Pith. sign in

REVIEW 3 major objections 5 minor 35 references

CoRAS picks a per-image acquisition rate from the early reconstruction path and still keeps reconstruction error below a target level with high probability.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-30 18:04 UTC pith:RRC47A3J

load-bearing objection Solid conformal method for image-specific stopping times; main caveat is that error-control experiments lean on unchecked loss monotonicity and a retrospective MRI residual. the 3 major comments →

arxiv 2607.26887 v1 pith:RRC47A3J submitted 2026-07-29 stat.ML cs.LGstat.APstat.ME

Conformalized Rate-Adaptive Sensing

Jiawei Yang , Yao Zhang This is my paper
classification stat.ML cs.LGstat.APstat.ME
keywords conformal predictionadaptive sensingimage reconstructionstopping timesrate-adaptive acquisitionMRIdistribution-free coverage
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

High-resolution imaging often cannot afford to collect every measurement for every image, yet the reconstructed image still has to meet a quality bar. This paper develops CoRAS, a stopping rule that watches how a reconstruction model improves as more measurements arrive, predicts when the error will first fall below a chosen target, and then calibrates that prediction against similar past images. The result is an image-specific acquisition rate with a finite-sample guarantee that the reconstruction loss stays at or under the target with probability at least 1−α. Fixed-rate conformal rules meet the same guarantee only by oversampling easy images; CoRAS spends fewer measurements on average and gives harder images more samples. The method is aimed at sensing and compression settings where cost is paid along a path and the true image is not yet fully observed when the stop decision must be made.

Core claim

From the early part of an image’s reconstruction path, CoRAS builds a horizontal estimate of the first time the reconstruction error falls below a target level, then applies a vertical conformal correction using calibration images with similar early states, yielding a stopping time that controls reconstruction error with marginal validity and approximate conditional validity across image complexity.

What carries the argument

Horizontal–vertical conformal stopping: a monotone polynomial fit to next-band residual history predicts the unresolved error tail; a state of that prediction plus decision-time entropy matches calibration images; full conformal residual ranks turn the corrected estimate into an upper bound on the target stopping time.

Load-bearing premise

Reconstruction error is assumed to never get worse as more measurements are added, so stopping after the true first good time still keeps error under the target.

What would settle it

On a held-out exchangeable test set with a nonincreasing reconstruction loss, check whether the fraction of images whose CoRAS stop still has loss above the target exceeds α, or whether average measurements no longer beat the fixed-rate conformal baseline while coverage holds.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Imaging systems can replace a single global sampling rate with per-image rates that still meet a stated error probability.
  • Harder images automatically receive more measurements; easier images stop earlier, lowering average acquisition cost.
  • The same path-plus-conformal template applies wherever quality improves step by step and the decision must be made before the outcome is fully observed.
  • Approximate conditional coverage is stated given the early reconstruction state the rule can actually see, not given unobserved true-image covariates.

Where Pith is reading between the lines

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

  • If reconstruction models with non-monotone error paths become common, the paper’s error-control step would need a different link from stopping time to loss, or a different score.
  • Prospective multi-coil MRI would require an online observable stand-in for next-band residuals; the current MRI results are retrospective on that point.
  • Sequential re-decisions along the path, flagged as future work, would need a new validity argument because naive reuse of conformal ranks can break exchangeability.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes Conformalized Rate-Adaptive Sensing (CoRAS), which chooses an image-specific acquisition stopping time so that reconstruction error stays below a user target c with probability at least 1−α. After a fixed early decision time t0, a horizontal step fits a monotone polynomial to the log next-band residual history and extrapolates a plug-in stopping time; a vertical step then applies full conformal prediction with state-dependent weights on (horizontal prediction, decision-time entropy), yielding a data-dependent upper bound Û^HV. Theorem 2 gives finite-sample marginal validity under exchangeability, symmetric bandwidth selection, and nonincreasing loss; Theorem 3 gives an approximate conditional guarantee under an RKHS bias model. Experiments on Fashion-MNIST and M4Raw report target stopping-time coverage, lower average/excess sampling than fixed-rate conformal rules, and higher rates on high-entropy images.

