Pith. sign in

REVIEW 2 major objections 2 minor 1 cited by

W-Flow trains a neural generator to compress a Wasserstein gradient flow into one-step sampling from reference to target distribution.

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

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

T0 review · grok-4.3

2026-06-30 22:07 UTC pith:GDCQ5IT2

load-bearing objection W-Flow gets a competitive 1.29 FID on one-step ImageNet 256 but the convergence claim rests on unverified assumptions that may not cover the actual neural parameterization. the 2 major comments →

arxiv 2605.11755 v2 pith:GDCQ5IT2 submitted 2026-05-12 cs.LG cs.CVstat.ML

One-Step Generative Modeling via Wasserstein Gradient Flows

Jiaqi Han , Puheng Li , Qiushan Guo , Renyuan Xu , Stefano Ermon , Emmanuel J. Cand\`es This is my paper
classification cs.LG cs.CVstat.ML
keywords generative modelingWasserstein gradient flowsone-step generationSinkhorn divergenceoptimal transportImageNet generationdiffusion models
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper establishes a two-stage process for one-step generative modeling. First, an evolution is defined from a simple reference distribution to the target data distribution by following the Wasserstein gradient flow of the Sinkhorn divergence energy functional. Second, a static neural generator is trained to approximate the entire evolution in a single forward pass. This yields a generator that captures global distributional discrepancy rather than local updates. The approach is shown to reach 1.29 FID on one-step ImageNet 256x256 generation while providing roughly 100 times faster sampling than comparable multi-step diffusion models, along with better mode coverage.

Core claim

W-Flow defines an evolution from the reference distribution to the target via a Wasserstein gradient flow minimizing the Sinkhorn divergence, then trains a static neural generator to compress this evolution into one forward pass, achieving state-of-the-art one-step ImageNet 256x256 performance at 1.29 FID with improved mode coverage and domain transfer.

What carries the argument

The Wasserstein gradient flow of the Sinkhorn divergence energy functional, which supplies an optimal-transport-based update rule that is then approximated by the trained neural generator.

Load-bearing premise

Finite-sample training dynamics of the generator converge to the continuous-time distributional dynamics of the Wasserstein flow under suitable assumptions.

What would settle it

Running the learned one-step generator on held-out ImageNet data and finding that its output distribution deviates substantially from the distribution obtained by iterating the full continuous-time Wasserstein flow to the same number of steps.

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

If this is right

  • Achieves new state of the art for one-step ImageNet 256x256 generation at 1.29 FID
  • Delivers approximately 100x faster sampling than multi-step diffusion models with similar FID scores
  • Improves mode coverage and domain transfer relative to prior one-step methods
  • Finite-sample training converges to continuous-time dynamics under the stated assumptions

Where Pith is reading between the lines

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

  • The same compression of a gradient flow into a static generator could be applied to other energy functionals beyond Sinkhorn divergence
  • One-step models of this form may enable deployment in latency-sensitive settings where iterative sampling is impractical
  • The convergence proof suggests that scaling the number of training samples could further close the gap to the ideal flow

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper introduces W-Flow, a two-stage framework that first evolves a reference distribution to the target via a Wasserstein gradient flow minimizing a Sinkhorn-divergence energy functional, then trains a static neural generator to compress the entire evolution into a single forward pass. It states a convergence theorem for finite-sample training dynamics to the continuous-time flow under suitable assumptions, and reports an empirical result of 1.29 FID on one-step ImageNet 256×256 generation together with claims of improved mode coverage, domain transfer, and ~100× faster sampling than comparable multi-step diffusion models.

Significance. If the convergence result holds for the actual neural parameterization and the 1.29 FID is reproducible under controlled protocols, the work would supply a principled optimal-transport foundation for one-step generators that improves mode coverage relative to standard diffusion baselines. The explicit use of Sinkhorn divergence as the driving energy is a concrete technical choice that could be reused; however, the current manuscript provides neither the derivation details nor the assumption-verification steps needed to transfer the theory to the reported high-dimensional experiments.

