pith. sign in

arxiv: 2604.01870 · v1 · submitted 2026-04-02 · 💻 cs.LG · cs.SY· eess.SY

Towards Intrinsically Calibrated Uncertainty Quantification in Industrial Data-Driven Models via Diffusion Sampler

Yiran Ma , Jerome Le Ny , Zhichao Chen , Zhihuan Song This is my paper

Pith reviewed 2026-05-13 22:11 UTC · model grok-4.3

classification 💻 cs.LG cs.SYeess.SY
keywords diffusion modelsuncertainty quantificationposterior samplingindustrial soft sensorscalibrated predictionsdata-driven modelingammonia synthesispredictive uncertainty
0
0 comments X

The pith

A diffusion-based posterior sampler produces inherently calibrated uncertainty estimates for industrial data-driven models without post-hoc calibration.

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

Data-driven models are essential for real-time monitoring in process industries where direct KPI measurements are impractical. Reliable uncertainty quantification matters for safety and decisions but often requires separate calibration steps that can be unreliable. The paper introduces a diffusion sampler that draws from the posterior distribution to generate calibrated uncertainties directly. Evaluations on synthetic distributions, a Raman phenylacetic acid soft sensor benchmark, and a real ammonia synthesis case show gains in both calibration quality and predictive accuracy over existing techniques.

Core claim

A diffusion-based posterior sampling framework inherently produces well-calibrated predictive uncertainty via faithful posterior sampling, eliminating the need for post-hoc calibration. This is demonstrated through improvements over existing UQ techniques in both uncertainty calibration and predictive accuracy across synthetic distributions, the Raman-based phenylacetic acid soft sensor benchmark, and a real ammonia synthesis case study.

What carries the argument

The diffusion-based posterior sampling framework that approximates and draws samples from the true posterior distribution to compute predictive uncertainties directly.

Load-bearing premise

The diffusion sampler faithfully approximates and samples from the true posterior distribution on complex real-world industrial datasets without extra assumptions or post-processing.

What would settle it

If the predicted uncertainty intervals on the ammonia synthesis case study fail to achieve the nominal coverage rate for observed errors, the claim of inherent calibration would not hold.

Figures

Figures reproduced from arXiv: 2604.01870 by Jerome Le Ny, Yiran Ma, Zhichao Chen, Zhihuan Song.

Figure 1
Figure 1. Figure 1: Overview of the DiffUQ Framework Algorithm 1 Training a diffusion sampler Require: Dataset D = {xi , yi} N y=1, probabilistic model p(·|x, θ), prior p(θ) = N (0, I). Ensure: uϕ(t, θ) parameterized by ϕ 1: Define: Augmented SDE drift fϕ(t, [θt, ct]) = [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Summary of the first principle-based mathematical [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison on the smiley-face (left) and funnel (right) [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: High-Low Transformer unit from an ammonia synthesis [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Sensitivity Analysis. The shaded area indicates [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Impact of Drift Network Size 1) Drift Network Capacity and Discretization Error: From a theoretical perspective, existing results, e.g., Theorem 2 in [21], ensure that a sufficiently expressive neural drift can approximate the target distribution in KL to arbitrary accu￾racy under mild regularity assumptions. However, these are existence results and do not account for finite-step numerical discretization. … view at source ↗
read the original abstract

In modern process industries, data-driven models are important tools for real-time monitoring when key performance indicators are difficult to measure directly. While accurate predictions are essential, reliable uncertainty quantification (UQ) is equally critical for safety, reliability, and decision-making, but remains a major challenge in current data-driven approaches. In this work, we introduce a diffusion-based posterior sampling framework that inherently produces well-calibrated predictive uncertainty via faithful posterior sampling, eliminating the need for post-hoc calibration. In extensive evaluations on synthetic distributions, the Raman-based phenylacetic acid soft sensor benchmark, and a real ammonia synthesis case study, our method achieves practical improvements over existing UQ techniques in both uncertainty calibration and predictive accuracy. These results highlight diffusion samplers as a principled and scalable paradigm for advancing uncertainty-aware modeling in industrial applications.

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

2 major / 2 minor

Summary. The manuscript introduces a diffusion-based posterior sampling framework for uncertainty quantification in industrial data-driven models. It claims that this approach inherently produces well-calibrated predictive uncertainty via faithful sampling from the posterior distribution, eliminating post-hoc calibration. Evaluations on synthetic distributions, the Raman phenylacetic acid soft sensor benchmark, and a real ammonia synthesis case study report practical improvements in both uncertainty calibration and predictive accuracy over existing UQ techniques.