Significance. The problem—when to stop acquiring measurements under a reconstruction-error guarantee—is practically important in imaging and fits a broader class of pathwise cost–quality trade-offs. The technical contribution is genuine: path-based prediction of an unobserved stopping time, embedded in full conformal calibration with a carefully symmetrized bandwidth selector, plus a model-based approximate conditional theory. Marginal validity (Theorem 2) is standard but correctly specialized; the online O(Gn + G N_cand + n N_cand) bandwidth algorithm and released code support reproducibility. If the error-control claim holds in prospective settings with verified monotone loss, CoRAS is a useful addition to conformal risk/decision methods and adaptive sensing.

major comments (3)
  1. [§5; Theorems 1–3; Eq. (4); Appendix A.2] Theorems 1–3 convert stopping-time coverage {T ≤ Û} into reconstruction-error control {L(Û) ≤ c} only through the assumption that L_i(t) in (4) is nonincreasing in t with probability one. Section 5 defines empirical coverage as the fraction with Û_i ≥ T_i and states this is “equivalent” to L ≤ c under that assumption, but the paper never reports the fraction of paths with non-monotone MSE, nor the direct rate P{L_{n+1}(Û) ≤ c}. Appendix A.2 even trains the M4Raw U-Net with monotonicity and zero-filled-anchor penalties, which indicates non-monotonicity is a live modeling risk. Without these checks, the headline experimental support for the abstract’s error-control claim is indirect. Please report (i) empirical monotonicity-violation rates on both datasets/models and (ii) direct error-control coverage L(Û) ≤ c alongside Û ≥ T.
  2. [§5.2; Appendix A.2; Abstract] For multi-coil M4Raw, Appendix A.2 states that next-band residuals I^MRI_{i,t} are computed from the fully sampled RSS reference because coil combination is nonlinear, so the horizontal step is not online-observable and the evaluation is retrospective. The main text (§5.2, abstract, introduction) still presents M4Raw as evidence for adaptive sensing alongside Fashion-MNIST. This overstates what the MRI experiment demonstrates: calibration of a stopping rule given oracle residual histories, not prospective rate-adaptive acquisition. Either restrict M4Raw claims to retrospective validation, or supply an online-observable surrogate for I_t and re-run the adaptive rule with that surrogate.
  3. [§5.1–5.2 Figures 2–4; Theorem 3; Assumption 2] Figures 2(a) and 4(a) show coverage falling below 1−α on the highest true-entropy bins for all rules, including CoRAS, despite increased sampling on those bins. Theorem 3’s approximate conditional bound depends on residual RKHS bias Λ_{κ,n} and neighborhood MMD D_{κ,n}; the paper does not diagnose whether the high-entropy gap is large horizontal bias, poor state matching, discrete-grid effects, or insufficient t0. A short ablation (e.g., larger t0, state without entropy, calibration-only vs exact bandwidth) and explicit discussion of when approximate conditional validity fails would make the conditional claims falsifiable rather than only marginally reassuring.
minor comments (5)
  1. [§4.1.2; §5.1] Assumption 1 and Propositions 1–2 motivate the log-linear/polynomial residual model on a log-spaced grid, but Fashion-MNIST uses uniform column increments θ_t = t/32. A sentence in §4.1.2 already notes the polynomial remains a local approximation on general grids; briefly quantify fit quality (e.g., CV residual or calibration of T̂^H vs T) so readers can judge misspecification severity.
  2. [§5] The decision time t0 is a free design choice with large efficiency impact. Sensitivity of coverage and excess sampling to t0 (beyond the single values t0=6 and t0=9) would strengthen the experimental section.
  3. [§5; Appendix C.3] Clarify in the main text (not only Appendix C) whether experiments use exact candidate-specific bandwidth selection or the calibration-only approximation; Theorem 2 covers only the symmetric exact procedure.
  4. [§3–4] Minor notation: raw vs model losses L^raw vs L, and floored T^⋆ vs T, are clear once introduced but dense in §4.2; a small notation table would help.
  5. [§2; §6] Related work on active MRI and early time-series classification is appropriate; a brief pointer to anytime/sequential conformal testing literature would better situate the Discussion’s sequential-extension limitation.

Circularity Check

0 steps flagged

No significant circularity: coverage follows from exchangeability of conformal residual scores, not from fitted horizontal/vertical machinery being true by construction.

full rationale

CoRAS’s load-bearing guarantee (Theorem 2) is a standard full-conformal rank argument: under exchangeability of images, permutation-invariant bandwidth selection (15), and nonincreasing loss, the residual scores R_j(t*) are exchangeable, so p_HV(t*) is super-uniform and the largest retained candidate upper-bounds T*_n+1 with probability ≥1−α; monotonicity then converts stopping-time coverage into L≤c. The paper states explicitly that this does not require the horizontal extrapolation model to be correctly specified. Horizontal monotone polynomial fits (Section 4.1), the ordered-tail Assumption 1 / Propositions 1–2, entropy state features, and LOO bandwidth grids (Appendix C) are estimation machinery that shape efficiency and approximate conditional coverage (Theorem 3); they do not define c, α, or the coverage event, nor do they force ˆU≥T by construction. Self-citation of Zhang & Candès (2024) appears only as related work on local conformal guarantees and is not used as a uniqueness theorem or premise of Theorem 2. Experimental “coverage” as ˆU_i≥T_i is the conformal event itself (with L≤c claimed only under the separate monotonicity assumption)—a correctness/assumption gap, not circular reduction of a prediction to its fitted inputs. No equation equates a claimed first-principles prediction with a quantity fitted from the same target.