major comments (2)
  1. [Abstract] Abstract (convergence statement): the claim that 'finite-sample training dynamics converge to the continuous-time distributional dynamics under suitable assumptions' is load-bearing for the assertion that the trained static generator faithfully realizes the flow's mode-coverage and distributional properties, yet the manuscript supplies neither the explicit assumptions (regularity, function-class, discretization, or neural-net restrictions) nor any verification that they hold for the ImageNet-scale generator used in the experiments.
  2. [Experiments] Empirical results (FID claim): the reported 1.29 FID on ImageNet 256×256 is presented without error bars, run-to-run variance, exact dataset protocol, or ablation of post-hoc choices (e.g., Sinkhorn regularization schedule, network architecture), making it impossible to assess whether the number robustly supports the SOTA and 100× speed-up claims relative to multi-step baselines.
minor comments (2)
  1. [Methods] Notation for the energy functional and the Sinkhorn divergence should be introduced with an explicit equation number in the methods section rather than only in prose.
  2. [Abstract] The abstract's comparison to 'multi-step diffusion models with similar FID scores' would benefit from a table listing the exact baseline models, their step counts, and FID values under identical evaluation settings.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript to provide the requested clarifications and additional details.

read point-by-point responses

  1. Referee: [Abstract] Abstract (convergence statement): the claim that 'finite-sample training dynamics converge to the continuous-time distributional dynamics under suitable assumptions' is load-bearing for the assertion that the trained static generator faithfully realizes the flow's mode-coverage and distributional properties, yet the manuscript supplies neither the explicit assumptions (regularity, function-class, discretization, or neural-net restrictions) nor any verification that they hold for the ImageNet-scale generator used in the experiments.

    Authors: We agree that the assumptions underlying the convergence result require explicit statement. In the revision we will add a dedicated paragraph in Section 3 listing the precise conditions (Lipschitz continuity of the Sinkhorn energy, bounded second moments, and the generator belonging to a sufficiently rich neural function class) together with a short proof sketch in the appendix. We will also note that the theorem is asymptotic and that the ImageNet-scale network is treated as a universal approximator; a brief discussion of how the practical discretization approximates the continuous flow will be included. These additions directly address the load-bearing nature of the claim. revision: yes

  2. Referee: [Experiments] Empirical results (FID claim): the reported 1.29 FID on ImageNet 256×256 is presented without error bars, run-to-run variance, exact dataset protocol, or ablation of post-hoc choices (e.g., Sinkhorn regularization schedule, network architecture), making it impossible to assess whether the number robustly supports the SOTA and 100× speed-up claims relative to multi-step baselines.

    Authors: We acknowledge that the current experimental reporting lacks the statistical and procedural details needed for rigorous evaluation. In the revised manuscript we will report mean FID and standard deviation over three independent training runs, specify the exact ImageNet 256×256 preprocessing pipeline and train/validation split, and add an ablation table varying the Sinkhorn regularization schedule and generator depth/width. These changes will allow readers to assess the robustness of the 1.29 FID and the claimed speed-up. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained against external benchmarks

full rationale

The paper defines the Wasserstein gradient flow independently via the Sinkhorn divergence energy functional (an established OT metric), then separately trains a neural generator to approximate the resulting trajectory. The convergence claim is stated as holding under suitable assumptions without reducing the target result to a fitted parameter or self-referential definition. No load-bearing step equates a prediction to its own input by construction, and no self-citation chain substitutes for an independent derivation. The empirical 1.29 FID result is presented as an outcome of this procedure rather than a renamed input.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the existence of a Wasserstein gradient flow that minimizes the Sinkhorn energy and on the unstated suitable assumptions required for the finite-sample convergence statement.

axioms (1)
  • domain assumption The finite-sample training dynamics converge to the continuous-time distributional dynamics under suitable assumptions.
    Explicitly invoked in the abstract as the basis for the convergence proof.

pith-pipeline@v0.9.1-grok · 5772 in / 1311 out tokens · 31778 ms · 2026-06-30T22:07:06.722277+00:00 · methodology

0 comments
read the original abstract

Diffusion models and flow-based methods have shown impressive generative capability, especially for images, but their sampling is expensive because it requires many iterative updates. We introduce W-Flow, a framework for training a generator that transforms samples from a simple reference distribution into samples from a target data distribution in a single step. This is achieved in two steps: we first define an evolution from the reference distribution to the target distribution through a Wasserstein gradient flow that minimizes an energy functional; second, we train a static neural generator to compress this evolution into one-step generation. We instantiate the energy functional with the Sinkhorn divergence, which yields an efficient optimal-transport-based update rule that captures global distributional discrepancy and improves coverage of the target distribution. We further prove that the finite-sample training dynamics converge to the continuous-time distributional dynamics under suitable assumptions. Empirically, W-Flow sets a new state of the art for one-step ImageNet 256$\times$256 generation, achieving 1.29 FID, with improved mode coverage and domain transfer. Compared to multi-step diffusion models with similar FID scores, our method yields approximately 100$\times$ faster sampling. These results show that Wasserstein gradient flows provide a principled and effective foundation for fast and high-fidelity generative modeling.