Significance. If the faithfulness claim holds, the work offers a principled, scalable alternative to post-hoc calibration for UQ in process industries, where reliable uncertainty estimates are critical for safety and decision-making when direct KPI measurements are unavailable. The diffusion-sampler approach for posterior sampling in this domain is novel and could generalize beyond the reported benchmarks.

major comments (2)
  1. [Methods] The central claim of faithful posterior sampling (i.e., draws from the true p(θ|D) without post-hoc adjustment) is load-bearing but unsupported by convergence diagnostics, discretization-error bounds, or posterior-predictive checks. On complex industrial data (Raman benchmark and ammonia synthesis), finite score-matching training introduces approximation error whose magnitude is not quantified; downstream ECE improvements alone cannot distinguish faithful sampling from biased but well-tuned approximations.
  2. [Experiments] In the real-data experiments, only aggregate calibration metrics (ECE, sharpness) and predictive accuracy are reported. No comparison to exact posterior sampling baselines (e.g., MCMC) or diagnostics confirming sampler convergence to the target posterior is provided, leaving the 'intrinsically calibrated' assertion unverified for the ammonia synthesis case study.
minor comments (2)
  1. [Abstract] Define acronyms on first use (e.g., ECE, UQ) and ensure consistent notation for posterior p(θ|D) versus posterior predictive throughout.
  2. [Figures] Figure captions should explicitly state the number of diffusion steps, training epochs, and any hyperparameter choices used for the reported runs to aid reproducibility.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for their constructive comments, which help clarify the evidentiary requirements for our central claim of faithful posterior sampling. We address each major point below and have incorporated revisions to strengthen the validation of the diffusion sampler's behavior.

read point-by-point responses

  1. Referee: [Methods] The central claim of faithful posterior sampling (i.e., draws from the true p(θ|D) without post-hoc adjustment) is load-bearing but unsupported by convergence diagnostics, discretization-error bounds, or posterior-predictive checks. On complex industrial data (Raman benchmark and ammonia synthesis), finite score-matching training introduces approximation error whose magnitude is not quantified; downstream ECE improvements alone cannot distinguish faithful sampling from biased but well-tuned approximations.

    Authors: We agree that explicit convergence diagnostics and quantification of score-matching approximation error are needed to support the faithfulness claim. In the revised manuscript we have added posterior-predictive checks on the synthetic distributions, effective sample size and trace diagnostics for the diffusion sampler on both the Raman and ammonia cases, and a brief discussion of the discretization error incurred by the finite-step reverse SDE. We also report the magnitude of the score-matching loss on held-out data as a proxy for approximation quality. Exact non-asymptotic discretization bounds remain technically difficult for the non-log-concave posteriors encountered in industrial settings, but the added diagnostics allow readers to assess the practical fidelity of the samples. revision: yes

  2. Referee: [Experiments] In the real-data experiments, only aggregate calibration metrics (ECE, sharpness) and predictive accuracy are reported. No comparison to exact posterior sampling baselines (e.g., MCMC) or diagnostics confirming sampler convergence to the target posterior is provided, leaving the 'intrinsically calibrated' assertion unverified for the ammonia synthesis case study.

    Authors: We acknowledge that direct MCMC comparisons are absent from the real-data sections. For the high-dimensional ammonia synthesis model, standard MCMC is computationally prohibitive (mixing times exceed practical limits on the available hardware), which motivated the diffusion approach. In the revision we have added a new subsection that (i) validates the sampler against MCMC on lower-dimensional synthetic and Raman subsets where exact sampling is feasible, and (ii) reports convergence diagnostics (ESS, Gelman-Rubin statistic, and autocorrelation times) for the ammonia runs. These additions provide indirect but quantitative support for the claim that the reported ECE improvements arise from faithful sampling rather than post-hoc tuning. revision: partial

standing simulated objections not resolved
  • Exact MCMC sampling on the full ammonia synthesis posterior remains computationally intractable, precluding a direct head-to-head verification on that specific case study.

