pith. sign in

arxiv: 2605.27352 · v1 · pith:2QE6QNTDnew · submitted 2026-05-26 · 💻 cs.LG · stat.ML

From Scores to Gibbs Correctors: Accelerating Uniform-Rate Discrete Diffusion Models

Yuchen Liang , Ness Shroff , Yingbin Liang This is my paper

Pith reviewed 2026-06-29 18:53 UTC · model grok-4.3

classification 💻 cs.LG stat.ML
keywords discrete diffusionGibbs correctorsampling complexityuniform-rate modelspredictor-correctorscore estimationtext generationmusic generation
0
0 comments X

The pith

GADD constructs Gibbs posterior likelihoods directly from concrete scores to reach O(polylog(ε^{-1})) sampling complexity for uniform-rate discrete diffusion models.

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

The paper introduces Gibbs-Accelerated Discrete Diffusion (GADD), a predictor-corrector method that builds Gibbs posterior likelihoods from the structure of the concrete score function without any extra training beyond ordinary score estimation. This construction is used to accelerate sampling in uniform-rate discrete diffusion models, which previously needed many steps for generation in text and symbolic data. The authors prove that the overall sampling complexity is O(polylog(ε^{-1})), the first such bound for diffusion-based samplers in this setting. They support the result with an induction argument that tracks error propagation across predictor steps while incorporating corrector inaccuracies, and they validate the approach on synthetic data plus zero-shot tasks in text and conditional music generation.

Core claim

GADD leverages the structure of the concrete score function to construct Gibbs posterior likelihoods directly, without requiring any additional training beyond standard score estimation. The resulting predictor-corrector scheme achieves an overall sampling complexity of O(polylog(ε^{-1})), the first such rate for diffusion-based samplers on uniform-rate discrete diffusion models. The theoretical analysis relies on an induction argument that tracks error propagation across predictor iterations while accounting for inaccuracies in the corrector updates, providing a new framework for analyzing predictor-corrector methods that differs from Girsanov-based approaches.

What carries the argument

The Gibbs-based corrector, which constructs posterior likelihoods directly from the concrete score function and plugs into the predictor-corrector iteration to accelerate mixing.

If this is right

  • Uniform-rate discrete diffusion models can now be sampled in polylogarithmic steps rather than the linear or slower rates of prior diffusion-based methods.
  • The same score estimates already computed for training suffice for both the predictor and the new Gibbs corrector, eliminating the need for auxiliary models.
  • The induction-based error analysis supplies a template for bounding predictor-corrector schemes that avoids change-of-measure arguments.
  • Empirical gains appear in wall-clock time and sample quality on zero-shot text sampling and conditional music generation relative to Euler and CTMC baselines.

Where Pith is reading between the lines

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

  • The direct score-to-Gibbs construction may extend to other discrete diffusion schedules whose transition kernels admit similar posterior forms.
  • The induction framework for tracking predictor-corrector errors could be adapted to analyze acceleration in continuous-state diffusion models.
  • Because the method re-uses existing score estimates, it lowers the barrier to deploying accelerated discrete diffusion in new symbolic domains.

Load-bearing premise

The structure of the concrete score function permits direct construction of Gibbs posterior likelihoods without any additional training beyond standard score estimation, and the induction argument correctly tracks error propagation while accounting for corrector inaccuracies.

What would settle it

An explicit counterexample or numerical run in which the total number of steps required to reach accuracy ε scales worse than any polylog(ε^{-1}) while using only the score-based Gibbs corrector would falsify the complexity claim.

Figures

Figures reproduced from arXiv: 2605.27352 by Ness Shroff, Yingbin Liang, Yuchen Liang.

