pith. sign in

arxiv: 2510.02706 · v3 · submitted 2025-10-03 · 🧮 math.OC

Flow Matching for Measure Transport and Feedback Stabilization of Control-Affine Systems

Karthik Elamvazhuthi This is my paper

Pith reviewed 2026-05-18 10:57 UTC · model grok-4.3

classification 🧮 math.OC
keywords flow matchingmeasure transportcontrol-affine systemsfeedback stabilizationcontinuity equationPontryagin maximum principledenoisingpath planning
0
0 comments X

The pith

Flow matching from the continuity equation lets classical control tools generate regression-based feedback for measure transport under control-affine dynamics.

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

The paper constructs exact and approximate flow matchings starting from the continuity equation of the system dx/dt = f0(x) plus sum ui fi(x) to generate interpolations between probability measures. This construction directly supports importing feedback design, oscillatory inputs, and trajectory steering to produce sample-efficient controllers that steer one measure to another. A parallel noising-plus-time-reversal view recasts stabilization as a denoising task, with noising processes built either from randomized regular controls through the endpoint map or from randomized adjoints in the Hamiltonian system of Pontryagin's principle. The same ideas extend to output flow matching when only output distributions need to align and are illustrated on linear and nonlinear examples including steering to target sets and obstacle-aware path planning.

Core claim

Starting from the continuity equation of the control-affine system, the paper builds measure interpolations via exact or approximate flow matching and shows that these interpolations preserve enough structure to apply standard control synthesis methods, thereby producing regression-based feedback controllers for measure-to-measure transport. Stabilization is reinterpreted as a denoising problem in which controlled excitations generate the noising process and time reversal yields the stabilizing feedback; two concrete constructions are given, one using endpoint maps with arbitrary controls and one using randomized adjoints from the maximum principle.

What carries the argument

Flow matching constructed directly from the continuity equation of dx/dt = f0(x) + sum ui fi(x), which produces the interpolating curves between measures while keeping the vector fields available for classical feedback and steering design.

If this is right

  • Standard feedback design and oscillatory control inputs become directly usable for measure-to-measure transport tasks.
  • Sample-efficient regression controllers can be trained from trajectory samples rather than solving infinite-dimensional optimal transport problems.
  • Control constraints are accommodated naturally through the randomized-control noising construction.
  • For linear systems with convex costs the PMP-based noising recovers the known optimal feedback law.

Where Pith is reading between the lines

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

  • The output-flow-matching variant suggests a route to controllers that act on partial observations or sensor distributions.
  • The framework may combine with existing sampling-based planners to reduce the number of rollouts needed for high-dimensional navigation.
  • Numerical comparisons on the same benchmark systems could quantify how much sample efficiency is gained relative to direct optimal-control or diffusion baselines.

Load-bearing premise

That flow matchings built from the continuity equation can be made to preserve the structure needed for classical control techniques to apply without losing their guarantees or efficiency.

What would settle it

A low-dimensional linear control-affine system in which the regression-based controller obtained from the flow-matched interpolation fails to transport the initial measure to the target measure within the expected error bounds.

Figures

Figures reproduced from arXiv: 2510.02706 by Karthik Elamvazhuthi.

Figure 1
Figure 1. Figure 1: Comparison of final output positions projections of target distribution and trained six-state [PITH_FULL_IMAGE:figures/full_fig_p021_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of final positions projections of target distribution and trained driftless system [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Trajectories of the time-reverse system visualized in 3D, stabilized to the origin. [PITH_FULL_IMAGE:figures/full_fig_p023_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Trajectories of the time-reverse system visualized in 3D, stabilized to the unit sphere. [PITH_FULL_IMAGE:figures/full_fig_p023_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Time-reversed trajectories of the Martinet system visualized in 3D. The learned policy [PITH_FULL_IMAGE:figures/full_fig_p024_5.png] view at source ↗
read the original abstract