Figures

Figures reproduced from arXiv: 2605.11755 by Emmanuel J. Cand\`es, Jiaqi Han, Puheng Li, Qiushan Guo, Renyuan Xu, Stefano Ermon.

Figure 1
Figure 1. Figure 1: (Left) 1-NFE samples from W-Flow-L/2 trained from scratch on ImageNet-256×256. (Right) Sample quality (measured by FID) vs. effective sampling compute [39] (billion parameters × number of function evaluations during sampling) evaluated on ImageNet 256×256. target distribution in one step. This would combine the efficiency of one-step generation with the flexibility of a distributional evolution during trai… view at source ↗
Figure 2
Figure 2. Figure 2: (a) The conceptual diagram of W-Flow. (b) Visualization of the training dynamics projected onto the Sinkhorn divergence landscape on 8 Gaussian mixtures, shown on a logarithmic scale. ing a few/one-step generator from scratch, typically by enforcing certain self-consistency conditions on the trajectory [18, 19, 4, 55] or the intermediate marginals [70]. These methods largely inherit their training signal f… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison between one￾batch and two-batch estimators on learning a 2D Gaussian. where Π(qbt, pb) is the set of matrices with prescribed marginals. Denote the optimal solution π ε,∗ qbt,pb . Two-batch estimate for self-transport. Naïvely estimating the self-entropic OT term OTε(qbt, qbt) from a single empirical batch introduces a self-matching artifact: since each particle can be matched to itself at zero … view at source ↗
Figure 4
Figure 4. Figure 4: Classifier-free guidance. Left: The FID and Inception Score curve when sweeping over CFG scales. Right: Image samples by W-Flow, L/2 with CFG increasing from 0.0 to 2.0. 1-NFE sampling, W-Flow outperforms most diffusion models requiring up to 250 steps, such as LightningDiT-XL/2. These strong empirical results support our central claim that principled WGF dynamics can translate into exceptional generation … view at source ↗
Figure 5
Figure 5. Figure 5: (a) Oval-to-circle domain transfer. Source and target are constructed by sampling angles uniformly from [0, 2π) with parametric curves corrupted by Gaussian noise. (b) & (c) One-step facial age translation on FFHQ, mapping older faces to younger ones. (b) Histogram of the latent ℓ2 distance between 2,000 source images and their generated targets. (c) Visual comparison. (a) Drifting (b) W-Flow Drifting W-Fl… view at source ↗
Figure 6
Figure 6. Figure 6: Evaluation of mode coverage under imbalanced target distributions. (a) Evaluation of mode coverage on a 2D Gaussian mixtures dataset featuring six dominant modes and two distant minority modes. (b) PCA scatter plot of generated latent codes for an artificially imbalanced FFHQ target distribution (95% senior faces, 5% child faces). See Appendix F for generated samples showing the comparison of mode coverage… view at source ↗
Figure 7
Figure 7. Figure 7: Evaluation of self-transport estimators on a 2D Gaussian mixtures dataset featuring six [PITH_FULL_IMAGE:figures/full_fig_p026_7.png] view at source ↗
Figure 7
Figure 7. Figure 7: Evaluation of self-transport estimators on a 2D Gaussian mixture dataset featuring six [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of velocity guidance and distribution guidance for conditional generation on a [PITH_FULL_IMAGE:figures/full_fig_p029_8.png] view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of velocity guidance and distribution guidance for conditional generation on a [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Illustrations on the difference in the velocity field computation between Drifting Model [PITH_FULL_IMAGE:figures/full_fig_p029_9.png] view at source ↗
Figure 9
Figure 9. Figure 9: Illustrations of the difference in the velocity field computation between Drifting Model (Alg. [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Uncurated samples generated by W-Flow, L/2 with CFG [PITH_FULL_IMAGE:figures/full_fig_p033_10.png] view at source ↗
Figure 10
Figure 10. Figure 10: Uncurated samples generated by W-Flow, L/2 with CFG [PITH_FULL_IMAGE:figures/full_fig_p036_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Uncurated samples generated by W-Flow, XL/2 [PITH_FULL_IMAGE:figures/full_fig_p034_11.png] view at source ↗
Figure 11
Figure 11. Figure 11: Uncurated samples generated by W-Flow, XL/2 [PITH_FULL_IMAGE:figures/full_fig_p037_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Uncurated samples generated by W-Flow, XL/2 with CFG [PITH_FULL_IMAGE:figures/full_fig_p036_12.png] view at source ↗
Figure 12
Figure 12. Figure 12: Uncurated samples generated by W-Flow, XL/2 with CFG [PITH_FULL_IMAGE:figures/full_fig_p038_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Uncurated samples generated by Drifting Model in the mode coverage experiment [PITH_FULL_IMAGE:figures/full_fig_p037_13.png] view at source ↗
Figure 13
Figure 13. Figure 13: Uncurated samples generated by Drifting Model in the mode coverage experiment [PITH_FULL_IMAGE:figures/full_fig_p039_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Uncurated samples generated by W-Flow in the mode coverage experiment (Sec. [PITH_FULL_IMAGE:figures/full_fig_p038_14.png] view at source ↗
Figure 14
Figure 14. Figure 14: Uncurated samples generated by W-Flow in the mode coverage experiment (Sec. [PITH_FULL_IMAGE:figures/full_fig_p040_14.png] view at source ↗