Figure 1
Figure 1. Figure 1: Comparison of GADD on synthetic data against the Euler method, the [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
read the original abstract

Discrete diffusion models have achieved strong empirical performance in text and other symbolic domains, but, especially for uniform-rate models, they often require many steps to generate a single sample. Existing acceleration methods either rely on training additional quantities or suffer from slow mixing. In this work, we propose a novel Gibbs-based corrector for discrete diffusion models, termed Gibbs-Accelerated Discrete Diffusion (GADD). GADD leverages the structure of the concrete score function to construct Gibbs posterior likelihoods directly, without requiring any additional training beyond standard score estimation. We show that GADD achieves an overall sampling complexity of $\mathcal{O}(\mathrm{polylog} (\varepsilon^{-1}))$, yielding the first such rate for diffusion-based samplers for uniform-rate discrete diffusion models. We also conduct numerical experiments demonstrating the practical advantages of GADD across synthetic data, zero-shot text sampling, and zero-shot conditional music generation. These results corroborate the theory and show that GADD consistently improves sample quality and wall-clock efficiency over standard baselines, including vanilla Euler methods and CTMC correctors. Beyond this, our theoretical analysis introduces a novel framework for analyzing predictor-corrector methods in discrete diffusion models, which may be of independent interest. Unlike existing approaches that rely on the Girsanov change-of-measure technique, our method is based on an induction argument that tracks error propagation across predictor iterations while accounting for inaccuracies in the corrector updates.

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 / 3 minor

Summary. The paper proposes Gibbs-Accelerated Discrete Diffusion (GADD), a predictor-corrector scheme for uniform-rate discrete diffusion models. It constructs Gibbs posterior likelihoods directly from the concrete score function without extra training, proves an overall sampling complexity of O(polylog(ε^{-1})) via an induction argument that tracks predictor error propagation while accounting for corrector inaccuracies, and reports empirical gains in sample quality and wall-clock time versus Euler and CTMC baselines on synthetic data, zero-shot text sampling, and conditional music generation. The analysis framework is presented as novel relative to Girsanov-based approaches.

Significance. If the claimed complexity bound is rigorously established, the result would be significant: it supplies the first polylogarithmic rate for diffusion-based sampling in the uniform-rate discrete setting and introduces an induction-based error analysis that may be reusable. The absence of additional training and the reported practical speed-ups are further strengths. The contribution is tempered by the need for explicit verification of the error-control constants in the discrete setting.

major comments (2)
  1. [§4] §4 (main complexity theorem): the induction argument claims to bound the accumulation of corrector inaccuracies across predictor steps, yet the manuscript does not exhibit the concrete contraction or Lipschitz constant of the score function nor the explicit dependence on state-space cardinality |X|; without these, it is unclear whether the per-step discrepancy remains small enough to preserve the polylog(ε^{-1}) scaling rather than degrading to poly(1/ε).
  2. [§4.3] §4.3 (error-propagation lemma): the claim that corrector error is controlled to O(ε) per step is used to close the induction, but the proof sketch does not show how the uniform-rate transition kernel interacts with the Gibbs posterior approximation; a concrete counter-example or numerical check on a small discrete space would strengthen the argument.
minor comments (3)
  1. Notation for the concrete score and the Gibbs posterior should be introduced with a single consistent symbol table; current usage mixes p_θ and the score notation across sections.
  2. Figure 3 (music generation) lacks axis labels on the wall-clock vs. quality plot and does not report the number of independent runs used for the error bars.
  3. The related-work discussion omits recent analyses of CTMC correctors that also target uniform-rate models; a brief comparison paragraph would clarify novelty.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback. We appreciate the acknowledgment of the potential significance of the polylogarithmic complexity result and the induction-based analysis. We address each major comment below, indicating where revisions will be made to improve clarity and rigor.

read point-by-point responses

  1. Referee: [§4] §4 (main complexity theorem): the induction argument claims to bound the accumulation of corrector inaccuracies across predictor steps, yet the manuscript does not exhibit the concrete contraction or Lipschitz constant of the score function nor the explicit dependence on state-space cardinality |X|; without these, it is unclear whether the per-step discrepancy remains small enough to preserve the polylog(ε^{-1}) scaling rather than degrading to poly(1/ε).

    Authors: We agree that the current presentation leaves the contraction constants implicit. In the revision we will explicitly derive the Lipschitz constant of the concrete score under the uniform-rate dynamics and state its (polynomial) dependence on |X|. The induction is closed by choosing the number of predictor steps T = O(polylog(ε^{-1})) and step size h = O(1/T) such that the per-step discrepancy factor remains bounded by 1 + O(h); the accumulated error therefore stays O(ε) rather than growing polynomially in 1/ε. The expanded §4 will contain these calculations. revision: yes

  2. Referee: [§4.3] §4.3 (error-propagation lemma): the claim that corrector error is controlled to O(ε) per step is used to close the induction, but the proof sketch does not show how the uniform-rate transition kernel interacts with the Gibbs posterior approximation; a concrete counter-example or numerical check on a small discrete space would strengthen the argument.

