Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Multi-Scale Conformal Prediction: A Theoretical Framework with Coverage Guarantees

T0 review · 4 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read The paper establishes that intersecting conformal prediction sets built at different scales preserves the marginal coverage guarantee and yields sets that are never larger than any single-scale set.

desk verdict Theorem 1 is a correct Bonferroni union-bound observation; Theorems 3-5 are unsound, and Theorem 5's optimality claim is mathematically false under its own assumptions. read the letter →

arxiv 2502.05565 v1 pith:SSA6OGAN submitted 2025-02-08 math.ST cs.SYeess.SYstat.TH

Ali Baheri , Marzieh Amiri Shahbazi This is my paper · ORCID
classification math.STcs.SYeess.SYstat.TH MSC 62G1562G20
keywords conformalpredictionmulti-scaleuncertaintyquantificationcoverageguaranteesexchangeabilitysetsmiscoverageallocationhierarchicalmodeling
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 a multi-scale version of conformal prediction: instead of one conformity score, define a distinct score for each scale or resolution of the data, build one conformal prediction set per scale at a miscoverage level $\alpha_k$, and intersect them. It tries to establish that this intersection keeps the standard marginal coverage guarantee, $\mathbb{P}(Y_{n+1} \in C_n(X_{n+1})) \ge 1-\alpha$, whenever $\sum_k \alpha_k = \alpha$, and that the intersection is always a subset of every scale-specific set, so it can only be smaller. The paper further argues that allocating smaller $\alpha_k$ to more informative scales reduces expected set size, and that under consistency conditions the multi-scale set asymptotically converges to the minimal conditional prediction set. A sympathetic reader cares because valid and small prediction sets are the practical target of uncertainty quantification, and this is a simple recipe for combining multiple views of the data.

What carries the argument

The central object is the multi-scale prediction set $C_n(X_{n+1})=\bigcap_{k=1}^K C_n^{(k)}(X_{n+1})$, where each $C_n^{(k)}$ is an ordinary conformal set built from a scale-specific conformity function $A^{(k)}$ at miscoverage level $\alpha_k$ with $\sum_k\alpha_k=\alpha$. The coverage argument is carried by the union bound: the event of missing the true label is the union of per-scale miscoverage events, each bounded by $\alpha_k$. The efficiency argument is carried by subset inclusion: the intersection is contained in every scale-specific set. The asymptotic-optimality argument is carried by the assumption that all $A^{(k)}$ converge uniformly to the same Bayes score, so that the limiting intersection reduces to a level set $\{y: p(y) > \max_k \alpha_k\}$ of the limiting conformal p-value.

What would settle it

Simulate a finite-label classification problem with known posterior probabilities, two scales with $\alpha_1=\alpha_2=\alpha/2$, and conformity scores that already equal the Bayes posterior at every sample; as $n\to\infty$ the multi-scale intersection becomes $\{y: p(y) > \alpha/2\}$, whose conditional coverage is $1-\alpha/2$, not $1-\alpha$, so it cannot equal the minimal set $C^*$ with coverage $1-\alpha$. This calculation would settle that Theorem 5's conclusion does not follow from its stated assumptions.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the multi-scale set $C_n(X_{n+1})=\bigcap_{k=1}^K C_n^{(k)}(X_{n+1})$, with $C_n^{(k)}$ the standard conformal set at level $\alpha_k$ and $\sum_k \alpha_k = \alpha$, inherits the conformal guarantee: $\mathbb{P}(Y_{n+1}\in C_n(X_{n+1}))\ge 1-\alpha$ by the union bound over per-scale miscoverage events. Because the intersection is contained in each $C_n^{(k)}$, its size is never larger than any single-scale set, so efficiency is automatic. The paper's strongest additional claim is asymptotic: if every scale's conformity score converges almost surely and uniformly to the Bayes score $-P(Y\mid X)$, then the intersection converges almost surely to the minimal set $C^*(X_{n+1})$ with conditional coverage $1-\alpha$. The allocation principle (Theorem 3) asserts that the expected size is minimized when the negative elasticities $f_k'(\alpha_k)/f_k(\alpha_k)$ are equalized across scales, which assigns smaller $\alpha_k$ to more informative scales.

Load-bearing premise

The paper's strongest claim, asymptotic convergence to the minimal conditional set, rests on the premise that every scale's conformity scores converge to the same Bayes posterior probabilities; if that premise fails, the claimed minimality does not follow.