We develop a \emph{flow-matching framework} for transporting probability measures under control-affine dynamics and for steering systems to points or target sets. Starting from the continuity equation associated with the control affine system of the form dx/dt = f_0(x) + \sum_{i=1}^m u_i f_i(x), we construct measure interpolations through exact, approximate flow matching, and extend the approach to output flow matching when only output distributions must align. These constructions allow to directly import standard control tools, such as feedback design, oscillatory inputs, and trajectory steering, and yield sample-efficient, regression-based feedback controllers for measure-to-measure transport. We also introduce a complementary ``noising + time-reversal'' perspective for classical state or set stabilization, inspired by denoising diffusion models. Here stabilization is interpreted as a denoising problem: noising corresponds to destabilizing the system through excitations, while denoising corresponds to stabilization via time reversal. We propose two methods for constructing the noising process: (i) Randomized-control noising, which employs regular (non-white noise) controls through the endpoint map and naturally accommodates control constraints. (iI) PMP-based noising, which leverages the Hamiltonian system from Pontryagin's Maximum Principle, corresponding to fixed or variable end-point optimal control problems, to explore the configuration space by randomizing the adjoint vectors and recovers the optimal controller for linear systems with convex costs, while providing feasible feedback laws in the nonlinear case. Finally, we numerically illustrate the framework on linear and nonlinear systems, demonstrating its effectiveness for measure transport, steering systems to target sets and path planning in a domain with obstacles.

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 develops a flow-matching framework for transporting probability measures under control-affine dynamics dx/dt = f_0(x) + sum u_i f_i(x). Starting from the associated continuity equation, it constructs measure interpolations via exact and approximate flow matching, extends the approach to output flow matching, and introduces a complementary noising + time-reversal perspective for stabilization using randomized-control noising or PMP-based noising. These are claimed to allow direct import of classical control tools (feedback design, oscillatory inputs, trajectory steering) for sample-efficient regression-based controllers, with numerical illustrations on linear and nonlinear systems for measure transport, set steering, and obstacle-avoiding path planning.

Significance. If the central claims hold, the work offers a promising bridge between flow-matching/generative modeling techniques and nonlinear control, potentially enabling practical, data-driven feedback for measure-to-measure transport and stabilization tasks. Credit is due for the dual perspectives (flow matching and noising/time-reversal), the explicit handling of control-affine structure in the continuity equation, and the numerical demonstrations across linear and nonlinear examples. The significance is tempered by the need to confirm that approximate constructions preserve admissibility of the resulting controls.

major comments (2)
  1. [§3.2] §3.2 (Approximate flow matching): the velocity field v_t is obtained via regression to the continuity equation without an explicit projection onto the span of {f_i} or quantitative error bounds on the resulting closed-loop vector field. This directly affects the abstract claim that the constructions 'directly import standard control tools, such as feedback design'; the regressed feedback may exit the admissible control distribution, and no uniform bounds are supplied showing preservation of stability or steering guarantees.
  2. [§5.1] §5.1 (Randomized-control noising): while the endpoint-map construction accommodates control constraints, the manuscript does not provide a stability or reachability analysis showing that the time-reversed denoising process recovers the original stabilization objective when the noising controls are only approximately realizable.
minor comments (3)
  1. [Abstract] Abstract: the two noising methods are both labeled (i); the second should be (ii).
  2. [§3] Notation: the distinction between exact and approximate matching velocity fields could be made clearer in the main text by consistent use of subscripts or superscripts (e.g., v_t^exact vs. v_t^approx).
  3. [Numerical experiments] Numerical section: figure captions would benefit from explicit statements of the system dimensions, sampling sizes, and regression method (e.g., neural network architecture) to aid reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thorough review and constructive comments on our manuscript. We address each major comment point by point below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (Approximate flow matching): the velocity field v_t is obtained via regression to the continuity equation without an explicit projection onto the span of {f_i} or quantitative error bounds on the resulting closed-loop vector field. This directly affects the abstract claim that the constructions 'directly import standard control tools, such as feedback design'; the regressed feedback may exit the admissible control distribution, and no uniform bounds are supplied showing preservation of stability or steering guarantees.

