Pith. sign in

REVIEW 104 references

From stability of Langevin diffusion to convergence of proximal MCMC for non-log-concave sampling

T0 review · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read PSGLA is proven to converge for non-convex composite potentials, up to a step-size bias, via a new drift-stability bound for inexact ULA.

arxiv 2505.14177 v2 pith:DWMFZA4D submitted 2025-05-20 stat.ML cs.CVcs.LG

classification stat.MLcs.CVcs.LG
keywords algorithmlangevinpotentialsconvergencenon-convexpsglasamplingstability
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

Sampling from a distribution proportional to e^{-V} is hard when the potential V is non-convex. The Unadjusted Langevin Algorithm (ULA) performs noisy gradient descent on V. When V splits into a smooth part f and a nonsmooth part g, PSGLA replaces the gradient of g with a proximal step, which is often a denoising operation. For a long time, convergence guarantees for PSGLA existed only for convex or strongly convex potentials.

This paper removes that convexity restriction. It first proves that two inexact Langevin chains with similar drifts have stationary laws that are close, with no extra discretization error term. It then rewrites PSGLA as a standard inexact ULA on a shadow chain driven by the drift b_gamma, and uses Moreau envelope calculus to show this drift is regular and dissipative. The resulting Theorem 3 gives exponential convergence to the smoothed target mu_gamma proportional to e^{-f-g_gamma}, with a bias of order gamma^{1/(2p)}, and a separate proposition shows mu_gamma approaches the true target pi as gamma goes to zero.

Experiments on 2D Gaussian mixtures and image inpainting suggest PnP-PSGLA mixes faster than PnP-ULA and restores images competitively. However, the proof relies on technical conditions, especially strong convexity at infinity of the Moreau envelope, which the authors admit are hard to verify and which are not checked for the neural denoiser used in the experiments.

Extended reading notes

Core claim

Theorem 3: Under Assumptions 2-3, for all gamma below a threshold and all k, Wp(pYk, mu_gamma) <= C1 r^{k gamma} + C2 gamma^{1/(2p)} and Wp(pXk, nu_gamma) <= C3 r^{k gamma} + C4 gamma^{1/(2p)}, where mu_gamma is proportional to e^{-f-g_gamma} and nu_gamma is its pushforward by the proximal operator. If correct, PSGLA samples the smoothed target geometrically fast, with bias that vanishes as gamma goes to zero.

Load-bearing premise

Assumption 3(ii): the Moreau envelope g_gamma is mu-strongly convex at infinity with mu >= 8 Lf + 4 Lg uniformly for small gamma. This is the only mechanism that makes the shadow drift b_gamma weakly dissipative (Lemma 21) and hence the shadow chain geometrically ergodic; the authors call it technical and hard to verify in practice (Appendix B). If it fails, Theorem 3's exponential convergence bound is not established. The companion Assumption 3(i), requiring g to be smooth on the prox image, is also restrictive but the authors acknowledge it.

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.

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

No free parameters are fitted in the theory. The main assumptions are explicit; Assumption 3(ii) is the most restrictive and is introduced specifically to force weak dissipativity of the shadow drift. Assumption 3(i) is technical. The paper relies on standard stochastic calculus and on external ergodicity results.

assumptions (6)
  • domain assumption Assumption 1: drifts are L-Lipschitz and weakly dissipative at infinity with constants L, R, m.
    Imposed on iULA drifts; ensures geometric ergodicity and moment bounds used in Theorems 1 and 2.
  • domain assumption Assumption 2: f is Lf-smooth; g is rho-weakly convex with gamma rho < 1.
    Composite potential framework for PSGLA; standard for forward-backward methods.
  • domain assumption Assumption 3(i): g is Lg-smooth on Prox_gamma g(R^d).
    Ensures g_gamma is 2Lg-smooth via Lemma 15, making the shadow drift b_gamma Lipschitz; restrictive for nonsmooth regularizers.
  • ad hoc to paper Assumption 3(ii): g_gamma is mu-strongly convex at infinity with mu >= 8 Lf + 4 Lg.
    Imposed specifically so b_gamma satisfies weak dissipativity; the paper admits it is hard to verify and proposes a projected variant to avoid it.
  • standard math Girsanov theorem and strong-solution existence for Lipschitz SDEs.
    Used in Lemma 20 to convert drift differences into KL-type bounds; standard stochastic calculus.
  • standard math Geometric ergodicity and moment bounds for weakly dissipative ULA chains from [23].
    External results invoked in Theorem 1 and Theorem 2 proofs; not re-derived in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of From stability of Langevin diffusion to convergence of proximal MCMC for non-log-concave sampling." pith.science (2026). https://pith.science/paper/DWMFZA4D