Review history (2 revisions) →

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Condition-Wise Sinkhorn Drifting for One-Shot Learned Channel Simulation

    eess.SP 2026-06 unverdicted novelty 6.0

    Condition-wise Sinkhorn drifting is presented as a one-shot conditional channel simulator using a Sinkhorn objective trained via barycentric velocities and detached particle regression, outperforming other one-shot va...

Reference graph

Works this paper leans on

75 extracted references · 34 canonical work pages · cited by 1 Pith paper · 22 internal anchors

  1. [1]

    Building Normalizing Flows with Stochastic Interpolants

    Michael S Albergo and Eric Vanden-Eijnden. Building normalizing flows with stochastic interpolants.arXiv preprint arXiv:2209.15571, 2022. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  2. [2]

    LightSBB-M: Bridging Schr\"odinger and Bass for Generative Diffusion Modeling

    Alexandre Alouadi, Pierre Henry-Labordère, Grégoire Loeper, Othmane Mazhar, Huyên Pham, and Nizar Touzi. Lightsbb-m: Bridging Schrödinger and bass for generative diffusion modeling. arXiv preprint arXiv:2601.19312, 2026. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  3. [3]

    Birkhäuser, 2008

    Luigi Ambrosio, Nicola Gigli, and Giuseppe Savaré.Gradient Flows: In Metric Spaces and in the Space of Probability Measures. Birkhäuser, 2008. 25

    2008

  4. [4]

    Wasserstein generative adversarial networks

    Martin Arjovsky, Soumith Chintala, and Léon Bottou. Wasserstein generative adversarial networks. InInternational conference on machine learning, pages 214–223. PMLR, 2017. 2, 3

    2017

  5. [5]

    M., Albergo, M

    Nicholas M Boffi, Michael S Albergo, and Eric Vanden-Eijnden. How to build a consistency model: Learning flow maps via self-distillation.arXiv preprint arXiv:2505.18825, 2025. 3

    work page arXiv 2025

  6. [6]

    Large Scale GAN Training for High Fidelity Natural Image Synthesis

    Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale GAN training for high fidelity natural image synthesis.arXiv preprint arXiv:1809.11096, 2018. 10

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  7. [7]

    Gradient flow drifting: Generative mod- eling via wasserstein gradient flows of KDE-approximated divergences.arXiv preprint arXiv:2603.10592, 2026

    Jiarui Cao, Zixuan Wei, and Yuxin Liu. Gradient flow drifting: Generative modeling via Wasserstein gradient flows of KDE-approximated divergences.arXiv preprint arXiv:2603.10592,

    work page arXiv

  8. [8]

    Displacement smoothness of entropic optimal transport.ESAIM: Control, Optimisation and Calculus of Variations, 30:25, 2024

    Guillaume Carlier, Lénaïc Chizat, and Maxime Laborde. Displacement smoothness of entropic optimal transport.ESAIM: Control, Optimisation and Calculus of Variations, 30:25, 2024. 32

    2024

  9. [9]

    Springer, 2018

    René Carmona and François Delarue.Probabilistic Theory of Mean Field Games with Applica- tions I. Springer, 2018. 25

    2018

  10. [10]

    Maskgit: Masked generative image transformer

    Huiwen Chang, Han Zhang, Lu Jiang, Ce Liu, and William T Freeman. Maskgit: Masked generative image transformer. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11315–11325, 2022. 10

