pith. sign in

arxiv: 2607.00149 · v1 · pith:2U3RBKRVnew · submitted 2026-06-30 · 🧮 math.PR · stat.ML

Uniform-in-time Propagation-of-Chaos for Stein Variational Gradient Descent

Krishnakumar Balasubramanian , Sayan Banerjee , Anna Korba This is my paper

Pith reviewed 2026-07-02 17:28 UTC · model grok-4.3

classification 🧮 math.PR stat.ML
keywords propagation of chaosStein variational gradient descentmean-field limitsuniform in timekernel Stein discrepancyWasserstein distanceinteracting particle systemsconjugacy
0
0 comments X

The pith

SVGD particle systems remain close to their mean-field limit uniformly in time, via cutoff arguments or moment closure.

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

The paper establishes two kinds of uniform-in-time propagation-of-chaos bounds for continuous-time Stein Variational Gradient Descent. For general metrics a cutoff splices short-time particle-mean-field closeness with separate long-time convergence estimates for each system, producing time-averaged closeness in kernel Stein discrepancy and Wasserstein distances at logarithmic rates. For matrix-valued kernels on Gaussian targets the dynamics close exactly on low-dimensional moments, delivering parametric N^{-1/2} closeness that holds for every physical time and transfers to other targets by conjugacy under diffeomorphisms. The results separate generic distributional metrics, where rates remain logarithmic, from closed finite-dimensional observables, where rates stay parametric.

Core claim

We obtain two complementary classes of uniform-in-time propagation-of-chaos results for continuous-time SVGD. For broad distributional metrics, we introduce a cutoff strategy which combines finite-time propagation-of-chaos estimates up to an N-dependent horizon with independent quantitative long-time convergence estimates for the finite-particle and mean-field SVGD flows. This yields uniform-in-averaging-time propagation-of-chaos bounds in Langevin kernel Stein discrepancy, Wasserstein-1 distance, and Wasserstein-2 distance, with logarithmic or iterated-logarithmic rates depending on the metric, target and kernel class. We also develop a finite-dimensional theory for matrix-valued finite-ran

What carries the argument

Cutoff strategy that splices finite-time propagation-of-chaos estimates with independent long-time convergence bounds for both systems, together with a conjugacy principle that transfers moment-closure estimates across orientation-preserving diffeomorphisms.

If this is right

  • Uniform-in-averaging-time propagation-of-chaos holds in Langevin kernel Stein discrepancy, Wasserstein-1 and Wasserstein-2 with logarithmic or iterated-logarithmic rates.
  • Genuine uniform-in-physical-time parametric N^{-1/2} rates hold in finite-dimensional Stein-feature metrics when the dynamics close on moments.
  • The feature-level estimates extend to broad classes of nonlinear and multimodal targets via the conjugacy principle.
  • Generic distributional metrics yield only logarithmic rates while closed finite-dimensional Stein observables retain parametric rates uniformly in time.

Where Pith is reading between the lines

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

  • The cutoff technique may extend to other mean-field interacting particle systems where separate long-time convergence results are already known.
  • Kernel design that forces exact closure on a few low-dimensional statistics could produce sampling algorithms whose error does not accumulate over long runs.
  • The distinction between time-averaged and physical-time uniformity suggests that practitioners should choose metrics adapted to the observables they actually track.

Load-bearing premise

Quantitative long-time convergence estimates exist independently for both the finite-particle and mean-field SVGD flows.

What would settle it

A concrete target and kernel for which the Wasserstein distance between the N-particle empirical measure and the mean-field limit grows unbounded as time increases while both systems individually converge to their stationary distributions.