Editorial extensions

If this is right

  • The multi-scale predictor can be dropped into any existing conformal pipeline: split the budget $\alpha$ among scales, build per-scale sets on the calibration data, and intersect; the finite-sample marginal guarantee holds for free.
  • Because the intersection is a subset of each single-scale set, on every dataset the multi-scale set is no larger than the smallest of the individual sets, so it is never worse in terms of set size.
  • If scale-specific expected sizes are decreasing convex functions of $\alpha_k$ and scale-wise conformity scores are independent, equalizing the negative elasticities $f_k'(\alpha_k)/f_k(\alpha_k)$ minimizes the expected intersection size; this gives a concrete miscoverage-allocation rule.
  • When scale-wise miscoverage events are positively dependent, the union bound is loose and the actual coverage is strictly above $1-\alpha$, so the multi-scale method is conservative rather than over-eager.
  • Under the consistency conditions of Theorem 5, the asymptotic multi-scale set reaches the minimal set with conditional coverage $1-\alpha$, meaning that no further set-size reduction is possible at that coverage level.

Reading between the lines

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

  • A practical extension suggested by Theorem 3 is to estimate each scale's expected-size curve $f_k$ on calibration data and solve the equal-elasticity condition numerically, giving a data-driven $\alpha_k$ allocation rather than a uniform split.
  • The conservatism under positive dependence could be exploited by calibrating the product bound $1-\prod_k(1-\alpha_k)$ to equal the target $\alpha$, which would allow smaller per-scale miscoverage levels and hence tighter intersections.
  • The same intersection construction transfers directly to regression by intersecting conformal intervals or bands built at different resolutions; the finite-sample coverage proof is agnostic to the output space.
  • It would be useful to test empirically how much of the set-size reduction survives when conformity scores at different scales are strongly correlated, since the theoretical efficiency bound comes from independence or curvature assumptions.
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

4 major / 5 minor

Summary. The paper proposes a multi-scale conformal prediction framework in which K conformity scores, each associated with a different scale or resolution, produce K conformal prediction sets whose intersection forms the final prediction set. The total miscoverage level α is split as ∑ α_k = α. The main theoretical claims are: (i) Theorem 1, the intersection preserves marginal coverage at level 1−α via the union bound; (ii) Theorem 2, the intersection is no larger than any single-scale set; (iii) Theorem 3, an optimal allocation of α_k assigns smaller miscoverage to more informative scales; (iv) Theorem 4, positive dependence between scales yields conservative coverage; and (v) Theorem 5, under convergence of each scale's conformity score to the Bayes-optimal score, the multi-scale set converges to the minimal conditional-coverage set C*. A synthetic classification experiment is reported to illustrate coverage and efficiency.

Significance. The basic idea of intersecting conformal prediction sets from multiple scales is natural, and the marginal coverage guarantee in Theorem 1 is a correct and immediate consequence of the union bound; this part is sound and could be useful as a building block. However, the paper's more substantive claims—the optimal allocation theorem, the conservative-coverage result under positive dependence, and especially the asymptotic optimality theorem—are not established. Theorem 5 is, in fact, false under the paper's own assumptions, because the limiting p-value is monotonically increasing in the posterior probability, not decreasing as the proof asserts. The manuscript therefore does not deliver a reliable theoretical framework beyond the elementary coverage bound. The numerical experiments, while suggestive, are synthetic and do not compensate for the unsupported theoretical results.