    2022

  11. [11]

    Scalable Wasserstein gradient flow for generative modeling through unbalanced optimal transport

    Jaemoo Choi, Jaewoong Choi, and Myungjoo Kang. Scalable Wasserstein gradient flow for generative modeling through unbalanced optimal transport. InProceedings of the 41st Inter- national Conference on Machine Learning, volume 235 ofProceedings of Machine Learning Research, pages 8629–8650. PMLR, 21–27 Jul 2024. 3, 24

    2024

  12. [12]

    Diffusion Schrödinger bridge with applications to score-based generative modeling.arXiv preprint arXiv:2106.01357,

    Valentin De Bortoli, James Thornton, Jeremy Heng, and Arnaud Doucet. Diffusion Schrödinger bridge with applications to score-based generative modeling.arXiv preprint arXiv:2106.01357,

    work page arXiv

  13. [13]

    ImageNet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. ImageNet: A large-scale hierarchical image database. InCVPR, pages 248–255. IEEE, 2009. 8

    2009

  14. [14]

    Generative Modeling via Drifting

    Mingyang Deng, He Li, Tianhong Li, Yilun Du, and Kaiming He. Generative modeling via drifting.arXiv preprint arXiv:2602.04770, 2026. 2, 3, 4, 6, 8, 9, 10, 21, 24, 33, 35

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  15. [15]

    Diffusion models beat GANs on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat GANs on image synthesis. NeurIPS, 34:8780–8794, 2021. 8, 10

    2021

  16. [16]

    Variational Wasser- stein gradient flow.arXiv preprint arXiv:2112.02424, 2021

    Jiaojiao Fan, Qinsheng Zhang, Amirhossein Taghvaei, and Yongxin Chen. Variational Wasser- stein gradient flow.arXiv preprint arXiv:2112.02424, 2021. 3 12

    work page arXiv 2021

  17. [17]

    Interpolating between optimal transport and mmd using Sinkhorn divergences

    Jean Feydy, Thibault Séjourné, François-Xavier Vialard, Shun-ichi Amari, Alain Trouvé, and Gabriel Peyré. Interpolating between optimal transport and mmd using Sinkhorn divergences. InThe 22nd international conference on artificial intelligence and statistics, pages 2681–2690. PMLR, 2019. 6

    2019

  18. [18]

    One Step Diffusion via Shortcut Models

    Kevin Frans, Danijar Hafner, Sergey Levine, and Pieter Abbeel. One step diffusion via shortcut models.arXiv preprint arXiv:2410.12557, 2024. 10

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  19. [19]

    Drift-Based Policy Optimization: Native One-Step Policy Learning for Online Robot Control

    Yuxuan Gao, Yedong Shen, Shiqi Zhang, Wenhao Yu, Yifan Duan, Jiajia Wu, Jiajun Deng, Yanyong Zhang, et al. Drift-based policy optimization: Native one-step policy learning for online robot control.arXiv preprint arXiv:2604.03540, 2026. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  20. [20]

    Learning generative models with Sinkhorn divergences

    Aude Genevay, Gabriel Peyré, and Marco Cuturi. Learning generative models with Sinkhorn divergences. InInternational Conference on Artificial Intelligence and Statistics, pages 1608–

  21. [21]

    3, 6, 24

    PMLR, 2018. 3, 6, 24

    2018

  22. [22]

    Mean Flows for One-step Generative Modeling

    Zhengyang Geng, Mingyang Deng, Xingjian Bai, J Zico Kolter, and Kaiming He. Mean flows for one-step generative modeling.arXiv preprint arXiv:2505.13447, 2025. 3, 10

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  23. [23]

    Improved Mean Flows: On the Challenges of Fastforward Generative Models

    Zhengyang Geng, Yiyang Lu, Zongze Wu, Eli Shechtman, J Zico Kolter, and Kaiming He. Improved mean flows: On the challenges of fastforward generative models.arXiv preprint arXiv:2512.02012, 2025. 3, 10

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  24. [24]

    Generative adversarial nets.NeurIPS, 2014

    Ian J Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets.NeurIPS, 2014. 2, 3

    2014

  25. [25]

    Improved training of Wasserstein GANs.Advances in neural information processing systems, 30, 2017

    Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron C Courville. Improved training of Wasserstein GANs.Advances in neural information processing systems, 30, 2017. 3

    2017

  26. [26]