@misc{pith2026250514177,
  author       = {Pith},
  title        = {Pith review of: From stability of Langevin diffusion to convergence of proximal MCMC for non-log-concave sampling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DWMFZA4D}},
  note         = {Machine review of arXiv:2505.14177}
}
read the original abstract

We consider the problem of sampling distributions stemming from non-convex potentials with Unadjusted Langevin Algorithm (ULA). We prove the stability of the discrete-time ULA to drift approximations under the assumption that the potential is strongly convex at infinity. In many context, e.g. imaging inverse problems, potentials are non-convex and non-smooth. Proximal Stochastic Gradient Langevin Algorithm (PSGLA) is a popular algorithm to handle such potentials. It combines the forward-backward optimization algorithm with a ULA step. Our main stability result combined with properties of the Moreau envelope allows us to derive the first proof of convergence of the PSGLA for non-convex potentials. We empirically validate our methodology on synthetic data and in the context of imaging inverse problems. In particular, we observe that PSGLA exhibits faster convergence rates than Stochastic Gradient Langevin Algorithm for posterior sampling while preserving its restoration properties.

Figures

Figures reproduced from arXiv: 2505.14177 by the authors.

Figure 1
Figure 1. Posterior Sampling with PnP-ULA (top row) and PSGLA (bottow row) with three different [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Qualitative result for image inpainting with 50% masked pixels and a noise level of [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Qualitative result for image inpainting with 50% masked pixels and a noise level of [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Restoration of PSGLA with Prox DRUNet for [PITH_FULL_IMAGE:figures/full_fig_p022_4.png]
Figure 5
Figure 5. Figure 5: PnP-ULA and PSGLA with DnCNN for 50% missing pixels and σ = 1/255 with various number of iterations N ∈ {104 , 105 , 106}. The standard deviation of the Markov Chains are shown for each number of iterations and the evolution of the PSNR for N ∈ [0, 106 ]. Note that the…
Figure 6
Figure 6. Figure 6: PSGLA with DnCNN for 50% missing pixels and σ = 1/255 with two number of iter￾ations N ∈ {104 , 105}. We run PSGLA with 100 random seeds in the algorithm randomness for N = 104 and 10 random seeds in the algorithm randomness for N = 105 on the same observation. We can …
Figure 7
Figure 7. Figure 7: Representation of the function f defined in Equation (32). int(dom(g)). Denoting as Γ ⊂ int(dom(g)) the subset on which g is differentiable, we thus have Leb(int(dom(g)) \ Γ) = 0. Moreover, for x ∈ Γ, g is differentiable at x, so by Corollary 2, we have x = Proxγg (x +…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

104 extracted references · 69 canonical work pages

  1. [1]

    Credibility intervals for the reproduction number of the Covid-19 pandemic using proximal Langevin samplers

    Patrice Abry, Gersende Fort, Barbara Pascal, and Nelly Pustelnik. Credibility intervals for the reproduction number of the Covid-19 pandemic using proximal Langevin samplers. In 31th European Signal Processing Conference (EUSIPCO), 2023

  2. [2]

    What regularized auto-encoders learn from the data- generating distribution

    Guillaume Alain and Yoshua Bengio. What regularized auto-encoders learn from the data- generating distribution. The Journal of Machine Learning Research, 15(1):3563–3593, 2014

  3. [3]

    Cor- rection to: convex analysis and monotone operator theory in Hilbert spaces

    Heinz H Bauschke, Patrick L Combettes, Heinz H Bauschke, and Patrick L Combettes. Cor- rection to: convex analysis and monotone operator theory in Hilbert spaces. Springer, 2017

  4. [4]

    Langevin Monte Carlo beyond Lipschitz gradient continuity

    Matej Benko, Iwona Chlebicka, Jorgen Endal, and Bła˙zej Miasojedow. Langevin Monte Carlo beyond Lipschitz gradient continuity. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 15541–15549, 2025

  5. [5]

    Langevin Monte Carlo and JKO splitting

    Espen Bernton. Langevin Monte Carlo and JKO splitting. In Conference on Learning Theory, pages 1777–1798. PMLR, 2018

  6. [6]

    Alternating proximal-gradient steps for (stochastic) nonconvex-concave minimax problems

    Radu Ioan Bo¸ t and Axel Böhm. Alternating proximal-gradient steps for (stochastic) nonconvex-concave minimax problems. SIAM Journal on Optimization , 33(3):1884–1913, 2023. 10

  7. [7]

    Sampling from a log- concave distribution with compact support with proximal Langevin Monte Carlo

    Nicolas Brosse, Alain Durmus, Éric Moulines, and Marcelo Pereyra. Sampling from a log- concave distribution with compact support with proximal Langevin Monte Carlo. In Confer- ence on Learning Theory, pages 319–342. PMLR, 2017