major comments (4)
  1. [Section 4, Theorem 5, Step 7] The monotonicity claim used in Step 7 is backwards. From the paper's own limiting expression, p(y) = P(P(Y|X) ≤ P(y|X_{n+1})) is an increasing function of P(y|X_{n+1}), since any cumulative distribution function is increasing. Larger posterior mass therefore gives a larger p-value, not a smaller one. Consequently the asymptotic intersection set is {y : p(y) > max_k α_k}, which selects labels whose posterior lies below a fixed quantile, not the labels with the highest posterior mass. The claimed convergence to the minimal conditional-coverage set C* fails. Moreover, when max_k α_k < α (which is the typical case for K ≥ 2 with all α_k < α), the limiting intersection set has conditional coverage 1 − max_k α_k, not 1 − α, so the multi-scale set is asymptotically conservative rather than minimal.
  2. [Section 4, Theorem 5, Conditions 1–3] Even if the monotonicity error were corrected, the theorem's conclusion is already contained in its assumptions. Conditions 1–3 require every scale's conformity score to converge to the same Bayes-optimal score s(X,Y) = −P(Y|X). Under those conditions, a single-scale conformal predictor at level α already achieves the minimal set C* in the limit, and intersecting K copies of the same limiting set cannot improve on it. The intersection plays no role in the claimed optimality, so Theorem 5 does not establish a benefit of multi-scale aggregation; at best it shows that an already-optimal single-scale predictor retains its optimality when intersected with itself.
  3. [Section 4, Theorem 3, Step 2] The approximation P_k(y) ≈ E[|C_n^{(k)}|]/|Y| is unjustified and generally false. It assumes that every label has roughly the same probability of being included in the scale-k prediction set, which is not implied by any condition in the theorem and contradicts the very idea that more informative scales have concentrated prediction sets. The subsequent Lagrange multiplier argument optimizes the approximate objective ∏ f_k(α_k), but the first-order condition ψ_k(α_k) = ψ_j(α_j) only characterizes a stationary point of that approximate problem. The conclusion that 'more informative scales get smaller α_k' does not follow from this condition without additional assumptions on the shape of f_k; for a decreasing convex f_k, the sign of the cross-effect between informativeness and optimal α_k is not determined by the stated conditions.
  4. [Section 4, Theorem 4] The proof of conservative coverage relies on transferring a positive-dependence assumption on the conformity scores to a positive-dependence assumption on the miscoverage events E_k = {Y_{n+1} ∉ C_n^{(k)}(X_{n+1})}. The mapping from score dependence to event dependence is nontrivial and is not established. Even if positive dependence of the events is granted, the inequality P(∪ E_k) ≤ 1 − ∏ (1 − P(E_k)) is valid for positively associated events, but the paper does not define the notion of positive dependence used, nor does it show that conformal p-values inherit this property. The theorem's conclusion that some α_dep < α exists is weaker than claimed and is not an effective bound; in particular, the proof does not identify α_dep or relate it to the α_k.
minor comments (5)
  1. [Section 5, Numerical Results] The experiment description states that two features (coarse and fine) are used and two logistic regression models are trained, yet the results discuss 'Scale 3' and compare three single-scale methods. This inconsistency makes it impossible to map the experimental setup to the reported figures and tables.
  2. [Section 3, Methodology] The label space is denoted by 'Y' without a formal definition; it is later used as the set of all possible labels. The notation should distinguish the random variable Y from the label space, e.g., using script Y.
  3. [Section 4, Theorem 4] The statement introduces α_dep as a constant satisfying P(Y_{n+1} ∉ C_n(X_{n+1})) ≤ α_dep < α, but the proof never defines α_dep; the final display simply restates the inequality. A concrete bound, such as 1 − ∏(1 − α_k), would be more informative.
  4. [Throughout] There are numerous typographical errors and missing words, including 'otentially' and 'ationale' in Section 3, and the proof of Theorem 5 jumps from Step 4 to Step 6 without a Step 5, which disrupts readability.
  5. [References] Several references in the related-work discussion are cited in a generic way (e.g., [16], [21], [27]) without a precise connection to multi-scale conformal prediction; additional context for how each cited work relates to the proposed framework would help the reader assess novelty.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity. The coverage theorem is a standard union-bound proof; the allocation theorem follows from its stated convexity assumptions. Theorem 5's optimality claim is invalid (Step 7 inverts CDF monotonicity), but that is a correctness flaw, not a reduction of the conclusion to its inputs.