    The Wasserstein gradient flow of the Sinkhorn divergence between gaussian distributions.arXiv preprint arXiv:2602.10726, 2026

    Mathis Hardion and Théo Lacombe. The Wasserstein gradient flow of the Sinkhorn divergence between gaussian distributions.arXiv preprint arXiv:2602.10726, 2026. 5

    work page arXiv 2026

  27. [27]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InCVPR, pages 770–778, 2016. 10

    2016

  28. [28]

    arXiv preprint arXiv:2603.12366 , year =

    Ping He, Om Khangaonkar, Hamed Pirsiavash, Yikun Bai, and Soheil Kolouri. Sinkhorn-drifting generative models.arXiv preprint arXiv:2603.12366, 2026. 3, 5, 7

    work page arXiv 2026

  29. [29]

    GANs trained by a two time-scale update rule converge to a local Nash equilibrium.NeurIPS,

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. GANs trained by a two time-scale update rule converge to a local Nash equilibrium.NeurIPS,

  30. [30]

    Denoising diffusion probabilistic models.NeurIPS, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models.NeurIPS, 33:6840–6851, 2020. 1, 2

    2020

  31. [31]

    Classifier-Free Diffusion Guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022. 8

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  32. [32]

    The variational formulation of the Fokker– Planck equation.SIAM journal on mathematical analysis, 29(1):1–17, 1998

    Richard Jordan, David Kinderlehrer, and Felix Otto. The variational formulation of the Fokker– Planck equation.SIAM journal on mathematical analysis, 29(1):1–17, 1998. 3

    1998

  33. [33]

    Scaling up GANs for text-to-image synthesis

    Minguk Kang, Jun-Yan Zhu, Richard Zhang, Jaesik Park, Eli Shechtman, Sylvain Paris, and Taesung Park. Scaling up GANs for text-to-image synthesis. InCVPR, pages 10124–10134,

  34. [34]

    Marlowe: Stanford’s gpu-based computational instrument, 2025

    Craig Kapfer, Kurt Stine, Balasubramanian Narasimhan, Christopher Mentzel, and Emmanuel Candes. Marlowe: Stanford’s gpu-based computational instrument, 2025. 12

    2025

  35. [35]

    A style-based generator architecture for generative adversarial networks

    Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4401–4410, 2019. 10 13

    2019

  36. [36]

    A Unified View of Score-Based and Drifting Models

    Chieh-Hsin Lai, Bac Nguyen, Naoki Murata, Yuhta Takida, Toshimitsu Uesaka, Yuki Mitsufuji, Stefano Ermon, and Molei Tao. A unified view of drifting and score-based models.arXiv preprint arXiv:2603.07514, 2026. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  37. [37]

    Autoregressive image generation without vector quantization.NeurIPS, 37:56424–56445, 2024

    Tianhong Li, Yonglong Tian, He Li, Mingyang Deng, and Kaiming He. Autoregressive image generation without vector quantization.NeurIPS, 37:56424–56445, 2024. 10

    2024

  38. [38]

    Generative moment matching networks

    Yujia Li, Kevin Swersky, and Rich Zemel. Generative moment matching networks. InICML, pages 1718–1727. PMLR, 2015. 5, 24

    2015

  39. [39]

    Generative Drifting for Conditional Medical Image Generation

    Zirong Li, Siyuan Mei, Weiwen Wu, Andreas Maier, Lina Gölz, and Yan Xia. Generative drifting for conditional medical image generation.arXiv preprint arXiv:2604.19736, 2026. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  40. [40]

    Adversarial Flow Models

    Shanchuan Lin, Ceyuan Yang, Zhijie Lin, Hao Chen, and Haoqi Fan. Adversarial flow models. arXiv preprint arXiv:2511.22475, 2025. 3, 10

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  41. [41]

    Flow Matching for Generative Modeling

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling.arXiv preprint arXiv:2210.02747, 2022. 1, 2, 3

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  42. [42]

    Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow.arXiv preprint arXiv:2209.03003, 2022. 1, 2, 3

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  43. [43]

    Simplifying, Stabilizing and Scaling Continuous-Time Consistency Models

    Cheng Lu and Yang Song. Simplifying, stabilizing and scaling continuous-time consistency models.arXiv preprint arXiv:2410.11081, 2024. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  44. [44]

    Schr ¨odinger bridge for generative AI: Soft-constrained formulation and convergence analysis,