read the original abstract

We study uniform-in-time propagation-of-chaos for continuous-time Stein Variational Gradient Descent (SVGD). Classical finite-time propagation-of-chaos estimates for mean-field systems typically deteriorate rapidly with time and therefore do not directly explain the long-time relation between the finite-particle system and its mean-field limit. We obtain two complementary classes of uniform-in-time propagation-of-chaos results. For broad distributional metrics, we introduce a cutoff strategy which combines finite-time propagation-of-chaos estimates up to an $N$-dependent horizon with independent quantitative long-time convergence estimates for the finite-particle and mean-field SVGD flows. This yields uniform-in-averaging-time propagation-of-chaos bounds in Langevin kernel Stein discrepancy, Wasserstein-1 distance, and Wasserstein-2 distance, with logarithmic or iterated-logarithmic rates depending on the metric, target and kernel class. We also develop a finite-dimensional theory for matrix-valued finite-rank kernels. For Gaussian targets with bilinear kernels, the SVGD dynamics close exactly on first and second moments, yielding genuine uniform-in-physical-time parametric propagation-of-chaos rates in finite-dimensional Stein-feature metrics. We then prove a conjugacy principle showing that these feature-level estimates transfer to conjugate target-kernel pairs under orientation-preserving diffeomorphisms, thereby extending the theory to broad classes of nonlinear, including multimodal, targets. Together, these results highlight the contrast between generic distributional metrics, for which our general approach yields logarithmic rates, and closed finite-dimensional Stein observables, for which parametric $N^{-1/2}$ propagation-of-chaos rates persist uniformly in time.

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

1 major / 1 minor

Summary. The paper claims two complementary classes of uniform-in-time propagation-of-chaos (PoC) results for continuous-time Stein Variational Gradient Descent (SVGD). The first uses a cutoff strategy combining finite-time PoC estimates up to an N-dependent horizon T_N with independent quantitative long-time convergence estimates for the N-particle and mean-field flows, yielding uniform-in-averaging-time PoC bounds in Langevin kernel Stein discrepancy, Wasserstein-1, and Wasserstein-2 distances with logarithmic or iterated-logarithmic rates. The second develops a finite-dimensional theory for matrix-valued finite-rank kernels: for Gaussian targets with bilinear kernels the dynamics close on first and second moments, giving genuine uniform-in-physical-time parametric PoC rates in Stein-feature metrics; a conjugacy principle then extends these to conjugate target-kernel pairs under orientation-preserving diffeomorphisms, covering broad classes of nonlinear targets.

Significance. If the results hold, the work addresses a central limitation of classical finite-time PoC estimates by establishing long-time control for SVGD particle systems. The contrast between generic distributional metrics (logarithmic rates) and closed finite-dimensional Stein observables (parametric N^{-1/2} rates) is conceptually useful, and the conjugacy principle broadens applicability to multimodal targets. The approach of leveraging independent long-time estimates and exact moment closure is a strength when the required uniformity can be verified.

major comments (1)
  1. [Abstract (cutoff strategy paragraph)] The cutoff argument for the first class of results (uniform-in-averaging-time PoC) closes only if quantitative long-time convergence rates to equilibrium exist independently for both the finite-particle and mean-field SVGD flows and remain uniform in N in the metrics Langevin KSD, W1, and W2. The abstract invokes these estimates as given; the manuscript must explicitly establish or cite their N-uniformity (particularly whether particle-system rates deteriorate through the interaction kernel) to justify the claimed logarithmic and iterated-logarithmic rates after averaging.
minor comments (1)
  1. [Abstract] The abstract outlines proof strategies but supplies no derivations, error bounds, or verification steps for the long-time estimates or conjugacy; the full manuscript should include at least one concrete example verifying the moment closure and conjugacy transfer.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading and for identifying this important point about the cutoff argument. We address the comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract (cutoff strategy paragraph)] The cutoff argument for the first class of results (uniform-in-averaging-time PoC) closes only if quantitative long-time convergence rates to equilibrium exist independently for both the finite-particle and mean-field SVGD flows and remain uniform in N in the metrics Langevin KSD, W1, and W2. The abstract invokes these estimates as given; the manuscript must explicitly establish or cite their N-uniformity (particularly whether particle-system rates deteriorate through the interaction kernel) to justify the claimed logarithmic and iterated-logarithmic rates after averaging.