full rationale

Theorem 1 proves coverage by combining the standard conformal validity lemma (scale-specific p-values are stochastically larger than uniform) with the union bound; it does not assume the target coverage. Theorem 2 is definitional (an intersection is a subset). Theorem 3 minimizes an explicit expected-size model under stated convexity and independence assumptions; the conclusion that more informative scales receive smaller alpha_k is the equal-elasticity first-order condition, not a fitted parameter renamed as a prediction. Theorem 4 is another union-bound-style claim that is questionable but not circular. Theorem 5 is not a circularity but a false derivation: under its own Condition 1 the limiting p-value is p(y) = P(P(Y|X) <= P(y|X_{n+1})), a CDF that is increasing in the candidate posterior, whereas Step 7 asserts the opposite ('larger values of P(y|X_{n+1}) correspond to smaller p(y)'). Consequently the asymptotic set is {y : p(y) > max_k alpha_k}, whose coverage is 1 - max_k alpha_k, not the nominal 1 - alpha when K >= 2. This invalidates the claimed convergence to C*, but the conclusion is not equivalent to the assumptions by construction; the proof's monotonicity assertion is simply wrong. No load-bearing self-citation chain appears: the authors' prior multi-fidelity work is cited for context only. Therefore the circularity score is 0.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The theoretical proofs contain no fitted parameters. The simulation uses user-selected settings such as alpha=0.1, alpha_k=alpha/2, and noise levels, which are not fitted. The additional assumptions that do real work in Theorems 3, 4, and 5 are listed as axioms. No new physical or model entities are introduced; the scales are a modeling choice equivalent to choosing K conformity functions.

assumptions (6)
  • domain assumption The extended sequence {(X_i,Y_i)}_{i=1}^{n+1} is exchangeable.
    Invoked in Theorem 1 and throughout Section 4; standard in conformal prediction but unverifiable from the paper's experiments.
  • ad hoc to paper For each scale k, f_k(alpha_k) = E[|C_n^{(k)}|] is a decreasing convex function of alpha_k.
    Assumption 1 of Theorem 3; no proof is given that conformal set sizes have this shape, and the elasticity argument does not follow from convexity alone.
  • ad hoc to paper Conformity scores at different scales are independent in Theorem 3 and positively dependent in Theorem 4.
    The two theorems use contradictory dependence assumptions, and neither is linked to the data-generating process presented in Section 5.
  • ad hoc to paper Positive dependence of conformity scores transfers to positive dependence of the miscoverage indicator events E_k.
    Theorem 4 depends on this transfer to bound the union, but the transfer is never proved.
  • ad hoc to paper For each scale k, A^(k)(X_i,Y_i) converges almost surely to s(X_i,Y_i) = -P(Y_i|X_i), uniformly over Y, with the same limit for all k.
    Theorem 5, Conditions 1-3; this already assumes each scale is Bayes-optimal and contains the claimed optimality conclusion.
  • ad hoc to paper The per-label inclusion probability satisfies P_k(y) approximately E[|C_n^{(k)}|]/|Y| for every label y.
    Theorem 3, Step 2; the expected set size divided by label count is not generally the per-label inclusion probability.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multi-Scale Conformal Prediction: A Theoretical Framework with Coverage Guarantees." pith.science (2026). https://pith.science/paper/SSA6OGAN

@misc{pith2026250205565,
  author       = {Pith},
  title        = {Pith review of: Multi-Scale Conformal Prediction: A Theoretical Framework with Coverage Guarantees},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SSA6OGAN}},
  note         = {Machine review of arXiv:2502.05565}
}
read the original abstract

We propose a multi-scale extension of conformal prediction, an approach that constructs prediction sets with finite-sample coverage guarantees under minimal statistical assumptions. Classic conformal prediction relies on a single notion of conformity, overlooking the multi-level structures that arise in applications such as image analysis, hierarchical data exploration, and multi-resolution time series modeling. In contrast, the proposed framework defines a distinct conformity function at each relevant scale or resolution, producing multiple conformal predictors whose prediction sets are then intersected to form the final multi-scale output. We establish theoretical results confirming that the multi-scale prediction set retains the marginal coverage guarantees of the original conformal framework and can, in fact, yield smaller or more precise sets in practice. By distributing the total miscoverage probability across scales in proportion to their informative power, the method further refines the set sizes. We also show that dependence between scales can lead to conservative coverage, ensuring that the actual coverage exceeds the nominal level. Numerical experiments in a synthetic classification setting demonstrate that multi-scale conformal prediction achieves or surpasses the nominal coverage level while generating smaller prediction sets compared to single-scale conformal methods.

Figures

Figures reproduced from arXiv: 2502.05565 by the authors.

Figure 1
Figure 1. Comparison of multi-scale conformal prediction (MSCP) with single-scale methods achieves a smaller distribution of average set sizes compared to any single-scale method, indicating more precise predictions at a fixed miscoverage level. Scales 1 and 2 exhibit moderate variability, with Scale 1 occasionally producing relatively large sets under some seeds and comparatively small sets under others, reflecting its inter… view at source ↗

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. Split Conformal Prediction in the Function Space with Neural Operators

    cs.LG 2025-09 conditional novelty 5.0 of 10

    A split conformal prediction method for function-valued outputs is proposed for neural operators, using a weighted L2 norm and a resolution-transfer heuristic to maintain calibrated coverage.