Axiom & Free-Parameter Ledger

6 free parameters · 7 axioms · 2 invented entities

Finite-sample marginal validity rests on classical exchangeability plus monotone loss and symmetric score construction. Efficiency and approximate conditional claims lean on modeling choices (ordered-tail residuals, RKHS bias, state features, fixed t0, bandwidth grid) and, for MRI, retrospective residual access. No new physical entities; free parameters are procedural knobs (t0, c, α, polynomial order cap, bandwidth grid, entropy bins).

free parameters (6)
  • decision time t0 = 6 (Fashion-MNIST), 9 (M4Raw)
    Fixed early stopping of acquisition for all images before the adaptive decision; chosen so most images still exceed c (t0=6 FMNIST; t0=9 M4Raw).
  • target error level c = 0.003 (Fashion-MNIST), 0.0015 (M4Raw)
    User threshold defining the stopping time T; directly sets the quality/budget tradeoff.
  • miscoverage level α = 0.1
    Conformal level for the upper bound on stopping time.
  • bandwidth grid B / G = {0.0125×2^m : m=-4..4}^2 (81 pairs)
    Prespecified squared-bandwidth pairs for state matching; selected by leave-one-out squared residual loss.
  • polynomial order cap q_max and CV folds = q_max=5
    Controls flexibility of horizontal log-residual extrapolation; order chosen image-wise by CV on pre-t0 history.
  • entropy binning (16 bins) = 16 equal bins on [0,1]
    Defines decision-time complexity feature Ω̂ used in the matching state.
axioms (7)
  • domain assumption Images Y1,...,Yn+1 are exchangeable (and reconstruction model trained on separate data).
    Load-bearing for Theorems 1–2 conformal rank validity (§3–4).
  • domain assumption Reconstruction loss L_i(t) is nonincreasing in t with probability one.
    Converts stopping-time coverage into error control at the chosen rate (Theorems 1–3).
  • standard math Bandwidth-selection algorithm is permutation-invariant on the augmented sample (eq. 15).
    Required so residual scores remain exchangeable for Theorem 2.
  • ad hoc to paper Assumption 1: transform residual spectrum follows A_i x^{-p_i(τ)} on a log-spaced band grid.
    Motivates linear/local-polynomial log next-band residual trends for horizontal prediction (Prop. 1–2); not needed for marginal validity.
  • ad hoc to paper Assumption 2: horizontal residual bias is an RKHS function of state plus i.i.d. sub-Gaussian noise with bounded density.
    Used only for approximate conditional coverage (Theorem 3).
  • domain assumption Target level attainable on the grid; stopping at t_max always controls loss.
    Stated in §3 so the discrete stopping problem is well-posed.
  • domain assumption For multi-coil MRI, next-band residuals may be computed from fully sampled RSS reference (retrospective).
    Appendix A.2; weakens the claim of online adaptive sensing on M4Raw.
invented entities (2)
  • CoRAS horizontal–vertical stopping rule (Û^HV) and state S=(T̂^H, Ω̂) independent evidence
    purpose: Produce image-specific conformal upper bounds on reconstruction stopping times from early paths.
    Methodological construct, not a physical entity; defined operationally from residuals, entropy, and calibration weights.
  • Continuous ordered-tail residual model (Assumption 1) no independent evidence
    purpose: Justify monotone polynomial extrapolation of next-band residual energy.
    Idealized generative assumption for residual spectra; paper notes real textures/artifacts deviate.

pith-pipeline@v1.2.0-daily-grok45 · 33567 in / 3659 out tokens · 71424 ms · 2026-07-30T18:04:01.618911+00:00 · methodology

0 comments
read the original abstract

Many high-resolution imaging systems face the same fundamental question: when have enough measurements been collected to reconstruct an image accurately? We develop Conformalized Rate-Adaptive Sensing (CoRAS), a method that adaptively chooses an acquisition or compression rate for each image while keeping the reconstruction error below a target level with high probability. As measurements are collected, an image reconstruction model gradually recovers the true image, producing a reconstruction path over acquisition rates. CoRAS uses this path up to an early decision time to estimate the target stopping time, defined as the first time at which the reconstruction error falls below the target level. It then calibrates this estimate using images with similar early reconstruction behavior, producing an upper bound on the stopping time with marginal and approximate conditional coverage guarantees. Experiments on image datasets show that CoRAS attains the target stopping-time coverage, uses fewer measurements on average than fixed-rate stopping rules, and assigns more measurements to images that are harder to reconstruct.

