Xu, Sihong Xie, Rick S Blum, and Parv Venkitasubramaniam. Reaction-diffusion graph ordinary differential equation networks: Traffic-law-informed speed prediction under mismatched data. The 12th International Workshop on Urban Computing, held in conjunction with . . . , 2023

    2023

  67. [67]

    Multicenter studies: relevance, design and implementation.Indian pediatrics, 59(7):571–579, 2022

    Manoja Kumar Das. Multicenter studies: relevance, design and implementation.Indian pediatrics, 59(7):571–579, 2022

    2022

  68. [68]

    Estimating diagnostic uncertainty in artificial intelligence assisted pathology using conformal prediction.Nature communications, 13(1):7761, 2022

    Henrik Olsson, Kimmo Kartasalo, Nita Mulliqi, Marco Capuccini, Pekka Ruusuvuori, Hema- mali Samaratunga, Brett Delahunt, Cecilia Lindskog, Emiel AM Janssen, Anders Blilie, et al. Estimating diagnostic uncertainty in artificial intelligence assisted pathology using conformal prediction.Nature communications, 13(1):7761, 2022

    2022

  69. [69]

    Pandemic Intervals Framework (PIF) — cdc.gov

    CDC. Pandemic Intervals Framework (PIF) — cdc.gov. https://www.cdc.gov/ pandemic-flu/php/national-strategy/intervals-framework.html?CDC_AAref_ Val=https://www.cdc.gov/flu/pandemic-resources/national-strategy/ intervals-framework.html. [Accessed 28-04-2025]

    2025

  70. [70]

    Tiberiu Harko, Francisco SN Lobo, and Man Kwong Mak. Exact analytical solutions of the susceptible-infected-recovered (sir) epidemic model and of the sir model with equal death and birth rates.Applied Mathematics and Computation, 236:184–194, 2014

    2014

  71. [71]

    Analysis of sir epidemic model with information spreading of awareness.Chaos, Solitons & Fractals, 119:118–125, 2019

    KM Ariful Kabir, Kazuki Kuga, and Jun Tanimoto. Analysis of sir epidemic model with information spreading of awareness.Chaos, Solitons & Fractals, 119:118–125, 2019

    2019

  72. [72]

    A restricted epidemic sir model with elementary solutions.Physica A: Statistical Mechanics and its Applications, 600:127570, 2022

    Mustafa Turkyilmazoglu. A restricted epidemic sir model with elementary solutions.Physica A: Statistical Mechanics and its Applications, 600:127570, 2022

    2022

  73. [73]

    Classification with valid and adaptive coverage.Advances in Neural Information Processing Systems, 33:3581–3591, 2020

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

    2020

  74. [74]

    The expressive power of a class of normalizing flow models, 2020

    Zhifeng Kong and Kamalika Chaudhuri. The expressive power of a class of normalizing flow models, 2020. 13

    2020

  75. [75]

    Why and when can deep-but not shallow-networks avoid the curse of dimensionality: a review

    Tomaso Poggio, Hrushikesh Mhaskar, Lorenzo Rosasco, Brando Miranda, and Qianli Liao. Why and when can deep-but not shallow-networks avoid the curse of dimensionality: a review. International Journal of Automation and Computing, 14(5):503–519, 2017

    2017

  76. [76]

    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(1):71–96, 2014

    2014

  77. [77]

    Uncertainty sets for image classifiers using conformal prediction.arXiv:2009.14193,

    Anastasios Angelopoulos, Stephen Bates, Jitendra Malik, and Michael I Jordan. Uncertainty sets for image classifiers using conformal prediction.arXiv preprint arXiv:2009.14193, 2020

    work page arXiv 2009

  78. [78]

    On temperature scaling and conformal prediction of deep classifiers, 2025

    Lahav Dabah and Tom Tirer. On temperature scaling and conformal prediction of deep classifiers, 2025

    2025

  79. [79]

    Does confidence calibration improve conformal prediction?, 2024

    Huajun Xi, Jianguo Huang, Kangdao Liu, Lei Feng, and Hongxin Wei. Does confidence calibration improve conformal prediction?, 2024

    2024

  80. [80]

    Selection and aggregation of conformal prediction sets, 2024

    Yachong Yang and Arun Kumar Kuchibhotla. Selection and aggregation of conformal prediction sets, 2024

    2024

Showing first 80 references.