Reference graph

Works this paper leans on

30 extracted references · 19 canonical work pages · cited by 1 Pith paper

  1. [1]

    Integrating machine learning and multiscale modeling—perspectives, challenges, and opportunities in the biolog- ical, biomedical, and behavioral sciences

    Mark Alber, Adrian Buganza Tepole, William R Cannon, Suvranu De, Salvador Dura-Bernal, Krishna Garikipati, George Karniadakis, William W Lytton, Paris Perdikaris, Linda Petzold, et al. Integrating machine learning and multiscale modeling—perspectives, challenges, and opportunities in the biolog- ical, biomedical, and behavioral sciences. NPJ digital medic...

    work page 2019

  2. [2]

    Conformal prediction: A gentle introduction

    Anastasios N Angelopoulos, Stephen Bates, et al. Conformal prediction: A gentle introduction. Foun- dations and Trends® in Machine Learning, 16(4):494–591, 2023

    work page 2023

  3. [3]

    Exploring the role of simulator fidelity in the safety validation of learning-enabled au- tonomous systems

    Ali Baheri. Exploring the role of simulator fidelity in the safety validation of learning-enabled au- tonomous systems. AI Magazine, 44(4):453–459, 2023

    work page 2023

  4. [4]

    Safety validation of learning-based autonomous systems: A multi-fidelity approach

    Ali Baheri. Safety validation of learning-based autonomous systems: A multi-fidelity approach. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 15432–15432, 2023

    work page 2023

  5. [5]

    SMTL: A stratified logic for expressive multi-level temporal specifications

    Ali Baheri and Peng Wei. SMTL: A stratified logic for expressive multi-level temporal specifications. arXiv preprint arXiv:2501.02094, 2025

    arXiv 2025

  6. [6]

    Black-box safety validation of autonomous systems: A multi-fidelity reinforcement learning approach

    Jared J Beard and Ali Baheri. Black-box safety validation of autonomous systems: A multi-fidelity reinforcement learning approach. arXiv preprint arXiv:2203.03451, 2022

    arXiv 2022

  7. [7]

    Improved online conformal prediction via strongly adaptive online learning

    Aadyot Bhatnagar, Huan Wang, Caiming Xiong, and Yu Bai. Improved online conformal prediction via strongly adaptive online learning. In International Conference on Machine Learning, pages 2337–2363. PMLR, 2023

    2023

  8. [8]

    Multi-fidelity design framework inte- grating compositional kernels to facilitate early-stage design exploration of complex systems

    Nikoleta Dimitra Charisi, Hans Hopman, and Austin A Kana. Multi-fidelity design framework inte- grating compositional kernels to facilitate early-stage design exploration of complex systems. Journal of Mechanical Design, 147(1), 2025

    work page 2025

Show all 30 references
  1. [9]

    Distributional conformal prediction

    Victor Chernozhukov, Kaspar W ¨uthrich, and Yinchu Zhu. Distributional conformal prediction. Pro- ceedings of the National Academy of Sciences, 118(48):e2107794118, 2021

    2021

  2. [10]

    On the expected size of conformal pre- diction sets

    Guneet S Dhillon, George Deligiannidis, and Tom Rainforth. On the expected size of conformal pre- diction sets. In International Conference on Artificial Intelligence and Statistics , pages 1549–1557. PMLR, 2024

    2024

  3. [11]

    A review on multiscale-deep-learning applications

    Elizar Elizar, Mohd Asyraf Zulkifley, Rusdha Muharar, Mohd Hairi Mohd Zaman, and Seri Mastura Mustaza. A review on multiscale-deep-learning applications. Sensors, 22(19):7384, 2022

    2022

  4. [12]

    Review of multi-fidelity models

    M Giselle Fern ´andez-Godino. Review of multi-fidelity models. arXiv preprint arXiv:1609.07196, 2016

    2016 arXiv

  5. [13]

    Multi-fidelity optimization via sur- rogate modelling

    Alexander IJ Forrester, Andr ´as S ´obester, and Andy J Keane. Multi-fidelity optimization via sur- rogate modelling. Proceedings of the royal society a: mathematical, physical and engineering sciences , 463(2088):3251–3269, 2007