    Jin Ma, Ying Tan, and Renyuan Xu. Schrödinger bridge for generative ai: Soft-constrained formulation and convergence analysis.arXiv preprint arXiv:2510.11829, 2025. 3

    work page arXiv 2025

  45. [45]

    SiT: Exploring flow and diffusion-based generative models with scalable interpolant transformers

    Nanye Ma, Mark Goldstein, Michael S Albergo, Nicholas M Boffi, Eric Vanden-Eijnden, and Saining Xie. SiT: Exploring flow and diffusion-based generative models with scalable interpolant transformers. InECCV, pages 23–40. Springer, 2024. 10, 35

    2024

  46. [46]

    Large-scale Wasserstein gradient flows.Advances in Neural Information Processing Systems, 34:15243–15256, 2021

    Petr Mokrov, Alexander Korotin, Lingxiao Li, Aude Genevay, Justin M Solomon, and Evgeny Burnaev. Large-scale Wasserstein gradient flows.Advances in Neural Information Processing Systems, 34:15243–15256, 2021. 3

    2021

  47. [47]

    Entropic optimal transport: Convergence of potentials

    Marcel Nutz and Johannes Wiesel. Entropic optimal transport: Convergence of potentials. Probability Theory and Related Fields, 184(1):401–424, 2022. 6

    2022

  48. [48]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InCVPR, pages 4195–4205, 2023. 8, 10

    2023

  49. [49]

    Now Foundations and Trends, 2019

    Gabriel Peyré and Marco Cuturi.Computational optimal transport: With applications to data science. Now Foundations and Trends, 2019. 6

    2019

  50. [50]

    Transport equation with nonlocal velocity in Wasserstein spaces: convergence of numerical schemes.Acta applicandae mathematicae, 124(1):73–105,

    Benedetto Piccoli and Francesco Rossi. Transport equation with nonlocal velocity in Wasserstein spaces: convergence of numerical schemes.Acta applicandae mathematicae, 124(1):73–105,

  51. [51]

    Adversarial latent autoen- coders

    Stanislav Pidhorskyi, Donald A Adjeroh, and Gianfranco Doretto. Adversarial latent autoen- coders. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14104–14113, 2020. 10

    2020

  52. [52]

    Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks

    Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep convolutional generative adversarial networks.arXiv preprint arXiv:1511.06434, 2015. 2, 3

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  53. [53]

    High- resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High- resolution image synthesis with latent diffusion models. InCVPR, pages 10684–10695, 2022. 8

    2022

  54. [54]

    Progressive Distillation for Fast Sampling of Diffusion Models

    Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models. arXiv preprint arXiv:2202.00512, 2022. 2 14

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  55. [55]

    Multistep distillation of diffusion models via moment matching.NeurIPS, 37:36046–36070, 2024

    Tim Salimans, Thomas Mensink, Jonathan Heek, and Emiel Hoogeboom. Multistep distillation of diffusion models via moment matching.NeurIPS, 37:36046–36070, 2024. 2

    2024

  56. [56]

    StyleGAN-XL: Scaling StyleGAN to large diverse datasets

    Axel Sauer, Katja Schwarz, and Andreas Geiger. StyleGAN-XL: Scaling StyleGAN to large diverse datasets. InSIGGRAPH, pages 1–10, 2022. 10

    2022

  57. [57]

    Concerning nonnegative matrices and doubly stochastic matrices.Pacific Journal of Mathematics, 21(2):343–348, 1967

    Richard Sinkhorn and Paul Knopp. Concerning nonnegative matrices and doubly stochastic matrices.Pacific Journal of Mathematics, 21(2):343–348, 1967. 7

    1967

  58. [58]

    Deep unsu- pervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsu- pervised learning using nonequilibrium thermodynamics. InICML, pages 2256–2265. PMLR,

  59. [59]

    Improved Techniques for Training Consistency Models

    Yang Song and Prafulla Dhariwal. Improved techniques for training consistency models.arXiv preprint arXiv:2310.14189, 2023. 2, 10

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  60. [60]

    Consistency models

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models. 2023. 2, 3

    2023

  61. [61]

    Generative modeling by estimating gradients of the data distribution.Advances in neural information processing systems, 32, 2019

    Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution.Advances in neural information processing systems, 32, 2019. 1

    2019

  62. [62]

    Score-Based Generative Modeling through Stochastic Differential Equations

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations.arXiv preprint arXiv:2011.13456, 2020. 1, 2

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  63. [63]

    Visual autoregressive modeling: Scalable image generation via next-scale prediction.NeurIPS, 37:84839–84865,