    Authors: We agree that explicit verification of N-uniformity is necessary to close the cutoff argument and justify the rates. The manuscript draws the long-time estimates from the existing literature on SVGD and Langevin dynamics (e.g., results establishing exponential or polynomial convergence in KSD/Wasserstein metrics under standard assumptions on the target and kernel). For the mean-field flow these rates are N-independent by construction. For the N-particle system, the rates remain uniform in N because the interaction occurs through the empirical measure and the kernel is fixed (positive definite, bounded derivatives); the particle-system convergence to equilibrium does not deteriorate with N beyond the mean-field limit. To make this fully explicit as requested, we will add a dedicated remark or subsection in the cutoff-strategy section that cites the precise long-time results, states the uniformity assumptions, and confirms that the interaction kernel does not cause rate deterioration in the relevant metrics. We will also update the abstract to reference this uniformity explicitly. revision: yes

Circularity Check

0 steps flagged

Cutoff strategy invokes independent long-time convergence estimates; finite-dimensional closure derived directly without reduction to inputs

full rationale

The paper's first class of results explicitly combines finite-time PoC with separate 'independent quantitative long-time convergence estimates' for N-particle and mean-field flows (abstract), which are treated as external inputs rather than derived or fitted within the PoC argument. The second class derives moment closure for Gaussian targets with bilinear kernels and a conjugacy principle directly from the dynamics, without self-referential fitting or renaming. No quoted step reduces a claimed prediction to a definition or self-citation chain by construction. This matches the default expectation of no significant circularity (score 0-2) when results rest on independent external estimates.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The work rests on standard domain assumptions in mean-field interacting particle systems and kernel methods; no free parameters or new invented entities are introduced in the abstract.

axioms (2)
  • domain assumption Existence, uniqueness, and well-posedness of the continuous-time SVGD particle and mean-field flows
    Required to state and compare the finite-particle and mean-field dynamics throughout the paper.
  • domain assumption Availability of quantitative long-time convergence estimates for both finite-particle and mean-field SVGD
    Directly invoked to close the cutoff argument for uniform-in-averaging-time bounds.

pith-pipeline@v0.9.1-grok · 5821 in / 1305 out tokens · 34609 ms · 2026-07-02T17:28:13.172738+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

45 extracted references · 13 canonical work pages · 10 internal anchors

  1. [1]

    Finite-Particle Convergence Rates for Conservative and Non-Conservative Drifting Models

    Krishnakumar Balasubramanian. Finite-particle convergence rates for conservative and non- conservative drifting models.arXiv preprint arXiv:2605.22795, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  2. [2]

    Improved finite-particle convergence rates for Stein variational gradient descent

    Sayan Banerjee, Krishnakumar Balasubramanian, and Promit Ghosal. Improved finite-particle convergence rates for Stein variational gradient descent. InInternational Conference on Learn- ing Representations, 2025

    2025

  3. [3]

    Flocking under Fast and Large Jumps: Stability, Chaos, and Traveling Waves.arXiv preprint arXiv:2404.13117, 2024

    Sayan Banerjee, Amarjit Budhiraja, and Dilshad Imon. Flocking under Fast and Large Jumps: Stability, Chaos, and Traveling Waves.arXiv preprint arXiv:2404.13117, 2024

    work page arXiv 2024

  4. [4]

    Targeted separation and convergence with kernel discrepancies.Journal of Machine Learning Research, 25(378):1–50, 2024

    Alessandro Barp, Carl-Johann Simon-Gabriel, Mark Girolami, and Lester Mackey. Targeted separation and convergence with kernel discrepancies.Journal of Machine Learning Research, 25(378):1–50, 2024

    2024

  5. [5]

    Bobkov and Michel Ledoux.One-Dimensional Empirical Measures, Order Statistics, and Kantorovich Transport Distances, volume 261 ofMemoirs of the American Mathematical Society

    Sergey G. Bobkov and Michel Ledoux.One-Dimensional Empirical Measures, Order Statistics, and Kantorovich Transport Distances, volume 261 ofMemoirs of the American Mathematical Society. American Mathematical Society, Providence, RI, 2019

    2019

  6. [6]

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

    work page arXiv 2026

  7. [7]

    Convergence and stability results for the particle system in the Stein gradient descent method.Mathematics of Computation, 94(354):1793– 1814, 2025