    2007

  6. [14]

    Adaptive conformal inference under distribution shift

    Isaac Gibbs and Emmanuel Candes. Adaptive conformal inference under distribution shift. Advances in Neural Information Processing Systems, 34:1660–1672, 2021

    2021

  7. [15]

    Verifiably robust conformal predic- tion

    Linus Jeary, Tom Kuipers, Mehran Hosseini, and Nicola Paoletti. Verifiably robust conformal predic- tion. arXiv preprint arXiv:2405.18942, 2024

    2024 arXiv

  8. [16]

    Quantifying deep learning model uncertainty in conformal predic- tion

    Hamed Karimi and Reza Samavi. Quantifying deep learning model uncertainty in conformal predic- tion. In Proceedings of the AAAI Symposium Series, volume 1, pages 142–148, 2023

    2023

  9. [17]

    Length optimization in conformal prediction

    Shayan Kiyani, George Pappas, and Hamed Hassani. Length optimization in conformal prediction. arXiv preprint arXiv:2406.18814, 2024

    2024 arXiv

  10. [18]

    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(523):1094–1111, 2018

    2018

  11. [19]

    Hierarchical deep learning of multiscale differential equation time-steppers

    Yuying Liu, J Nathan Kutz, and Steven L Brunton. Hierarchical deep learning of multiscale differential equation time-steppers. Philosophical Transactions of the Royal Society A, 380(2229):20210200, 2022

    2022

  12. [20]

    Multiscale modeling meets machine learning: What can we learn? Archives of Computational Methods in Engineer- ing, 28:1017–1037, 2021

    Grace CY Peng, Mark Alber, Adrian Buganza Tepole, William R Cannon, Suvranu De, Savador Dura- Bernal, Krishna Garikipati, George Karniadakis, William W Lytton, Paris Perdikaris, et al. Multiscale modeling meets machine learning: What can we learn? Archives of Computational Met...

    2021

  13. [21]

    Conformal prediction for hierarchical data

    Guillaume Principato, Yvenn Amara-Ouali, Yannig Goude, Bachir Hamrouche, Jean-Michel Poggi, and Gilles Stoltz. Conformal prediction for hierarchical data. arXiv preprint arXiv:2411.13479, 2024

    2024 arXiv

  14. [22]

    Conformalized quantile regression

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

    2019

  15. [23]

    Bayesian hierarchical multi-subject multiscale analysis of functional mri data

    Nilotpal Sanyal and Marco AR Ferreira. Bayesian hierarchical multi-subject multiscale analysis of functional mri data. NeuroImage, 63(3):1519–1531, 2012

    2012

  16. [24]

    A tutorial on conformal prediction

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

    2008

  17. [25]

    Falsification of learning-based controllers through multi-fidelity Bayesian optimization

    Zahra Shahrooei, Mykel J Kochenderfer, and Ali Baheri. Falsification of learning-based controllers through multi-fidelity Bayesian optimization. European Control Conference (ECC), 2023

    2023

  18. [26]

    Conformal time-series forecasting

    Kamile Stankeviciute, Ahmed M Alaa, and Mihaela van der Schaar. Conformal time-series forecasting. Advances in neural information processing systems, 34:6216–6228, 2021

    2021

  19. [27]

    Multi-modal conformal predic- tion regions with simple structures by optimizing convex shape templates

    R Tumu, M Cleaveland, R Mangharam, GJ Pappas, and L Lindemann. Multi-modal conformal predic- tion regions with simple structures by optimizing convex shape templates. 2024

    2024

  20. [28]

    Delving into temperature scaling for adaptive conformal prediction

    HuaJun Xi, Jianguo Huang, Kangdao Liu, Lei Feng, and Hongxin Wei. Delving into temperature scaling for adaptive conformal prediction. 2024

    2024

  21. [29]

    Provably robust conformal prediction with improved efficiency

    Ge Yan, Yaniv Romano, and Tsui-Wei Weng. Provably robust conformal prediction with improved efficiency. arXiv preprint arXiv:2404.19651, 2024

    2024 arXiv

  22. [30]

    Multi resolution analysis (MRA) for approximate self-attention

    Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, and Vikas Singh. Multi resolution analysis (MRA) for approximate self-attention. In International Conference on Machine Learning , pages 25955– 25972. PMLR, 2022

    2022

Pith tools

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