Circularity Check

0 steps flagged

No circularity: new diffusion sampling framework presented as independent method

full rationale

The paper introduces a diffusion-based posterior sampling approach claimed to produce calibrated uncertainty directly through faithful sampling, without post-hoc steps. No equations or steps in the abstract or described framework reduce by construction to fitted parameters renamed as predictions, self-citations as load-bearing premises, or ansatzes smuggled from prior author work. The derivation chain relies on the properties of score-based diffusion models applied to industrial data, with evaluations on synthetic distributions and real benchmarks (Raman phenylacetic acid, ammonia synthesis) serving as external checks rather than tautological re-expressions. This is self-contained against the stated assumptions of posterior approximation.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based on abstract only; no explicit free parameters, axioms, or invented entities are identifiable. The approach implicitly relies on standard diffusion model assumptions for posterior approximation.

pith-pipeline@v0.9.0 · 5445 in / 943 out tokens · 70207 ms · 2026-05-13T22:11:54.840584+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

40 extracted references · 40 canonical work pages

  1. [1]

    Data-driven soft sensors in the process industry,

    P. Kadlec, B. Gabrys, and S. Strandt, “Data-driven soft sensors in the process industry,”Computers & chemical engineering, vol. 33, no. 4, pp. 795–814, 2009

    work page 2009

  2. [2]

    Heat equation Stein variational ensemble: Rethinking and advancing uncertainty-aware soft sensor modeling,

    Y . Ma, Z. Chen, Z. Yang, X. Zhang, and Z. Song, “Heat equation Stein variational ensemble: Rethinking and advancing uncertainty-aware soft sensor modeling,”IEEE Transactions on Industrial Informatics, 2024

    work page 2024

  3. [3]

    M. J. Kochenderfer,Decision making under uncertainty: theory and application. MIT press, 2015

    work page 2015

  4. [4]

    Robust control and model uncertainty,

    L. P. Hansen and T. J. Sargent, “Robust control and model uncertainty,” American Economic Review, vol. 91, no. 2, pp. 60–66, 2001

    work page 2001

  5. [5]

    Optimization under uncertainty: state-of-the-art and opportunities,

    N. V . Sahinidis, “Optimization under uncertainty: state-of-the-art and opportunities,”Computers & chemical engineering, vol. 28, no. 6-7, pp. 971–983, 2004

    work page 2004

  6. [6]

    Deep Bayesian active learning with image data,

    Y . Gal, R. Islam, and Z. Ghahramani, “Deep Bayesian active learning with image data,” inInternational conference on machine learning. PMLR, 2017, pp. 1183–1192

    work page 2017

  7. [7]

    A practical Bayesian framework for backpropagation networks,

    D. J. MacKay, “A practical Bayesian framework for backpropagation networks,”Neural computation, vol. 4, no. 3, pp. 448–472, 1992

    work page 1992

  8. [8]

    Bayesian leaning for neural networks,

    R. M. Neal, “Bayesian leaning for neural networks,” 1996

    work page 1996

  9. [9]

    Process data analytics via probabilistic latent variable models: A tutorial review,

    Z. Ge, “Process data analytics via probabilistic latent variable models: A tutorial review,”Industrial & Engineering Chemistry Research, vol. 57, no. 38, pp. 12 646–12 661, 2018

    work page 2018

  10. [10]

    A survey on deep learning for data-driven soft sensors,

    Q. Sun and Z. Ge, “A survey on deep learning for data-driven soft sensors,”IEEE Transactions on Industrial Informatics, vol. 17, no. 9, pp. 5853–5866, 2021

    work page 2021

  11. [11]

    Understanding and accelerating particle-based variational inference,

    C. Liu, J. Zhuo, P. Cheng, R. Zhang, and J. Zhu, “Understanding and accelerating particle-based variational inference,” inInternational Conference on Machine Learning. PMLR, 2019, pp. 4082–4092

    work page 2019

  12. [12]

    Annealed Stein variational gradient de- scent,

    F. D’Angelo and V . Fortuin, “Annealed Stein variational gradient de- scent,” inThe Third Symposium on Advances in Approximate Bayesian Inference, 2021

    work page 2021

  13. [13]

    Remaining useful life prediction method for bearings based on lstm with uncertainty quantification,

    J. Yang, Y . Peng, J. Xie, and P. Wang, “Remaining useful life prediction method for bearings based on lstm with uncertainty quantification,” Sensors, vol. 22, no. 12, p. 4549, 2022

    work page 2022

  14. [14]

    A parallel GRU with dual- stage attention mechanism model integrating uncertainty quantification for probabilistic rul prediction of wind turbine bearings,