  8. [8]

    The tamed unadjusted Langevin algorithm

    Nicolas Brosse, Alain Durmus, Éric Moulines, and Sotirios Sabanis. The tamed unadjusted Langevin algorithm. Stochastic Processes and their Applications, 129(10):3638–3663, 2019

Show all 104 references
  1. [9]

    Finite-time analysis of projected Langevin Monte Carlo

    Sebastien Bubeck, Ronen Eldan, and Joseph Lehec. Finite-time analysis of projected Langevin Monte Carlo. Advances in Neural Information Processing Systems, 28, 2015

  2. [10]

    An overview of existing methods and recent advances in sequential monte carlo

    Olivier Cappé, Simon J Godsill, and Eric Moulines. An overview of existing methods and recent advances in sequential monte carlo. Proceedings of the IEEE, 95(5):899–924, 2007

  3. [11]

    Sampling is as easy as learning the score: theory for diffusion models with minimal data assumptions

    Sitan Chen, Sinho Chewi, Jerry Li, Yuanzhi Li, Adil Salim, and Anru R Zhang. Sampling is as easy as learning the score: theory for diffusion models with minimal data assumptions. arXiv preprint arXiv:2209.11215, 2022

  4. [12]

    Convergence of Langevin MCMC in KL-divergence

    Xiang Cheng and Peter Bartlett. Convergence of Langevin MCMC in KL-divergence. In Algorithmic learning theory, pages 186–211. PMLR, 2018

  5. [13]

    Analysis of Langevin Monte Carlo from Poincare to Log-Sobolev

    Sinho Chewi, Murat A Erdogdu, Mufan Li, Ruoqi Shen, and Shunshi Zhang. Analysis of Langevin Monte Carlo from Poincare to Log-Sobolev. In Po-Ling Loh and Maxim Raginsky, editors, Proceedings of Thirty Fifth Conference on Learning Theory, volume 178 of Proceed- ings of Machine L...

  6. [14]

    Score-based diffusion models for accelerated mri

    Hyungjin Chung and Jong Chul Ye. Score-based diffusion models for accelerated mri. Med- ical image analysis, 80:102479, 2022

  7. [15]

    Diffusion posterior sampling for general noisy inverse problems

    Hyungjin Chung, Jeongsol Kim, Michael Thompson Mccann, Marc Louis Klasky, and Jong Chul Ye. Diffusion posterior sampling for general noisy inverse problems. In The Eleventh International Conference on Learning Representations, 2022

  8. [16]

    Nonsmooth analysis and optimization

    Frank H Clarke. Nonsmooth analysis and optimization. In Proceedings of the international congress of mathematicians, volume 5, pages 847–853. Citeseer, 1983

  9. [17]

    Plug-and-play split gibbs sam- pler: embedding deep generative priors in bayesian inference

    Florentin Coeurdoux, Nicolas Dobigeon, and Pierre Chainais. Plug-and-play split gibbs sam- pler: embedding deep generative priors in bayesian inference. IEEE Transactions on Image Processing, 2024

  10. [18]

    User’s guide to viscosity solutions of second order partial differential equations

    Michael G Crandall, Hitoshi Ishii, and Pierre-Louis Lions. User’s guide to viscosity solutions of second order partial differential equations. Bulletin of the American mathematical society, 27(1):1–67, 1992

  11. [19]

    Optimal scaling results for Moreau-Yosida Metropolis-adjusted Langevin algorithms

    Francesca R Crucinio, Alain Durmus, Pablo Jiménez, and Gareth O Roberts. Optimal scaling results for Moreau-Yosida Metropolis-adjusted Langevin algorithms. Bernoulli, 31(3):1889– 1907, 2025

  12. [20]

    User-friendly guarantees for the langevin monte carlo with inaccurate gradient

    Arnak Dalalyan and Avetik Karagulyan. User-friendly guarantees for the langevin monte carlo with inaccurate gradient. Stochastic Processes and their Applications , 129(12):5278– 5311, 2019

  13. [21]

    Stochastic model-based minimization of weakly convex functions

    Damek Davis and Dmitriy Drusvyatskiy. Stochastic model-based minimization of weakly convex functions. SIAM Journal on Optimization, 29(1):207–239, 2019

  14. [22]

    Convergence of denoising diffusion models under the manifold hypoth- esis

    Valentin De Bortoli. Convergence of denoising diffusion models under the manifold hypoth- esis. Transactions on Machine Learning Research, 2022

  15. [23]

    Convergence of diffusions and their discretizations: from continuous to discrete processes and back, 2019

    Valentin De Bortoli and Alain Durmus. Convergence of diffusions and their discretizations: from continuous to discrete processes and back, 2019. URL https://arxiv.org/abs/ 1904.09808

