pith. sign in

arxiv: 2607.02173 · v1 · pith:VDFLUJQXnew · submitted 2026-07-02 · 📊 stat.ME · stat.ML

Conformal Bayes for Two-Sided Censored Gaussian Regression under Label Shift

Seungjin Choi This is my paper

Pith reviewed 2026-07-03 07:44 UTC · model grok-4.3

classification 📊 stat.ME stat.ML
keywords conformal predictionlabel shiftcensored regressionTobit modelBayesian predictionprediction setsmixed distributions
0
0 comments X

The pith

Weighted tilted conformal Bayes restores marginal coverage with smaller sets for two-sided censored Gaussian regression under label shift.

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

The paper develops conformal Bayes for prediction under label shift when responses are two-sided censored in a Gaussian model. Observed data form a mixed distribution with atoms at the lower and upper bounds plus a continuous density inside the interval. It combines posterior predictive tilting with weighted conformal calibration, deriving a mixed calibration weight from a latent exponential tilt that handles boundary atoms separately from the interior density ratio. Synthetic experiments show this approach achieves correct marginal coverage while producing smaller sets than weighted source-score calibration.

Core claim

In a two-sided Tobit Gaussian Bayesian prediction head with Laplace posterior approximation, the tilted predictive distribution has left-atom, interior, and right-atom components with a three-term closed-form normalizer; a latent exponential tilt induces tail-averaged atom weights at the censored boundaries while the interior ratio remains density based, producing a mixed observed-space calibration weight with two atom ratios and one interior density ratio that corrects the calibration measure for target-adapted mixed-HDR geometry.

What carries the argument

Mixed observed-space calibration weight with two atom ratios and one interior density ratio, obtained via latent exponential tilt on the censored scale.

If this is right

  • Prediction sets can combine boundary atoms with an interior interval or reduce to atom-only sets under strong censoring.
  • The method restores marginal coverage with smaller sets than weighted source-score calibration.
  • A trade-off exists between marginal coverage and component-wise behavior across atoms and interior observations.

Where Pith is reading between the lines

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

  • The mixed-weight construction could be adapted to other mixed discrete-continuous observation models beyond censoring.
  • Component-wise coverage diagnostics might be developed to balance the observed marginal versus per-component behavior.

Load-bearing premise

A latent exponential tilt induces tail-averaged atom weights at the censored boundaries while the interior ratio remains density based, allowing a mixed observed-space calibration weight with two atom ratios and one interior density ratio.

What would settle it

A simulation under the two-sided Tobit Gaussian model where the weighted tilted conformal Bayes sets fail to attain the nominal marginal coverage probability.

Figures

Figures reproduced from arXiv: 2607.02173 by Seungjin Choi.

Figure 1
Figure 1. Figure 1: Two-sided censored Gaussian observation model. A latent Gaussian response [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Schematic of split conformal Bayes on the censored mixed space. Censoring maps a latent [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Visual summary of the three calibration configurations in split conformal Bayes. UT uses [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
read the original abstract

Prediction under label shift becomes nonstandard when responses are censored. In a two-sided censored Gaussian model, latent values below $L$ and above $U$ are recorded at the boundary values, so the observed predictive distribution is mixed, with atoms at $L$ and $U$ and a continuous density on $(L,U)$. In this paper we develop conformal Bayes for this mixed-space setting by combining posterior predictive tilting with weighted conformal calibration. Under a two-sided Tobit Gaussian Bayesian prediction head with a Laplace posterior approximation, the tilted predictive distribution has left-atom, interior, and right-atom components, with a three-term closed-form normalizer. The resulting prediction set is a mixed highest density region that can combine boundary atoms with an interior interval and can reduce to atom-only sets under strong censoring. The main technical issue is that latent label shift does not directly give an ordinary density ratio on the observed censored scale. A latent exponential tilt induces tail-averaged atom weights at the censored boundaries, while the interior ratio remains density based. This yields a mixed observed-space calibration weight with two atom ratios and one interior density ratio. The weight corrects the calibration measure, while predictive tilting gives target-adapted mixed-HDR geometry. Synthetic experiments show that weighted tilted conformal Bayes restores marginal coverage with smaller sets than weighted source-score calibration, while revealing a trade-off between marginal coverage and component-wise behavior across atoms and interior observations.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

0 major / 3 minor

Summary. The manuscript develops conformal Bayes for two-sided censored Gaussian regression under label shift. In the Tobit model, the observed response is a mixed distribution with atoms at the censoring bounds L and U and a continuous density on (L,U). The method combines a Laplace-approximated Bayesian posterior predictive with exponential tilting to produce a three-component tilted predictive (left atom, interior density, right atom) whose normalizer is closed-form. Calibration weights are constructed via a latent exponential tilt that supplies tail-averaged atom ratios at the boundaries together with the ordinary density ratio in the interior; these weights are used in weighted conformal calibration. The resulting prediction sets are mixed highest-density regions that may include atoms and/or an interior interval. Synthetic experiments are reported to show that the weighted tilted procedure restores marginal coverage while producing smaller sets than weighted source-score calibration, with an observed trade-off between marginal and component-wise coverage.