    Authors: The Gibbs corrector is obtained by exponentiating the score to recover the exact conditional distribution of the uniform-rate kernel; consequently the one-step corrector error is bounded by the score estimation error, which is O(ε) by assumption. To make this interaction fully explicit we will enlarge the proof sketch in §4.3 and add a small-scale numerical verification (binary alphabet, |X|=4) confirming that the per-step corrector discrepancy remains O(ε) across predictor iterations. These additions will be included in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained via induction on error propagation

full rationale

The paper derives its O(polylog(ε^{-1})) sampling complexity claim from a novel induction argument that tracks predictor-corrector error propagation in the uniform-rate discrete setting. This argument is presented as independent of Girsanov techniques and does not reduce to fitted parameters, self-citations, or definitional equivalences. The Gibbs corrector is constructed directly from the concrete score without additional training, and the abstract explicitly contrasts the approach with prior methods. No load-bearing step is shown to collapse by construction to its inputs; the central result therefore retains independent analytical content.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Abstract provides limited detail on underlying assumptions; the induction argument and score structure are invoked but not expanded.

axioms (1)
  • domain assumption Induction argument tracks error propagation across predictor iterations while accounting for inaccuracies in the corrector updates.
    Invoked in the theoretical analysis section of the abstract.
invented entities (1)
  • GADD no independent evidence
    purpose: Gibbs-based corrector for accelerating discrete diffusion sampling
    New method introduced to leverage concrete score for Gibbs posteriors.

pith-pipeline@v0.9.1-grok · 5784 in / 1188 out tokens · 37736 ms · 2026-06-29T18:53:08.587913+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

49 extracted references · 10 canonical work pages · 3 internal anchors

  1. [1]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. InProceedings of the 32nd International Conference on Machine Learning, volume 37, pages 2256–2265, 2015

    2015

  2. [2]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. InAdvances in Neural Information Processing Systems, volume 33, pages 6840–6851, 2020

    2020

  3. [3]

    Johnson, Jonathan Ho, Daniel Tarlow, and Rianne van den Berg

    Jacob Austin, Daniel D. Johnson, Jonathan Ho, Daniel Tarlow, and Rianne van den Berg. Structured denoising diffusion models in discrete state-spaces. In A. Beygelzimer, Y . Dauphin, P. Liang, and J. Wortman Vaughan, editors,Advances in Neural Information Processing Systems, 2021

    2021

  4. [4]

    Large language diffusion models

    Shen Nie, Fengqi Zhu, Zebin You, Xiaolu Zhang, Jingyang Ou, Jun Hu, Jun Zhou, Yankai Lin, Ji-Rong Wen, and Chongxuan Li. Large language diffusion models. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025. 11

    2025

  5. [5]

    Digress: Discrete denoising diffusion for graph generation

    Clement Vignac, Igor Krawczuk, Antoine Siraudin, Bohan Wang, V olkan Cevher, and Pascal Frossard. Digress: Discrete denoising diffusion for graph generation. InThe Eleventh International Conference on Learning Representations, 2023

    2023

  6. [6]

    Generative diffusion models on graphs: methods and applications

    Chengyi Liu, Wenqi Fan, Yunqing Liu, Jiatong Li, Hang Li, Hui Liu, Jiliang Tang, and Qing Li. Generative diffusion models on graphs: methods and applications. InProceedings of the Thirty-Second International Joint Conference on Artificial Intelligence, 2023

    2023

  7. [7]

    A continuous time framework for discrete denoising models

    Andrew Campbell, Joe Benton, Valentin De Bortoli, Tom Rainforth, George Deligiannidis, and Arnaud Doucet. A continuous time framework for discrete denoising models. InAdvances in Neural Information Processing Systems, 2022

    2022

  8. [8]

    Diffusion models in de novo drug design

    Amira Alakhdar, Barnabas Poczos, and Newell Washburn. Diffusion models in de novo drug design. Journal of Chemical Information and Modeling, 2024

    2024

  9. [9]