    Jos´ e Carrillo and Jakub Skrzeczkowski. Convergence and stability results for the particle system in the Stein gradient descent method.Mathematics of Computation, 94(354):1793– 1814, 2025

    2025

  8. [8]

    The Stein-log-Sobolev inequality and the exponential rate of convergence for the continuous Stein variational gradient descent method.arXiv preprint arXiv:2412.10295, 2024

    Jos´ e A Carrillo, Jakub Skrzeczkowski, and Jethro Warnett. The Stein-log-Sobolev inequality and the exponential rate of convergence for the continuous Stein variational gradient descent method.arXiv preprint arXiv:2412.10295, 2024. 53

    work page arXiv 2024

  9. [9]

    Stein Variational Gradient Descent dynamics for highly concentrated kernels

    Jos´ e A Carrillo, Jakub Skrzeczkowski, and Jethro Warnett. Stein Variational Gradient Descent dynamics for highly concentrated kernels.arXiv preprint arXiv:2605.03627, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  10. [10]

    Probabilistic approach for granular media equations in the non-uniformly convex case.Probability Theory and Related Fields, 140(1–2):19–40, 2008

    Patrick Cattiaux, Arnaud Guillin, and Florent Malrieu. Probabilistic approach for granular media equations in the non-uniformly convex case.Probability Theory and Related Fields, 140(1–2):19–40, 2008

    2008

  11. [11]

    Propagation of chaos: A review of models, methods and applications

    Louis-Pierre Chaintron and Antoine Diez. Propagation of chaos: A review of models, methods and applications. I. Models and methods.Kinetic and Related Models, 15(6):895–1015, 2022

    2022

  12. [12]

    Quantitative Local Convergence of Mean-Field Stein Variational Gradient Flow

    L´ ena¨ ıc Chizat, Maria Colombo, Roberto Colombo, and Xavier Fern´ andez-Real. Quanti- tative Local Convergence of Mean-Field Stein Variational Gradient Flow.arXiv preprint arXiv:2605.09456, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  13. [13]

    A kernel test of goodness of fit

    Kacper Chwialkowski, Heiko Strathmann, and Arthur Gretton. A kernel test of goodness of fit. InInternational conference on machine learning, pages 2606–2615. PMLR, 2016

    2016

  14. [14]

    Uniform in Time Weak Propagation of Chaos on the Torus

    Fran¸ cois Delarue and Alvin Tse. Uniform in Time Weak Propagation of Chaos on the Torus. Annales de l’Institut Henri Poincar´ e, Probabilit´ es et Statistiques, 61(2):1021–1074, 2025

    2025

  15. [15]

    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

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  16. [16]

    On the geometry of Stein variational gradient descent.Journal of Machine Learning Research, 24(56):1–39, 2023

    Andrew Duncan, Nikolas N¨ usken, and Lukasz Szpruch. On the geometry of Stein variational gradient descent.Journal of Machine Learning Research, 24(56):1–39, 2023

    2023

  17. [17]

    An elementary ap- proach to uniform in time propagation of chaos.Proceedings of the American Mathematical Society, 148(12):5387–5398, 2020

    Alain Durmus, Andreas Eberle, Arnaud Guillin, and Raphael Zimmer. An elementary ap- proach to uniform in time propagation of chaos.Proceedings of the American Mathematical Society, 148(12):5387–5398, 2020

    2020

  18. [18]

    Learning to Draw Samples with Amortized Stein Variational Gradient Descent

    Yihao Feng, Dilin Wang, and Qiang Liu. Learning to draw samples with amortized Stein variational gradient descent.arXiv preprint arXiv:1707.06626, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  19. [19]

    On the rate of convergence in wasserstein distance of the empirical measure.Probability theory and related fields, 162(3):707–738, 2015

    Nicolas Fournier and Arnaud Guillin. On the rate of convergence in wasserstein distance of the empirical measure.Probability theory and related fields, 162(3):707–738, 2015

    2015

  20. [20]

    Drifting Fields are not Conservative

    Leonard T Franz, Sebastian Hoffmann, Tim Weiland, Bernhard Sch¨ olkopf, and Georg Martius. Drifting fields are not conservative.arXiv preprint arXiv:2604.06333, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  21. [21]