  16. [24]

    Ergodic bsdes under weak dissipative assumptions

    Arnaud Debussche, Ying Hu, and Gianmario Tessitore. Ergodic bsdes under weak dissipative assumptions. Stochastic Processes and their applications, 121(3):407–426, 2011. 11

  17. [25]

    Mulog, or how to apply gaussian denoisers to multi-channel sar speckle reduction? IEEE Transactions on Image Processing, 26(9):4389–4403, 2017

    Charles-Alban Deledalle, Loic Denis, Sonia Tabti, and Florence Tupin. Mulog, or how to apply gaussian denoisers to multi-channel sar speckle reduction? IEEE Transactions on Image Processing, 26(9):4389–4403, 2017

  18. [26]

    Markov chains: Basic definitions

    Randal Douc, Eric Moulines, Pierre Priouret, Philippe Soulier, Randal Douc, Eric Moulines, Pierre Priouret, and Philippe Soulier. Markov chains: Basic definitions. Springer, 2018

  19. [27]

    Nonasymptotic convergence analysis for the unadjusted Langevin algorithm

    Alain Durmus and Eric Moulines. Nonasymptotic convergence analysis for the unadjusted Langevin algorithm. The Annals of Applied Probability, 2017

  20. [28]

    Efficient bayesian computation by prox- imal markov chain monte carlo: when langevin meets moreau

    Alain Durmus, Eric Moulines, and Marcelo Pereyra. Efficient bayesian computation by prox- imal markov chain monte carlo: when langevin meets moreau. SIAM Journal on Imaging Sciences, 11(1):473–506, 2018

  21. [29]

    Analysis of Langevin Monte Carlo via convex optimization

    Alain Durmus, Szymon Majewski, and Bła ˙zej Miasojedow. Analysis of Langevin Monte Carlo via convex optimization. Journal of Machine Learning Research, 20(73):1–46, 2019

  22. [30]

    Reflection couplings and contraction rates for diffusions

    Andreas Eberle. Reflection couplings and contraction rates for diffusions. Probability theory and related fields, 166:851–886, 2016

  23. [31]

    On the kantorovich–rubinstein theorem

    David A Edwards. On the kantorovich–rubinstein theorem. Expositiones Mathematicae, 29 (4):387–398, 2011

  24. [32]

    Tweedie’s formula and selection bias

    Bradley Efron. Tweedie’s formula and selection bias. Journal of the American Statistical Association, 106(496):1602–1614, 2011

  25. [33]

    Proximal Langevin sam- pling with inexact proximal mapping

    Matthias J Ehrhardt, Lorenz Kuger, and Carola-Bibiane Schönlieb. Proximal Langevin sam- pling with inexact proximal mapping. SIAM Journal on Imaging Sciences, 17(3):1729–1760, 2024

  26. [34]

    Proximal interacting particle langevin algorithms, 2024

    Paula Cordero Encinar, Francesca R Crucinio, and O Deniz Akyildiz. Proximal interacting particle langevin algorithms, 2024. URL https://arxiv.org/abs/2406.14292

  27. [35]

    Pot: Python optimal transport

    Rémi Flamary, Nicolas Courty, Alexandre Gramfort, Mokhtar Z Alaya, Aurélie Boisbunon, Stanislas Chambon, Laetitia Chapel, Adrien Corenflos, Kilian Fatras, Nemo Fournier, et al. Pot: Python optimal transport. Journal of Machine Learning Research, 22(78):1–8, 2021

  28. [36]

    Covid19 reproduction number: Credibility intervals by blockwise proximal monte carlo samplers

    Gersende Fort, Barbara Pascal, Patrice Abry, and Nelly Pustelnik. Covid19 reproduction number: Credibility intervals by blockwise proximal monte carlo samplers. IEEE Transac- tions on Signal Processing, 71:888–900, 2023

  29. [37]

    Convexity in ReLU Neural Networks: beyond ICNNs?, 2025

    Anne Gagneux, Mathurin Massias, Emmanuel Soubies, and Rémi Gribonval. Convexity in ReLU Neural Networks: beyond ICNNs?, 2025. URL https://arxiv.org/abs/2501. 03017

  30. [38]

    Stochastic relaxation, gibbs distributions, and the bayesian restoration of images

    Stuart Geman and Donald Geman. Stochastic relaxation, gibbs distributions, and the bayesian restoration of images. IEEE Transactions on pattern analysis and machine intelligence, 1984

  31. [39]

    Adaptive rejection metropolis sampling within gibbs sampling

    Wally R Gilks, Nicky G Best, and Keith KC Tan. Adaptive rejection metropolis sampling within gibbs sampling. Journal of the Royal Statistical Society Series C: Applied Statistics , 44(4):455–472, 1995