    Unlocking guidance for discrete state-space diffusion and flow models

    Hunter Nisonoff, Junhao Xiong, Stephan Allenspach, and Jennifer Listgarten. Unlocking guidance for discrete state-space diffusion and flow models. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  10. [10]

    Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole

    Yang Song, Jascha Sohl-Dickstein, Diederik P. Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. InThe Ninth International Conference on Learning Representations, 2021

    2021

  11. [11]

    Stochastic Runge-Kutta methods: Provable acceleration of diffusion models.arXiv preprint arXiv:2410.04760, 2024

    Yuhui Wu, Yuxin Chen, and Yuting Wei. Stochastic Runge-Kutta methods: Provable acceleration of diffusion models.arXiv preprint arXiv:2410.04760, 2024

    work page arXiv 2024

  12. [12]

    Broadening target distributions for accelerated diffusion models via a novel analysis approach

    Yuchen Liang, Peizhong Ju, Yingbin Liang, and Ness Shroff. Broadening target distributions for accelerated diffusion models via a novel analysis approach. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  13. [13]

    Faster diffusion models via higher-order approxima- tion.arXiv preprint arXiv:2506.24042, 2025

    Gen Li, Yuchen Zhou, Yuting Wei, and Yuxin Chen. Faster diffusion models via higher-order approxima- tion.arXiv preprint arXiv:2506.24042, 2025

    work page arXiv 2025

  14. [14]

    Masked diffusion models are secretly time-agnostic masked models and exploit inaccurate categorical sampling

    Kaiwen Zheng, Yongxin Chen, Hanzi Mao, Ming-Yu Liu, Jun Zhu, and Qinsheng Zhang. Masked diffusion models are secretly time-agnostic masked models and exploit inaccurate categorical sampling. InInternational Conference on Learning Representations (ICLR), 2025

    2025

  15. [15]

    Informed correctors for discrete diffusion models

    Yixiu Zhao, Jiaxin Shi, Feng Chen, Shaul Druckmann, Lester Mackey, and Scott Linderman. Informed correctors for discrete diffusion models. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

    2025

  16. [16]

    Learning flexible forward trajectories for masked molecular diffusion

    Hyunjin Seo, Taewon Kim, Sihyun Yu, and SungSoo Ahn. Learning flexible forward trajectories for masked molecular diffusion.arXiv preprint arXiv: 2505.16790, 2025

    work page arXiv 2025

  17. [17]

    Mg-diff: A novel molecular graph diffusion model for molecular generation and optimization.PLOS ONE, 2025

    Xiaochen Zhang, Shuangxi Wang, Ying Fang, and Qiankun Zhang. Mg-diff: A novel molecular graph diffusion model for molecular generation and optimization.PLOS ONE, 2025

    2025

  18. [18]

    The Diffusion Duality, Chapter II: $\Psi$-Samplers

    Justin Deschenaux, Caglar Gulcehre, and Subham Sekhar Sahoo. The diffusion duality, chapter ii: ψ-samplers and efficient curriculum.arXiv preprint arXiv:2602.21185, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  19. [19]

    Convergence of score-based discrete diffusion models: A discrete-time analysis

    Zikun Zhang, Zixiang Chen, and Quanquan Gu. Convergence of score-based discrete diffusion models: A discrete-time analysis. InThe Thirteenth International Conference on Learning Representations, 2025. 12

    2025

  20. [20]

    Discrete diffusion models: Novel analysis and new sampler guarantees

    Yuchen Liang, Yingbin Liang, Lifeng Lai, and Ness Shroff. Discrete diffusion models: Novel analysis and new sampler guarantees. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

    2025

  21. [21]

    Rotskoff, Molei Tao, and Lexing Ying

    Yinuo Ren, Haoxuan Chen, Yuchen Zhu, Wei Guo, Yongxin Chen, Grant M. Rotskoff, Molei Tao, and Lexing Ying. Fast solvers for discrete diffusion models: Theory and applications of high-order algorithms. InFrontiers in Probabilistic Inference: Learning meets Sampling, 2025

    2025

  22. [22]

    Rotskoff, and Lexing Ying

    Yinuo Ren, Haoxuan Chen, Grant M. Rotskoff, and Lexing Ying. How discrete and continuous diffusion meet: Comprehensive analysis of discrete diffusion models via a stochastic integral framework. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  23. [23]