    L. Cao, H. Zhang, Z. Meng, and X. Wang, “A parallel GRU with dual- stage attention mechanism model integrating uncertainty quantification for probabilistic rul prediction of wind turbine bearings,”Reliability Engineering & System Safety, vol. 235, p. 109197, 2023

    work page 2023

  15. [15]

    Dropout as a Bayesian approximation: Representing model uncertainty in deep learning,

    Y . Gal and Z. Ghahramani, “Dropout as a Bayesian approximation: Representing model uncertainty in deep learning,” ininternational conference on machine learning. PMLR, 2016, pp. 1050–1059

    work page 2016

  16. [16]

    Uncertainty-aware soft sensor using Bayesian recurrent neural networks,

    M. Lee, J. Bae, and S. B. Kim, “Uncertainty-aware soft sensor using Bayesian recurrent neural networks,”Advanced Engineering Informatics, vol. 50, p. 101434, 2021

    work page 2021

  17. [17]

    Nonasymptotic estimates for stochastic gradient Langevin dynamics under local condi- tions in nonconvex optimization,

    Y . Zhang, ¨O. D. Akyildiz, T. Damoulas, and S. Sabanis, “Nonasymptotic estimates for stochastic gradient Langevin dynamics under local condi- tions in nonconvex optimization,”Applied Mathematics & Optimization, vol. 87, no. 2, p. 25, 2023

    work page 2023

  18. [18]

    A survey of the Schr ¨odinger problem and some of its con- nections with optimal transport,

    C. L ´eonard, “A survey of the Schr ¨odinger problem and some of its con- nections with optimal transport,”Discrete and Continuous Dynamical Systems, vol. 34, no. 4, pp. 1533–1574, 2013

    work page 2013

  19. [19]

    Probability densities with given marginals,

    S. Kullback, “Probability densities with given marginals,”The Annals of Mathematical Statistics, vol. 39, no. 4, pp. 1236–1243, 1968

    work page 1968

  20. [20]

    Schr ¨odinger- F¨ollmer sampler,

    J. Huang, Y . Jiao, L. Kang, X. Liao, J. Liu, and Y . Liu, “Schr ¨odinger- F¨ollmer sampler,”IEEE Transactions on Information Theory, vol. 71, no. 2, pp. 1283–1299, 2025

    work page 2025

  21. [21]

    Path integral sampler: A stochastic control approach for sampling,

    Q. Zhang and Y . Chen, “Path integral sampler: A stochastic control approach for sampling,” inThe Tenth International Conference on Learning Representations, ICLR, 2022

    work page 2022

  22. [22]

    Denoising diffusion samplers,

    F. Vargas, W. S. Grathwohl, and A. Doucet, “Denoising diffusion samplers,” inThe Eleventh International Conference on Learning Rep- resentations, ICLR, 2023

    work page 2023

  23. [23]

    Adjoint sampling: Highly scalable diffusion samplers via adjoint matching,

    A. J. Havens, B. K. Miller, B. Yan, C. Domingo-Enrich, A. Sriram, D. S. Levine, B. M. Wood, B. Hu, B. Amos, B. Karreret al., “Adjoint sampling: Highly scalable diffusion samplers via adjoint matching,” in International Conference on Machine Learning. PMLR, 2025, pp. 22 204–22 237

    work page 2025

  24. [24]

    Adjoint Schr¨odinger bridge sampler,

    G.-H. Liu, J. Choi, Y . Chen, B. K. Miller, and R. T. Chen, “Adjoint Schr¨odinger bridge sampler,” inThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

    work page 2025

  25. [25]

    Accurate uncertainties for deep learning using calibrated regression,

    V . Kuleshov, N. Fenner, and S. Ermon, “Accurate uncertainties for deep learning using calibrated regression,” inInternational conference on machine learning. PMLR, 2018, pp. 2796–2804

    work page 2018

  26. [26]

    What uncertainties do we need in Bayesian deep learning for computer vision?