  32. [40]

    Learning weakly convex regulariz- ers for convergent image-reconstruction algorithms

    Alexis Goujon, Sebastian Neumayer, and Michael Unser. Learning weakly convex regulariz- ers for convergent image-reconstruction algorithms. SIAM Journal on Imaging Sciences, 17 (1):91–115, 2024

  33. [41]

    A characterization of proximity operators

    Rémi Gribonval and Mila Nikolova. A characterization of proximity operators. Journal of Mathematical Imaging and Vision, 62(6):773–789, 2020

  34. [42]

    Agem: Solving linear inverse problems via deep priors and sampling

    Bichuan Guo, Yuxing Han, and Jiangtao Wen. Agem: Solving linear inverse problems via deep priors and sampling. Advances in Neural Information Processing Systems, 32, 2019

  35. [43]

    Provable benefit of annealed Langevin Monte Carlo for non-log-concave sampling

    Wei Guo, Molei Tao, and Yongxin Chen. Provable benefit of annealed Langevin Monte Carlo for non-log-concave sampling. International Conference of Learning Representation, 2025. 12

  36. [44]

    Penalized overdamped and under- damped langevin monte carlo algorithms for constrained sampling.Journal of machine learn- ing research, 25(263):1–67, 2024

    Mert Gurbuzbalaban, Yuanhan Hu, and Lingjiong Zhu. Penalized overdamped and under- damped langevin monte carlo algorithms for constrained sampling.Journal of machine learn- ing research, 25(263):1–67, 2024

  37. [45]

    Convex analysis and minimization al- gorithms I: Fundamentals, volume 305

    Jean-Baptiste Hiriart-Urruty and Claude Lemaréchal. Convex analysis and minimization al- gorithms I: Fundamentals, volume 305. Springer science & business media, 1996

  38. [46]

    On proximal point-type algorithms for weakly convex functions and their connection to the backward euler method

    Tim Hoheisel, Maxime Laborde, and Adam Oberman. On proximal point-type algorithms for weakly convex functions and their connection to the backward euler method. Optimization Online, 2010

  39. [47]

    Gradient step denoiser for con- vergent plug-and-play

    Samuel Hurault, Arthur Leclaire, and Nicolas Papadakis. Gradient step denoiser for con- vergent plug-and-play. In International Conference on Learning Representations (ICLR’22), 2022

  40. [48]

    Convergent plug-and-play with proximal denoiser and unconstrained regularization parameter

    Samuel Hurault, Antonin Chambolle, Arthur Leclaire, and Nicolas Papadakis. Convergent plug-and-play with proximal denoiser and unconstrained regularization parameter. Journal of Mathematical Imaging and Vision, pages 1–23, 2024

  41. [49]

    The performance of the unadjusted langevin algorithm without smoothness assumptions

    Tim Johnston, Iosif Lytras, Nikolaos Makras, and Sotirios Sabanis. The performance of the unadjusted langevin algorithm without smoothness assumptions. arXiv preprint arXiv:2502.03458, 2025

  42. [50]

    Local convergence of an inexact proximal algorithm for weakly convex functions, 2023

    Valdines Leite de Sousa Junior, Lucas Vidal de Meireles, Samara Costa Lima, and Gilson do Nascimento Silva. Local convergence of an inexact proximal algorithm for weakly convex functions, 2023. URL https://arxiv.org/abs/2308.02014

  43. [51]

    Stochastic solutions for linear inverse problems using the prior implicit in a denoiser

    Zahra Kadkhodaie and Eero Simoncelli. Stochastic solutions for linear inverse problems using the prior implicit in a denoiser. Advances in Neural Information Processing Systems , 34:13242–13254, 2021

  44. [52]

    On a space of totally additive functions

    Leonid Vasilevich Kantorovich and SG Rubinshtein. On a space of totally additive functions. Vestnik of the St. Petersburg University: Mathematics, 13(7):52–59, 1958

  45. [53]

    Brownian motion and stochastic calculus, volume 113

    Ioannis Karatzas and Steven Shreve. Brownian motion and stochastic calculus, volume 113. Springer Science & Business Media, 1991

  46. [54]

    Denoising diffusion restora- tion models

    Bahjat Kawar, Michael Elad, Stefano Ermon, and Jiaming Song. Denoising diffusion restora- tion models. Advances in Neural Information Processing Systems, 35:23593–23606, 2022

  47. [55]

    Accelerated Bayesian imaging by relaxed proximal-point Langevin sampling

    Teresa Klatzer, Paul Dobson, Yoann Altmann, Marcelo Pereyra, Jesús María Sanz-Serna, and Konstantinos C Zygalakis. Accelerated Bayesian imaging by relaxed proximal-point Langevin sampling. SIAM Journal on Imaging Sciences, 17(2):1078–1117, 2024