    Efficient sampling with discrete diffusion models: Sharp and adaptive guarantees

    Daniil Dmitriev, Zhihan Huang, and Yuting Wei. Efficient sampling with discrete diffusion models: Sharp and adaptive guarantees.arXiv preprint arXiv:2602.15008, 2026

    work page internal anchor Pith review arXiv 2026

  24. [24]

    Non-asymptotic convergence of discrete diffusion models: Masked and random walk dynamics

    Giovanni Conforti, Alain Durmus, Le-Tuyet-Nhi Pham, and Gael Raoul. Non-asymptotic convergence of discrete diffusion models: Masked and random walk dynamics.arXiv preprint arXiv:2512.00580, 2025

    work page arXiv 2025

  25. [25]

    Discrete Markov probabilistic models: An improved discrete score-based framework with sharp con- vergence bounds under minimal assumptions

    Le-Tuyet-Nhi Pham, Dario Shariatian, Antonio Ocello, Giovanni Conforti, and Alain Oliviero Durmus. Discrete Markov probabilistic models: An improved discrete score-based framework with sharp con- vergence bounds under minimal assumptions. InForty-second International Conference on Machine Learning, 2025

    2025

  26. [26]

    Convergence analysis of discrete diffusion model: Exact implementation through uniformization.arXiv preprint arXiv:2402.08095, 2024

    Hongrui Chen and Lexing Ying. Convergence analysis of discrete diffusion model: Exact implementation through uniformization.arXiv preprint arXiv:2402.08095, 2024

    work page arXiv 2024

  27. [27]

    Dimension-Free Convergence of Discrete Diffusion Models: Adjoint Equations Induce the Right Space

    Kelvin Kan, Xingjian Li, Benjamin J. Zhang, Tuhin Sahai, Stanley Osher, and Markos A. Katsoulakis. Dimension-free convergence of discrete diffusion models: Adjoint equations induce the right space. arXiv preprint arXiv:2605.17232, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  28. [28]

    Itai Gat, Tal Remez, Neta Shaul, Felix Kreuk, Ricky T. Q. Chen, Gabriel Synnaeve, Yossi Adi, and Yaron Lipman. Discrete flow matching. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

    2024

  29. [29]

    Discrete diffusion modeling by estimating the ratios of the data distribution

    Aaron Lou, Chenlin Meng, and Stefano Ermon. Discrete diffusion modeling by estimating the ratios of the data distribution. InProceedings of the 41st International Conference on Machine Learning, volume 235 ofProceedings of Machine Learning Research, pages 32819–32848, 2024

    2024

  30. [30]

    Kelly.Reversibility and Stochastic Networks

    Frank P. Kelly.Reversibility and Stochastic Networks. Cambridge University Press, 2011

    2011

  31. [31]

    Levin, Yuval Peres, and Elizabeth L

    David A. Levin, Yuval Peres, and Elizabeth L. Wilmer.Markov Chains and Mixing Times (Second Edition). American Mathematical Society, 2017

    2017

  32. [32]

    Reverse transition kernel: A flexible framework to accelerate diffusion inference

    Xunpeng Huang, Difan Zou, Hanze Dong, Yi Zhang, Yian Ma, and Tong Zhang. Reverse transition kernel: A flexible framework to accelerate diffusion inference. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

    2024

  33. [33]

    Sharp convergence rates for masked diffusion models, 2026 b

    Yuchen Liang, Zhiheng Tan, Ness Shroff, and Yingbin Liang. Sharp convergence rates for masked diffusion models.arXiv preprint arXiv:2602.22505, 2026

    work page arXiv 2026

  34. [34]

    Musegan: Multi-track sequential generative adversarial networks for symbolic music generation and accompaniment

    Hao-Wen Dong, Wen-Yi Hsiao, Li-Chia Yang, and Yi-Hsuan Yang. Musegan: Multi-track sequential generative adversarial networks for symbolic music generation and accompaniment. InProceedings of the AAAI Conference on Artificial Intelligence, 2018. 13

    2018

  35. [35]

    Theory on score-mismatched diffusion models and zero-shot conditional samplers

    Yuchen Liang, Peizhong Ju, Yingbin Liang, and Ness Shroff. Theory on score-mismatched diffusion models and zero-shot conditional samplers. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  36. [36]