Figures

Figures reproduced from arXiv: 2607.26887 by Jiawei Yang, Yao Zhang.

Figure 1
Figure 1. Figure 1: Performance of stopping rules on Fashion-MNIST over 50 independent runs. [PITH_FULL_IMAGE:figures/full_fig_p017_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Performance of stopping rules over true-image entropy in Fashion-MNIST. [PITH_FULL_IMAGE:figures/full_fig_p018_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Performance of stopping rules on M4Raw over 50 independent runs. [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Performance of stopping rules over true-image entropy in M4Raw. [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Reference Fashion-MNIST test images spanning the clothing classes. [PITH_FULL_IMAGE:figures/full_fig_p025_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Reconstruction entropy at t0 versus true-image entropy on the 6000 Fashion￾MNIST test images. Left: scatter plot with a linear fit. Right: hexagonal bin density. 0.935 and Spearman correlation 0.931. The decision-time entropy is therefore a reliable proxy for the complexity of the true image, which is not observed at test time, and this supports its use in the state Si . A.2 M4Raw brain MRI The reference r… view at source ↗
Figure 7
Figure 7. Figure 7: Reference RSS reconstructions from eight M4Raw test studies (columns) at [PITH_FULL_IMAGE:figures/full_fig_p027_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Reconstruction entropy at t0 versus true-image entropy on the 1008 M4Raw test slices. Left: scatter plot with a linear fit. Right: hexagonal bin density. B Additional results for the horizontal prediction This appendix collects two results that support the horizontal prediction model presented in Section 4.1. Appendix B.1 extends Proposition 1 to the case where the tail exponent varies with acquisition tim… view at source ↗

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

35 extracted references · 5 linked inside Pith

  1. [1]

    Proceedings of the 39th International Conference on Machine Learning , series=

    Image-to-image regression with distribution-free uncertainty quantification and applications in imaging , author=. Proceedings of the 39th International Conference on Machine Learning , series=. 2022 , publisher=

    2022

  2. [2]

    International Conference on Medical image computing and computer-assisted intervention , pages=

    U-net: Convolutional networks for biomedical image segmentation , author=. International Conference on Medical image computing and computer-assisted intervention , pages=. 2015 , organization=

    2015

  3. [3]

    arXiv preprint arXiv:2411.11824 , year=

    Theoretical Foundations of Conformal Prediction , author=. arXiv preprint arXiv:2411.11824 , year=

    Pith/arXiv arXiv

  4. [4]

    arXiv preprint arXiv:1708.07747 , year=

    Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms , author=. arXiv preprint arXiv:1708.07747 , year=

    Pith/arXiv arXiv

  5. [5]

    2005 , publisher=

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

    2005

  6. [6]

    IEEE Transactions on Radiation and Plasma Medical Sciences , volume=

    Deep learning for PET image reconstruction , author=. IEEE Transactions on Radiation and Plasma Medical Sciences , volume=. 2020 , publisher=

    2020

  7. [7]

    Journal of the American Statistical Association , volume=

    Distribution-free predictive inference for regression , author=. Journal of the American Statistical Association , volume=. 2018 , publisher=

    2018

  8. [8]

    Advances in Neural Information Processing Systems , volume=

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

  9. [9]

    Biometrika , volume=

    Localized conformal prediction: a generalized inference framework for conformal prediction , author=. Biometrika , volume=. 2023 , publisher=

    2023

  10. [10]

    Journal of the ACM , volume=

    Distribution-free, risk-controlling prediction sets , author=. Journal of the ACM , volume=. 2021 , publisher=

    2021

  11. [11]

    The Annals of Applied Statistics , volume=

    Learn then test: Calibrating predictive algorithms to achieve risk control , author=. The Annals of Applied Statistics , volume=

  12. [12]

    International Conference on Learning Representations , year=

    Conformal risk control , author=. International Conference on Learning Representations , year=

  13. [13]

    The economic costs of conflict: A case study of the

    Abadie, Alberto and Gardeazabal, Javier , journal=. The economic costs of conflict: A case study of the