    Uniform-in-Time Weak Propagation-of-Chaos in Shallow Neural Networks

    Margalit Glasgow and Joan Bruna. Uniform-in-Time Weak Propagation-of-Chaos in Shallow Neural Networks.arXiv preprint arXiv:2605.22010, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  22. [22]

    On the Wasserstein Gradient Flow Interpretation of Drifting Models

    Arthur Gretton, Li Kevin Wenliang, Alexandre Galashov, James Thornton, Valentin De Bor- toli, and Arnaud Doucet. On the Wasserstein Gradient Flow Interpretation of Drifting Models. arXiv preprint arXiv:2605.05118, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  23. [23]

    Uniform Long-Time and Propagation of Chaos Es- timates for Mean Field Kinetic Particles in Non-convex Landscapes.Journal of Statistical Physics, 185(2):15, 2021

    Arnaud Guillin and Pierre Monmarch´ e. Uniform Long-Time and Propagation of Chaos Es- timates for Mean Field Kinetic Particles in Non-convex Landscapes.Journal of Statistical Physics, 185(2):15, 2021

    2021

  24. [24]

    Finite-Particle Rates for Regularized Stein Variational Gradient Descent

    Ye He, Krishnakumar Balasubramanian, Sayan Banerjee, and Promit Ghosal. Finite-Particle Rates for Regularized Stein Variational Gradient Descent.arXiv preprint arXiv:2602.05172, 2026. 54

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  25. [25]

    Regular- ized Stein variational gradient flow.Foundations of Computational Mathematics, 25(4):1199– 1257, 2025

    Ye He, Krishnakumar Balasubramanian, Bharath K Sriperumbudur, and Jianfeng Lu. Regular- ized Stein variational gradient flow.Foundations of Computational Mathematics, 25(4):1199– 1257, 2025

    2025

  26. [26]

    Controlling mo- ments with kernel Stein discrepancies.The Annals of Applied Probability, 35(6):3818–3843, 2025

    Heishiro Kanagawa, Alessandro Barp, Arthur Gretton, and Lester Mackey. Controlling mo- ments with kernel Stein discrepancies.The Annals of Applied Probability, 35(6):3818–3843, 2025

    2025

  27. [27]

    A generalization of Caffarelli’s contraction theorem via (reverse) heat flow.Mathematische Annalen, 354(3):827–862, 2012

    Young-Heon Kim and Emanuel Milman. A generalization of Caffarelli’s contraction theorem via (reverse) heat flow.Mathematische Annalen, 354(3):827–862, 2012

    2012

  28. [28]

    A non-asymptotic analysis for Stein variational gradient descent

    Anna Korba, Adil Salim, Michael Arbel, Giulia Luise, and Arthur Gretton. A non-asymptotic analysis for Stein variational gradient descent. In H. Larochelle, M. Ranzato, R. Hadsell, M. F. Balcan, and H. Lin, editors,Advances in Neural Information Processing Systems, volume 33, pages 4672–4682. Curran Associates, Inc., 2020

    2020

  29. [29]

    Hierarchies, entropy, and quantitative propagation of chaos for mean field diffusions.Probability and Mathematical Physics, 4(2):377–432, 2023

    Daniel Lacker. Hierarchies, entropy, and quantitative propagation of chaos for mean field diffusions.Probability and Mathematical Physics, 4(2):377–432, 2023

    2023

  30. [30]

    Sharp uniform-in-time propagation of chaos.Probability Theory and Related Fields, 187:443–480, 2023

    Daniel Lacker and Luc Le Flem. Sharp uniform-in-time propagation of chaos.Probability Theory and Related Fields, 187:443–480, 2023

    2023

  31. [31]

    On optimal matching of Gaussian samples.Zap

    Michel Ledoux. On optimal matching of Gaussian samples.Zap. Nauchn. Sem. S.-Peterburg. Otdel. Mat. Inst. Steklov. (POMI), 457:226–264, 2017. Veroyatnost’ i Statistika. 25