    Keyu Tian, Yi Jiang, Zehuan Yuan, Bingyue Peng, and Liwei Wang. Visual autoregressive modeling: Scalable image generation via next-scale prediction.NeurIPS, 37:84839–84865,

  64. [64]

    Tolstikhin, O

    Ilya Tolstikhin, Olivier Bousquet, Sylvain Gelly, and Bernhard Schoelkopf. Wasserstein auto- encoders.arXiv preprint arXiv:1711.01558, 2017. 3

    work page arXiv 2017

  65. [65]

    Generative Drifting is Secretly Score Matching: a Spectral and Variational Perspective

    Erkan Turan and Maks Ovsjanikov. Generative drifting is secretly score matching: a spectral and variational perspective.arXiv preprint arXiv:2603.09936, 2026. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  66. [66]

    Pro- lificdreamer: High-fidelity and diverse text-to-3d generation with variational score distillation

    Zhengyi Wang, Cheng Lu, Yikai Wang, Fan Bao, Chongxuan Li, Hang Su, and Jun Zhu. Pro- lificdreamer: High-fidelity and diverse text-to-3d generation with variational score distillation. Advances in neural information processing systems, 36:8406–8441, 2023. 2

    2023

  67. [67]

    arXiv preprint arXiv:2509.04394 (2025) 18 M

    Zidong Wang, Yiyuan Zhang, Xiaoyu Yue, Xiangyu Yue, Yangguang Li, Wanli Ouyang, and Lei Bai. Transition models: Rethinking the generative learning objective.arXiv preprint arXiv:2509.04394, 2025. 10

    work page arXiv 2025

  68. [68]

    Flow-based generative models as iterative algorithms in probability space.arXiv preprint arXiv:2502.13394, 2025

    Yao Xie and Xiuyuan Cheng. Flow-based generative models as iterative algorithms in probability space.arXiv preprint arXiv:2502.13394, 2025. 3

    work page arXiv 2025

  69. [69]

    Reconstruction vs

    Jingfeng Yao, Bin Yang, and Xinggang Wang. Reconstruction vs. generation: Taming op- timization dilemma in latent diffusion models. InCVPR, pages 15703–15712, 2025. 10, 35

    2025

  70. [70]

    Improved distribution matching distillation for fast image synthesis

    Tianwei Yin, Michaël Gharbi, Taesung Park, Richard Zhang, Eli Shechtman, Fredo Durand, and William T Freeman. Improved distribution matching distillation for fast image synthesis. Advances in neural information processing systems, 37:47455–47487, 2024. 2

    2024

  71. [71]

    One-step diffusion with distribution matching distillation

    Tianwei Yin, Michaël Gharbi, Richard Zhang, Eli Shechtman, Fredo Durand, William T Freeman, and Taesung Park. One-step diffusion with distribution matching distillation. In CVPR, pages 6613–6623, 2024. 2

    2024

  72. [72]

    Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think

    Sihyun Yu, Sangkyung Kwak, Huiwon Jang, Jongheon Jeong, Jonathan Huang, Jinwoo Shin, and Saining Xie. Representation alignment for generation: Training diffusion transformers is easier than you think.arXiv preprint arXiv:2410.06940, 2024. 10

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  73. [73]

    arXiv preprint arXiv:2510.20771 , year=

    Huijie Zhang, Aliaksandr Siarohin, Willi Menapace, Michael Vasilkovsky, Sergey Tulyakov, Qing Qu, and Ivan Skorokhodov. AlphaFlow: Understanding and improving MeanFlow models. arXiv preprint arXiv:2510.20771, 2025. 10 15

    work page arXiv 2025

  74. [74]

    Diffusion Transformers with Representation Autoencoders

    Boyang Zheng, Nanye Ma, Shengbang Tong, and Saining Xie. Diffusion transformers with representation autoencoders.arXiv preprint arXiv:2510.11690, 2025. 10

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  75. [75]

    arXiv preprint arXiv:2503.07565 , year=

    Linqi Zhou, Stefano Ermon, and Jiaming Song. Inductive moment matching.arXiv preprint arXiv:2503.07565, 2025. 2, 3, 24 16 Appendix Table of Contents A Additional discussions 17 A.1 Wasserstein gradient flows of energy functionals . . . . . . . . . . . . . . . . . 17 A.2 More discussion on the estimators for self-transport . . . . . . . . . . . . . . . 21 ...

    work page arXiv 2025