  48. [56]

    Projected stochastic gradient Langevin algorithms for constrained sam- pling and non-convex learning

    Andrew Lamperski. Projected stochastic gradient Langevin algorithms for constrained sam- pling and non-convex learning. InConference on Learning Theory, pages 2891–2937. PMLR, 2021

  49. [57]

    A note on the measurability of convex sets

    Robert Lang. A note on the measurability of convex sets. Archiv der Mathematik, 47:90–92, 1986

  50. [58]

    On maximum a posteriori estimation with plug & play priors and stochastic gradient descent

    Rémi Laumont, Valentin De Bortoli, Andrés Almansa, Julie Delon, Alain Durmus, and Marcelo Pereyra. On maximum a posteriori estimation with plug & play priors and stochastic gradient descent. Journal of Mathematical Imaging and Vision, 65(1):140–163, 2023

  51. [59]

    Bayesian imaging using Plug & Play Priors: When Langevin meets Tweedie

    Rémi Laumont, Valentin De Bortoli, Andrés Almansa, Julie Delon, Alain Durmus, and Marcelo Pereyra. Bayesian imaging using Plug & Play Priors: When Langevin meets Tweedie. SIAM Journal on Imaging Sciences, 15(2):701–737, 2022

  52. [60]

    A proximal algorithm for sampling from non-smooth potentials

    Jiaming Liang and Yongxin Chen. A proximal algorithm for sampling from non-smooth potentials. In 2022 Winter Simulation Conference (WSC), pages 3229–3240, 2022

  53. [61]

    Statistics of random processes: I

    Robert Liptser, Albert Nikolaevich Shiriaev, and Albert N Shiryaev. Statistics of random processes: I. General theory, volume 1. Springer Science & Business Media, 2001. 13

  54. [62]

    Ergodic bsdes and related pdes with neumann boundary conditions under weak dissipative assumptions

    Pierre-Yves Madec. Ergodic bsdes and related pdes with neumann boundary conditions under weak dissipative assumptions. Stochastic Processes and their Applications , 125(5):1821– 1860, 2015

  55. [63]

    Stochastic differential equations and applications

    Xuerong Mao. Stochastic differential equations and applications. Elsevier, 2007

  56. [64]

    A database of human seg- mented natural images and its application to evaluating segmentation algorithms and measur- ing ecological statistics

    David Martin, Charless Fowlkes, Doron Tal, and Jitendra Malik. A database of human seg- mented natural images and its application to evaluating segmentation algorithms and measur- ing ecological statistics. In Proceedings eighth IEEE international conference on computer vision...

  57. [65]

    Proximité et dualité dans un espace hilbertien

    Jean-Jacques Moreau. Proximité et dualité dans un espace hilbertien. Bulletin de la Société mathématique de France, 93:273–299, 1965

  58. [66]

    Inf-convolution, sous-additivité, convexité des fonctions numériques

    Jean Jacques Moreau. Inf-convolution, sous-additivité, convexité des fonctions numériques. Journal de Mathématiques Pures et Appliquées, 1970

  59. [67]

    Improved bounds for discretization of Langevin diffusions: Near-optimal rates without convexity

    Wenlong Mou, Nicolas Flammarion, Martin J Wainwright, and Peter L Bartlett. Improved bounds for discretization of Langevin diffusions: Near-optimal rates without convexity. Bernoulli, 28(3):1577–1601, 2022

  60. [68]

    Unadjusted langevin algorithm for non-convex weakly smooth potentials, 2021

    Dao Nguyen, Xin Dang, and Yixin Chen. Unadjusted langevin algorithm for non-convex weakly smooth potentials, 2021. URL https://arxiv.org/abs/2101.06369

  61. [69]

    Second-order analysis of the moreau–yosida and the lasry–lions regulariza- tions

    Nina Ovcharova. Second-order analysis of the moreau–yosida and the lasry–lions regulariza- tions. Optimization Methods & Software, 25(1):109–116, 2010

  62. [70]

    Proximal markov chain monte carlo algorithms

    Marcelo Pereyra. Proximal markov chain monte carlo algorithms. Statistics and Computing, 26:745–760, 2016

  63. [71]

    Learning max- imally monotone operators for image recovery

    Jean-Christophe Pesquet, Audrey Repetti, Matthieu Terris, and Yves Wiaux. Learning max- imally monotone operators for image recovery. SIAM Journal on Imaging Sciences , 14(3): 1206–1237, 2021

  64. [72]

    Optimal scaling for the proximal Langevin algorithm in high dimensions

    Natesh S Pillai. Optimal scaling for the proximal Langevin algorithm in high dimensions. Journal of Machine Learning Research, 25(404):1–32, 2024