  14. [14]

    Synthetic control methods for comparative case studies: Estimating the effect of

    Abadie, Alberto and Diamond, Alexis and Hainmueller, Jens , journal=. Synthetic control methods for comparative case studies: Estimating the effect of. 2010 , publisher=

    2010

  15. [15]

    Journal of Economic Literature , volume=

    Using synthetic controls: Feasibility, data requirements, and methodological aspects , author=. Journal of Economic Literature , volume=

  16. [16]

    Experimental design for

    Bakker, Tim and van Hoof, Herke and Welling, Max , booktitle=. Experimental design for

  17. [17]

    International Conference on Learning Representations , volume=

    Scaling LLM test-time compute optimally can be more effective than scaling parameters for reasoning , author=. International Conference on Learning Representations , volume=

  18. [18]

    arXiv preprint arXiv:2404.17487 , year=

    Conformal prediction with learned features , author=. arXiv preprint arXiv:2404.17487 , year=

    Pith/arXiv arXiv

  19. [19]

    arXiv preprint arXiv:2409.19712 , year=

    Posterior conformal prediction , author=. arXiv preprint arXiv:2409.19712 , year=

    Pith/arXiv arXiv

  20. [20]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Conformal prediction with local weights: randomization enables robust guarantees , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2025 , publisher=

    2025

  21. [21]

    Nature machine intelligence , volume=

    Deep learning for tomographic image reconstruction , author=. Nature machine intelligence , volume=. 2020 , publisher=

    2020

  22. [22]

    ACM SIGGRAPH 2022 conference proceedings , pages=

    Palette: Image-to-image diffusion models , author=. ACM SIGGRAPH 2022 conference proceedings , pages=

    2022

  23. [23]

    A position paper by the

    Usability of irreversible image compression in radiological imaging. A position paper by the. Insights into Imaging , volume=. 2011 , doi=

    2011

  24. [24]

    Information , volume=

    The Current Role of Image Compression Standards in Medical Imaging , author=. Information , volume=. 2017 , doi=

    2017

  25. [25]

    BMC Medical Imaging , volume=

    Determining optimal medical image compression: psychometric and image distortion analysis , author=. BMC Medical Imaging , volume=. 2012 , doi=

    2012

  26. [26]

    Applied Sciences , volume=

    Error-Bounded Learned Scientific Data Compression with Preservation of Derived Quantities , author=. Applied Sciences , volume=. 2022 , doi=

    2022

  27. [27]

    Proceedings of the 31st International Symposium on High-Performance Parallel and Distributed Computing , pages=

    Ultrafast Error-bounded Lossy Compression for Scientific Datasets , author=. Proceedings of the 31st International Symposium on High-Performance Parallel and Distributed Computing , pages=. 2022 , doi=

    2022

  28. [28]

    IEEE Transactions on Biomedical Engineering , volume=

    Compression in Wearable Sensor Nodes: Impacts of Node Topology , author=. IEEE Transactions on Biomedical Engineering , volume=. 2014 , doi=

    2014

  29. [29]

    IEEE Sensors Letters , volume=

    Direct Lightweight Temporal Compression for Wearable Sensor Data , author=. IEEE Sensors Letters , volume=. 2021 , doi=

    2021

  30. [30]

    Machine Learning , volume=

    Early classification of time series: cost-based optimization criterion and algorithms , author=. Machine Learning , volume=. 2021 , doi=

    2021

  31. [31]

    Pineda, Luis and Basu, Sumana and Romero, Adriana and Calandra, Roberto and Drozdzal, Michal , booktitle=. Active. 2020 , publisher=

    2020

  32. [32]

    arXiv preprint arXiv:2602.13935 , year=

    Statistical Early Stopping for Reasoning Models , author=. arXiv preprint arXiv:2602.13935 , year=

    Pith/arXiv arXiv

  33. [33]

    M4Raw: A multi-contrast, multi-repetition, multi-channel

    Lyu, Mengye and Mei, Lifeng and Huang, Shoujin and Liu, Sixing and Li, Yi and Yang, Kexin and Liu, Yilong and Dong, Yu and Dong, Linzheng and Wu, Ed X , journal=. M4Raw: A multi-contrast, multi-repetition, multi-channel. 2023 , publisher=

    2023

  34. [34]

    Journal of the American Statistical Association , volume=

    The augmented synthetic control method , author=. Journal of the American Statistical Association , volume=. 2021 , publisher=

    2021

  35. [35]

    American Economic Review , volume=

    Synthetic difference-in-differences , author=. American Economic Review , volume=