Significance. If the derivations hold, the work supplies a concrete, closed-form construction for conformal prediction under label shift when responses are two-sided censored, a setting that arises in many applied regression problems. The explicit separation of atom and interior weights, the mixed-HDR geometry, and the comparison against a natural baseline are useful contributions. The paper also demonstrates that the latent-tilt construction yields exact marginal coverage in the observed space without requiring additional assumptions on the censoring mechanism beyond the model.

minor comments (3)
  1. [§3] §3 (or wherever the three-term normalizer is derived): the normalizer expression should be written out explicitly with the three components labeled, so that the reader can verify the cancellation that produces the closed form.
  2. [Experiments] The synthetic experiments section should report the precise values of L and U used, the degree of censoring (fraction of observations at each atom), and the label-shift strength (e.g., the tilt parameter), so that the coverage numbers can be reproduced.
  3. [Figures] Figure captions for the coverage and set-size plots should state whether the plotted intervals are pointwise or simultaneous and whether they are over replications or over test points.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive summary and significance assessment of the manuscript, as well as the recommendation for minor revision. No specific major comments were raised in the report.

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The provided abstract and description outline a construction that combines posterior predictive tilting (via latent exponential tilt inducing tail-averaged atom weights) with weighted conformal calibration to produce mixed observed-space weights and mixed-HDR prediction sets. No equations, self-citations, or fitted inputs are quoted that reduce any central prediction or normalizer to its own inputs by construction. The Laplace approximation and three-term normalizer are presented as standard applications rather than self-referential. Synthetic experiments are described as external validation of coverage and set size, not as the derivation itself. The chain is therefore self-contained against external conformal and tilting benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities can be identified from the provided text. The approach relies on a Laplace posterior approximation and a latent exponential tilt, but these are not detailed enough to ledger.

pith-pipeline@v0.9.1-grok · 5776 in / 1141 out tokens · 22367 ms · 2026-07-03T07:44:56.890778+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

29 extracted references · 1 canonical work pages · 1 internal anchor

  1. [1]

    J. H. Albert and S. Chib. Bayesian analysis of binary and polychotomous response data. Journal of the American Statistical Association, 88 0 (422): 0 669--679, 1993

    1993

  2. [2]

    A. M. Alexandari, A. Kundaje, and A. Shrikumar. Maximum likelihood with bias-corrected calibration is hard-to-beat at label shift adaptation. In Proceedings of the International Conference on Machine Learning (ICML), 2020

    2020

  3. [3]

    T. Amemiya. Tobit models: A survey. Journal of Econometrics, 24: 0 3--61, 1984

    1984

  4. [4]

    A. N. Angelopoulos and S. Bates. Conformal prediction: A gentle introduction. Foundations and Trends ^ in Machine Learning, 16 0 (4): 0 494--591, 2023

    2023

  5. [5]

    Azizzadenesheli, F

    K. Azizzadenesheli, F. Yang, A. Liu, and A. Anandkumar. Regularized learning for domain adaptation under label shifts. In Proceedings of the International Conference on Learning Representations (ICLR), 2019

    2019

  6. [6]

    R. F. Barber, E. J. Cand \`e s, A. Ramdas, and R. J. Tibshirani. Conformal prediction beyond exchangeability. The Annals of Statistics, 51 0 (2): 0 816--845, 2023

    2023

  7. [7]

    Cand \`e s, L

    E. Cand \`e s, L. Lei, and Z. Ren. Conformalized survival analysis. Journal of the Royal Statistical Society Series B, 85 0 (1): 0 24--45, 2023

    2023

  8. [8]

    S. Choi. Conformal Bayes under label shift: Post-hoc calibration vs. in-training adaptation. In The 2nd Workshop on Epistemic Intelligence in Machine Learning, 2026. URL https://arxiv.org/abs/2606.11865

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  9. [9]