    Accelerating convergence of score-based diffusion models, provably

    Gen Li, Yu Huang, Timofey Efimov, Yuting Wei, Yuejie Chi, and Yuxin Chen. Accelerating convergence of score-based diffusion models, provably. InProceedings of the 41st International Conference on Machine Learning, 2024

    2024

  37. [37]

    Stochastic relaxation, gibbs distributions, and the bayesian restoration of images.IEEE Transactions on Pattern Analysis and Machine Intelligence, 1984

    Stuart Geman and Donald Geman. Stochastic relaxation, gibbs distributions, and the bayesian restoration of images.IEEE Transactions on Pattern Analysis and Machine Intelligence, 1984

    1984

  38. [38]

    A polynomial time MCMC method for sampling from continuous determinantal point processes

    Alireza Rezaei and Shayan Oveis Gharan. A polynomial time MCMC method for sampling from continuous determinantal point processes. InProceedings of the 36th International Conference on Machine Learning, 2019

    2019

  39. [39]

    Yves F. Atchadé. Approximate spectral gaps for markov chain mixing times in high dimensions.SIAM Journal on Mathematics of Data Science, 3(3):854–872, 2021

    2021

  40. [40]

    Dimension-free mixing times of Gibbs samplers for Bayesian hierarchical models.The Annals of Statistics, 52(3):869 – 894, 2024

    Filippo Ascolani and Giacomo Zanella. Dimension-free mixing times of Gibbs samplers for Bayesian hierarchical models.The Annals of Statistics, 52(3):869 – 894, 2024

    2024

  41. [41]

    Parallel gibbs sampling: From colored fields to thin junction trees

    Joseph Gonzalez, Yucheng Low, Arthur Gretton, and Carlos Guestrin. Parallel gibbs sampling: From colored fields to thin junction trees. InProceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, 2011

    2011

  42. [42]

    Discrete predictor- corrector diffusion models for image synthesis

    Jose Lezama, Tim Salimans, Lu Jiang, Huiwen Chang, Jonathan Ho, and Irfan Essa. Discrete predictor- corrector diffusion models for image synthesis. InThe Eleventh International Conference on Learning Representations, 2023

    2023

  43. [43]

    Language models are unsupervised multitask learners.OpenAI, 2019

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language models are unsupervised multitask learners.OpenAI, 2019

    2019

  44. [44]

    Split gibbs discrete diffusion posterior sampling.arXiv preprint arXiv:2503.01161, 2025

    Wenda Chu, Zihui Wu, Yifan Chen, Yang Song, and Yisong Yue. Split gibbs discrete diffusion posterior sampling.arXiv preprint arXiv:2503.01161, 2025

    work page arXiv 2025

  45. [45]

    Improved analysis of score-based generative modeling: user-friendly bounds under minimal smoothness assumptions

    Hongrui Chen, Holden Lee, and Jianfeng Lu. Improved analysis of score-based generative modeling: user-friendly bounds under minimal smoothness assumptions. InProceedings of the 40th International Conference on Machine Learning, 2023

    2023

  46. [46]

    Bobkov and Prasad Tetali

    Sergey G. Bobkov and Prasad Tetali. Modified logarithmic Sobolev inequalities in discrete settings. Journal of Theoretical Probability, 19(2):289–336, 2006. 14 Appendix A Additional Related Works 15 B Variants of GADD and Posterior Likelihood Estimate 16 C Details of Numerical Experiments 17 C.1 Synthetic Experiment . . . . . . . . . . . . . . . . . . . ....

    2006

  47. [47]

    the outer-loop (or grand) initialization error at the start of the algorithm

  48. [48]

    the inner-loop initialization error at the beginning of each Gibbs sampling phase; and

  49. [49]

    X xτ ˜pT−t k,τ|η kℓ(xτ |xηkℓ)gτ(xτ)−g τ(xηkℓ) # ≲E xηk ℓ∼˜pT−t k ,ηk ℓ

    the inner-loop error resulting from imperfect score estimation. We analyze each of these error components separately and then combine them to derive the overall convergence guarantee for Algorithm 1. To begin, the following result gives us an upper bound on the grand (i.e., outer-loop) initialization error. From [19, Proposition 2] and [22, Theorem C.1], ...