  65. [73]

    A survey on sampling and probe methods for inverse problems

    Roland Potthast. A survey on sampling and probe methods for inverse problems. Inverse Problems, 22(2):R1, 2006

  66. [74]

    On the stability of the stochastic gradient Langevin algo- rithm with dependent data stream

    Miklós Rásonyi and Kinga Tikosi. On the stability of the stochastic gradient Langevin algo- rithm with dependent data stream. Statistics & Probability Letters, 182:109321, 2022

  67. [75]

    Plug-and-play posterior sampling under mismatched measurement and prior models

    Marien Renaud, Jiaming Liu, Valentin De Bortoli, Andrés Almansa, and Ulugbek S Kamilov. Plug-and-play posterior sampling under mismatched measurement and prior models. The Twelfth International Conference on Learning Representations, 2023

  68. [76]

    Plug-and-play image restoration with Stochastic deNOising REgularization

    Marien Renaud, Jean Prost, Arthur Leclaire, and Nicolas Papadakis. Plug-and-play image restoration with Stochastic deNOising REgularization. InForty-first International Conference on Machine Learning, 2024

  69. [77]

    Convergence analysis of a proximal stochastic denoising regularization algorithm

    Marien Renaud, Julien Hermant, and Nicolas Papadakis. Convergence analysis of a proximal stochastic denoising regularization algorithm. In International Conference on Scale Space and Variational Methods in Computer Vision, pages 17–29. Springer, 2025

  70. [78]

    Langevin diffusions and Metropolis-Hastings algo- rithms

    Gareth O Roberts and Osnat Stramer. Langevin diffusions and Metropolis-Hastings algo- rithms. Methodology and computing in applied probability, 4:337–357, 2002

  71. [79]

    Exponential convergence of Langevin distributions and their discrete approximations

    Gareth O Roberts and Richard L Tweedie. Exponential convergence of Langevin distributions and their discrete approximations. Bernoulli, pages 341–363, 1996

  72. [80]

    Variational analysis, volume 317

    R Tyrrell Rockafellar and Roger J-B Wets. Variational analysis, volume 317. Springer Sci- ence & Business Media, 2009. 14

  73. [81]

    The little engine that could: Regular- ization by denoising (red)

    Yaniv Romano, Michael Elad, and Peyman Milanfar. The little engine that could: Regular- ization by denoising (red). SIAM Journal on Imaging Sciences, 10(4):1804–1844, 2017

  74. [82]

    Solving linear inverse problems provably via posterior sampling with latent dif- fusion models

    Litu Rout, Negin Raoof, Giannis Daras, Constantine Caramanis, Alex Dimakis, and Sanjay Shakkottai. Solving linear inverse problems provably via posterior sampling with latent dif- fusion models. Advances in Neural Information Processing Systems, 36:49960–49990, 2023

  75. [83]

    Nonlinear total variation based noise removal algorithms

    Leonid I Rudin, Stanley Osher, and Emad Fatemi. Nonlinear total variation based noise removal algorithms. Physica D: nonlinear phenomena, 60(1-4):259–268, 1992

  76. [84]

    Primal dual interpretation of the proximal stochastic gradient langevin algorithm

    Adil Salim and Peter Richtarik. Primal dual interpretation of the proximal stochastic gradient langevin algorithm. Advances in Neural Information Processing Systems , 33:3786–3796, 2020

  77. [85]

    Stochastic proximal Langevin algorithm: Potential splitting and nonasymptotic rates

    Adil Salim, Dmitry Kovalev, and Peter Richtárik. Stochastic proximal Langevin algorithm: Potential splitting and nonasymptotic rates. Advances in Neural Information Processing Sys- tems, 32, 2019

  78. [86]

    Weakly convex regularisers for inverse problems: Convergence of critical points and primal- dual optimisation

    Zakhar Shumaylov, Jeremy Budd, Subhadip Mukherjee, and Carola-Bibiane Schönlieb. Weakly convex regularisers for inverse problems: Convergence of critical points and primal- dual optimisation. Forty-first International Conference on Machine Learning, 2024

  79. [87]

    Generative modeling by estimating gradients of the data distribution

    Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution. Advances in neural information processing systems, 32, 2019

  80. [88]

    Bayesian inference in thresh- old models using gibbs sampling

    D Andersen Sorensen, S Andersen, D Gianola, and I Korsgaard. Bayesian inference in thresh- old models using gibbs sampling. Genetics Selection Evolution, 27(3):229–249, 1995

  81. [89]

    Provable prob- abilistic imaging using score-based generative priors

    Yu Sun, Zihui Wu, Yifan Chen, Berthy T Feng, and Katherine L Bouman. Provable prob- abilistic imaging using score-based generative priors. IEEE Transactions on Computational Imaging, 2024