    Davidov, S

    H. Davidov, S. Feldman, G. Shamai, R. Kimmel, and Y. Romano. Conformalized survival analysis for general right-censored data. In Proceedings of the International Conference on Learning Representations (ICLR), 2025

    2025

  10. [10]

    Fong and C

    E. Fong and C. Holmes. Conformal Bayesian computation. In Advances in Neural Information Processing Systems (NeurIPS), 2021

    2021

  11. [11]

    S. Garg, Y. Wu, S. Balakrishnan, and Z. C. Lipton. A unified view of label shift estimation. In Advances in Neural Information Processing Systems (NeurIPS), 2020

    2020

  12. [12]

    Y. Gui, R. Hore, Z. Ren, and R. F. Barber. Conformalized survival analysis with adaptive cut-offs. Biometrika, 111 0 (2): 0 459--477, 2024

    2024

  13. [13]

    D. R. Helsel. Statistics for Censored Environmental Data Using Minitab and R. John Wiley \ Sons, Inc., 2011

    2011

  14. [14]

    Y. Jin, Z. Ren, and E. J. Cand \`e s. Sensitivity analysis of individual treatment effects: A robust conformal inference approach. Proceedings of the National Academy of Sciences, USA, 120 0 (6), 2023

    2023

  15. [15]

    R. J. Keizer, R. S. Jansen, H. Rosing, B. Thijssen, J. H. Beijnen, J. H. M. Schellens, and A. D. R. Huitema. Incorporation of concentration data below the limit of quantification in population pharmacokinetic analyses. Pharmacol Res Perspect., 3 0 (2), 2015

    2015

  16. [16]

    H. Lee, J. Kim, E. Jadamba, S. Choi, and H. Shin. Conformal prediction for molecular properties under label shift. In NeurIPS 2025 Workshop on Reliable ML from Unreliable Data, 2025

    2025

  17. [17]

    H. Lee, J. Kim, E. Jadamba, S. Choi, and H. Shin. Split conformal prediction with label-shift-adjusted Bayesian scores. In The 2nd Workshop on Epistemic Intelligence in Machine Learning, 2026

    2026

  18. [18]

    Z. C. Lipton, Y.-X. Wang, and A. J. Smola. Detecting and correcting for label shift with black box predictors. In Proceedings of the International Conference on Machine Learning (ICML), 2018

    2018

  19. [19]

    Papadopoulos, K

    H. Papadopoulos, K. Proedrou, V. Vovk, and A. Gammerman. Inductive confidence machines for regression. In Proceedings of the European Conference on Machine Learning (ECML), 2002

    2002

  20. [20]

    R. N. Rosett and F. D. Nelson. Estimation of the two-limit probit regression model. Econometrica, 43 0 (1): 0 141--146, 1975

    1975

  21. [21]

    Saerens, P

    M. Saerens, P. Latinne, and C. Decaestecker. Adjusting the outputs of a classifier to new a priori probabilities: A simple procedure. Neural Computation, 2002

    2002

  22. [22]

    Sesia and V

    M. Sesia and V. Svetnik. Doubly robust conformalized survival analysis with right-censored data. In Proceedings of the International Conference on Machine Learning (ICML), 2025

    2025

  23. [23]

    Shafer and V

    G. Shafer and V. Vovk. A tutorial on conformal prediction. Journal of Machine Learning Research, 9: 0 371--421, 2008

    2008

  24. [24]

    R. J. Tibshirani, R. F. Barber, E. J. Cand \`e s, and A. Ramdas. Conformal prediction under covariate shift. In Advances in Neural Information Processing Systems (NeurIPS), 2019

    2019

  25. [25]

    Tierney and J

    L. Tierney and J. B. Kadane. Accurate approximations for posterior moments and marginal densities. Journal of the American Statistical Association, 81 0 (393): 0 82--86, 1986

    1986

  26. [26]

    J. Tobin. Estimation of relationships for limited dependent variables. Econometrica, 26 0 (1): 0 24--36, 1958

    1958

  27. [27]

    V. Vovk, A. Gammerman, and G. Shafer. Algorithmic Learning in a Random World. Springer, 2005

    2005

  28. [28]

    Wasserman

    L. Wasserman. Frasian inference. Statistical Science, 26 0 (3): 0 322--325, 2011

    2011

  29. [29]

    J. R. Williams, H.-W. Kim, and C. M. Crespi. Modeling observations with a detection limit using a truncated normal distribution with censoring. BMC Medical Research Methodology, 20, 2020

    2020