    2017

  32. [32]

    On optimal matching of Gaussian samples III.Probability and Mathematical Statistics, 41(2):237–265, 2021

    Michel Ledoux and Jie-Xiang Zhu. On optimal matching of Gaussian samples III.Probability and Mathematical Statistics, 41(2):237–265, 2021

    2021

  33. [33]

    Stein variational gradient descent as gradient flow

    Qiang Liu. Stein variational gradient descent as gradient flow. InAdvances in Neural Infor- mation Processing Systems, volume 30, 2017

    2017

  34. [34]

    A kernelized Stein discrepancy for goodness-of-fit tests

    Qiang Liu, Jason Lee, and Michael Jordan. A kernelized Stein discrepancy for goodness-of-fit tests. InInternational conference on machine learning, pages 276–284. PMLR, 2016

    2016

  35. [35]

    Stein variational gradient descent: A general purpose Bayesian inference algorithm

    Qiang Liu and Dilin Wang. Stein variational gradient descent: A general purpose Bayesian inference algorithm. InAdvances in Neural Information Processing Systems, volume 29, 2016

    2016

  36. [36]

    Towards understanding the dynamics of Gaussian-Stein variational gradient descent.Advances in Neural Information Processing Systems, 36:61234–61291, 2023

    Tianle Liu, Promit Ghosal, Krishnakumar Balasubramanian, and Natesh Pillai. Towards understanding the dynamics of Gaussian-Stein variational gradient descent.Advances in Neural Information Processing Systems, 36:61234–61291, 2023

    2023

  37. [37]

    Scaling limit of the Stein variational gradient descent: the mean field regime.SIAM Journal on Mathematical Analysis, 51(2):648–671, 2019

    Jianfeng Lu, Yulong Lu, and James Nolen. Scaling limit of the Stein variational gradient descent: the mean field regime.SIAM Journal on Mathematical Analysis, 51(2):648–671, 2019

    2019

  38. [38]

    Convergence to equilibrium for granular media equations and their Euler schemes.The Annals of Applied Probability, 13(2):540–560, 2003

    Florent Malrieu. Convergence to equilibrium for granular media equations and their Euler schemes.The Annals of Applied Probability, 13(2):540–560, 2003

    2003

  39. [39]

    Integral probability metrics and their generating classes of functions.Advances in applied probability, 29(2):429–443, 1997

    Alfred M¨ uller. Integral probability metrics and their generating classes of functions.Advances in applied probability, 29(2):429–443, 1997

    1997

  40. [40]

    A convergence theory for SVGD in the popula- tion limit under Talagrand’s inequalityT 1

    Adil Salim, Lukang Sun, and Peter Richtarik. A convergence theory for SVGD in the popula- tion limit under Talagrand’s inequalityT 1. InInternational Conference on Machine Learning, pages 19139–19152. PMLR, 2022. 55

    2022

  41. [41]

    A finite-particle convergence rate for Stein variational gradient descent

    Jiaxin Shi and Lester Mackey. A finite-particle convergence rate for Stein variational gradient descent. InAdvances in Neural Information Processing Systems, volume 36, 2023

    2023

  42. [42]

    Convergence of stein variational gra- dient descent under a weaker smoothness condition

    Lukang Sun, Avetik Karagulyan, and Peter Richtarik. Convergence of stein variational gra- dient descent under a weaker smoothness condition. InInternational Conference on Artificial Intelligence and Statistics, pages 3693–3717. PMLR, 2023

    2023

  43. [43]

    Topics in propagation of chaos

    Alain-Sol Sznitman. Topics in propagation of chaos. InEcole d’´ et´ e de probabilit´ es de Saint- Flour XIX—1989, pages 165–251. Springer, 2006

    1989

  44. [44]

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

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

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  45. [45]

    Stein Variational Gradient De- scent with Matrix-Valued Kernels

    Dilin Wang, Ziyang Tang, Chandrajit Bajaj, and Qiang Liu. Stein Variational Gradient De- scent with Matrix-Valued Kernels. InAdvances in Neural Information Processing Systems, volume 32, 2019. 56

    2019