    A. Kendall and Y . Gal, “What uncertainties do we need in Bayesian deep learning for computer vision?”Advances in neural information processing systems, vol. 30, 2017

    work page 2017

  27. [27]

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

    E. H ¨ullermeier and W. Waegeman, “Aleatoric and epistemic uncertainty in machine learning: An introduction to concepts and methods,”Machine learning, vol. 110, no. 3, pp. 457–506, 2021

    work page 2021

  28. [28]

    On the expressiveness of approximate inference in Bayesian neural networks,

    A. Foong, D. Burt, Y . Li, and R. Turner, “On the expressiveness of approximate inference in Bayesian neural networks,”Advances in Neural Information Processing Systems, vol. 33, pp. 15 897–15 908, 2020

    work page 2020

  29. [29]

    On the relation between optimal transport and Schr ¨odinger bridges: A stochastic control viewpoint,

    Y . Chen, T. T. Georgiou, and M. Pavon, “On the relation between optimal transport and Schr ¨odinger bridges: A stochastic control viewpoint,” Journal of Optimization Theory and Applications, vol. 169, no. 2, pp. 671–691, 2016

    work page 2016

  30. [30]

    Theoretical guarantees for sampling and inference in generative models with latent diffusions,

    B. Tzen and M. Raginsky, “Theoretical guarantees for sampling and inference in generative models with latent diffusions,” inConference on Learning Theory. PMLR, 2019, pp. 3084–3114

    work page 2019

  31. [31]

    Bayesian learning via neural Schr ¨odinger–F¨ollmer flows,

    F. Vargas, A. Ovsianas, D. Fernandes, M. Girolami, N. D. Lawrence, and N. N ¨usken, “Bayesian learning via neural Schr ¨odinger–F¨ollmer flows,” Statistics and Computing, vol. 33, no. 1, p. 3, 2023

    work page 2023

  32. [32]

    Integrating autoencoder and heteroscedastic noise neural networks for the batch process soft-sensor design,

    S. Kay, H. Kay, M. Mowbray, A. Lane, C. Mendoza, P. Martin, and D. Zhang, “Integrating autoencoder and heteroscedastic noise neural networks for the batch process soft-sensor design,”Industrial & Engi- neering Chemistry Research, vol. 61, no. 36, pp. 13 559–13 569, 2022

    work page 2022

  33. [33]

    Infinitely deep Bayesian neural networks with stochastic differential equations,

    W. Xu, R. T. Chen, X. Li, and D. Duvenaud, “Infinitely deep Bayesian neural networks with stochastic differential equations,” inInternational Conference on Artificial Intelligence and Statistics. PMLR, 2022, pp. 721–738

    work page 2022

  34. [34]

    Scalable gradients for stochastic differential equations,

    X. Li, T.-K. L. Wong, R. T. Q. Chen, and D. Duvenaud, “Scalable gradients for stochastic differential equations,”International Conference on Artificial Intelligence and Statistics, 2020

    work page 2020

  35. [35]

    Neural SDEs as infinite-dimensional GANs,

    P. Kidger, J. Foster, X. Li, H. Oberhauser, and T. Lyons, “Neural SDEs as infinite-dimensional GANs,”International Conference on Machine Learning, 2021

    work page 2021

  36. [36]

    Obtaining well calibrated probabilities using Bayesian binning,

    M. P. Naeini, G. Cooper, and M. Hauskrecht, “Obtaining well calibrated probabilities using Bayesian binning,” inProceedings of the AAAI conference on artificial intelligence, vol. 29, no. 1, 2015

    work page 2015

  37. [37]

    Simple and scalable predictive uncertainty estimation using deep ensembles,

    B. Lakshminarayanan, A. Pritzel, and C. Blundell, “Simple and scalable predictive uncertainty estimation using deep ensembles,”Advances in neural information processing systems, vol. 30, 2017

    work page 2017

  38. [38]

    Bayesian learning via stochastic gradient Langevin dynamics,

    M. Welling and Y . W. Teh, “Bayesian learning via stochastic gradient Langevin dynamics,” inProceedings of the 28th international conference on machine learning (ICML-11), 2011, pp. 681–688

    work page 2011

  39. [39]

    Modern day monitoring and control challenges outlined on an industrial-scale benchmark fermentation process,

    S. Goldrick, C. A. Duran-Villalobos, K. Jankauskas, D. Lovett, S. S. Farid, and B. Lennox, “Modern day monitoring and control challenges outlined on an industrial-scale benchmark fermentation process,”Com- puters & Chemical Engineering, vol. 130, p. 106471, 2019

    work page 2019

  40. [40]

    P. E. Kloeden and E. Platen,Numerical Solution of Stochastic Differen- tial Equations, ser. Applications of Mathematics: Stochastic Modelling and Applied Probability. Berlin, Heidelberg: Springer-Verlag, 1992, vol. 23

    work page 1992