  82. [90]

    Deepinverse: A deep learning frame- work for inverse problems in imaging

    J Tachella, D Chen, S Hurault, M Terris, and A Wang. Deepinverse: A deep learning frame- work for inverse problems in imaging. url: https://deepinv. github. io/deepinv, 2023

  83. [91]

    Pareto smoothed importance sampling

    Aki Vehtari, Daniel Simpson, Andrew Gelman, Yuling Yao, and Jonah Gabry. Pareto smoothed importance sampling. Journal of Machine Learning Research, 25(72):1–58, 2024

  84. [92]

    Plug-and-play priors for model based reconstruction

    Singanallur V Venkatakrishnan, Charles A Bouman, and Brendt Wohlberg. Plug-and-play priors for model based reconstruction. In 2013 IEEE global conference on signal and infor- mation processing, pages 945–948. IEEE, 2013

  85. [93]

    Optimal transport: old and new, volume 338

    Cédric Villani et al. Optimal transport: old and new, volume 338. Springer, 2008

  86. [94]

    Split-and-augmented gibbs sam- pler—application to large-scale inference problems

    Maxime V ono, Nicolas Dobigeon, and Pierre Chainais. Split-and-augmented gibbs sam- pler—application to large-scale inference problems. IEEE Transactions on Signal Process- ing, 67(6):1648–1661, 2019

  87. [95]

    Image quality assess- ment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assess- ment: from error visibility to structural similarity. IEEE transactions on image processing , 13(4):600–612, 2004

  88. [96]

    Learning pseudo-contractive denoisers for inverse problems

    Deliang Wei, Peng Chen, and Fang Li. Learning pseudo-contractive denoisers for inverse problems. In Forty-first International Conference on Machine Learning, 2024

  89. [97]

    Sampling as optimization in the space of measures: The Langevin dynamics as a composite optimization problem

    Andre Wibisono. Sampling as optimization in the space of measures: The Langevin dynamics as a composite optimization problem. In Conference on Learning Theory, pages 2093–3027. PMLR, 2018

  90. [98]

    Convergence of the inexact Langevin algorithm and score-based generative models in KL divergence

    Kaylee Yingxi Yang and Andre Wibisono. Convergence of the inexact Langevin algorithm and score-based generative models in KL divergence. NeurIPS 2022 Workshop on Score- Based Methods, 2022. 15

  91. [99]

    Plug- and-play image restoration with deep denoiser prior

    Kai Zhang, Yawei Li, Wangmeng Zuo, Lei Zhang, Luc Van Gool, and Radu Timofte. Plug- and-play image restoration with deep denoiser prior. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(10):6360–6376, 2021

  92. [100]

    The unrea- sonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unrea- sonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 586–595, 2018

  93. [101]

    Denoising diffusion models for plug-and-play image restoration

    Yuanzhi Zhu, Kai Zhang, Jingyun Liang, Jiezhang Cao, Bihan Wen, Radu Timofte, and Luc Van Gool. Denoising diffusion models for plug-and-play image restoration. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 1219–1229, 2023. Appendix...

  94. [102]

    1 2 Z T 0 ⟨b1((wu)u∈[0,T ], t), dwt⟩ −1 4 Z T 0 ∥b1((wu)u∈[0,T ], t)∥2dt # dµW dµ2 ((wu)u∈[0,T ], t) = exp

    So we get that rγ ≤ R0 = max 8(R + 1) + 4∥∇g(0)∥ Lg , 2 s ginf − g(0) Lg + (R + 1)2 ! . Therefore, we get from (36) that ∀γ ∈ [0, 1 2Lg ] and ∀x ∈ Rd \ B(0, R0), g(x) ≤ g(0) + ⟨∇g(0), x⟩ + Lg 2 ∥x∥2 ≤ 1 2γ (∥x∥ −R − 1)2 + ginf . 37 By combining the previous inequality and Equa...

  95. [103]

    As a consequence, ∀t ∈ (t0, t1), u(t) < R2 0 and ∀t ∈ [0, 1] \ [t0, t1], u(t) > R2

  96. [104]

    By definition z0, z1 ∈ B(0, R0), therefore ∥z0 − z1∥ ≤2R0

    We denote z0 = x + t0(y − x) z1 = x + t1(y − x). By definition z0, z1 ∈ B(0, R0), therefore ∥z0 − z1∥ ≤2R0. Then t1 − t0 = ∥z1 − x∥ ∥y − x∥ − ∥z0 − x∥ ∥y − x∥ = ∥z1 − z0∥ ∥y − x∥ ≤ 2R0 ∥y − x∥ ≤ 1 2 , (83) because ∥y − x∥ ≥4R0. We can now estimate the quantity of interest ⟨∇gγ...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.