    Authors: We thank the referee for this important remark. In §3.2 the approximate flow matching obtains the velocity field v_t by regression against the continuity equation; controls are subsequently recovered by solving the linear system v = f_0(x) + F(x) u for u. We agree that an explicit projection step onto the span of the control vector fields is not detailed in the current text and that quantitative error bounds on the closed-loop vector field, together with uniform guarantees on admissibility or stability, are not supplied. In the revised manuscript we will clarify the control-recovery procedure and add a short discussion of observed approximation errors (supported by the existing numerical examples) together with a note on the absence of uniform theoretical guarantees. This is a partial revision. revision: partial

  2. Referee: [§5.1] §5.1 (Randomized-control noising): while the endpoint-map construction accommodates control constraints, the manuscript does not provide a stability or reachability analysis showing that the time-reversed denoising process recovers the original stabilization objective when the noising controls are only approximately realizable.

    Authors: We appreciate the referee highlighting this point. Section 5.1 constructs the noising process via the endpoint map so that control constraints are respected by construction; the time-reversal is formulated to invert the forward noising dynamics exactly when the noising controls are realizable. For approximately realizable controls the manuscript relies on the numerical demonstrations in §6. A rigorous stability or reachability analysis for the approximate case is not provided. In the revision we will insert a clarifying paragraph that states the limitation explicitly and points to the empirical evidence already present in the paper. This is a partial revision. revision: partial

Circularity Check

0 steps flagged

No significant circularity; framework derives from standard continuity equation without self-referential reduction.

full rationale

The paper begins from the established continuity equation for control-affine systems and constructs exact or approximate flow-matching interpolations, extending to output matching and a noising/time-reversal view inspired by diffusion models. These steps use regression for approximate cases and PMP for noising, but the derivations do not reduce the imported control tools or regression-based controllers to quantities defined by the result itself. The central claims rest on independent constructions from classical PDEs and optimal control, with no load-bearing self-citations or fitted inputs presented as predictions. This is a self-contained proposal of a framework rather than a closed tautological loop.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The approach rests on the standard domain assumption that the system obeys control-affine dynamics whose continuity equation can be used to construct measure interpolations.

axioms (1)
  • domain assumption The system obeys control-affine dynamics of the form dx/dt = f_0(x) + sum_{i=1}^m u_i f_i(x)
    Explicitly stated as the starting point for constructing the continuity equation and flow matching.

pith-pipeline@v0.9.0 · 5831 in / 1181 out tokens · 29389 ms · 2026-05-18T10:57:44.759830+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

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

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

Reference graph

Works this paper leans on

45 extracted references · 45 canonical work pages · 3 internal anchors

  1. [1]

    Cambridge University Press, 2019

    Andrei Agrachev, Davide Barilari, and Ugo Boscain.A comprehensive introduction to sub- Riemannian geometry, volume 181. Cambridge University Press, 2019

    work page 2019

  2. [2]

    Optimal transportation under nonholonomic constraints

    Andrei Agrachev and Paul Lee. Optimal transportation under nonholonomic constraints. Transactions of the American Mathematical Society, 361(11):6019–6047, 2009

    work page 2009

  3. [3]

    Any sub-riemannian metric has points of smoothness

    Andrei A Agrachev. Any sub-riemannian metric has points of smoothness. InDoklady Mathe- matics, volume 79, pages 45–47. Springer, 2009

    work page 2009

  4. [4]

    Stochastic Interpolants: A Unifying Framework for Flows and Diffusions

    Michael S Albergo, Nicholas M Boffi, and Eric Vanden-Eijnden. Stochastic interpolants: A unifying framework for flows and diffusions.arXiv preprint arXiv:2303.08797, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  5. [5]

    Continuity equations and ode flows with non-smooth velocity.Proceedings of the Royal Society of Edinburgh Section A: Mathematics, 144(6):1191–1244, 2014

    Luigi Ambrosio and Gianluca Crippa. Continuity equations and ode flows with non-smooth velocity.Proceedings of the Royal Society of Edinburgh Section A: Mathematics, 144(6):1191–1244, 2014

    work page 2014

  6. [6]

    Sub-riemannianstructuresongroupsofdiffeomorphisms

    SylvainArguillereandEmmanuelTrélat. Sub-riemannianstructuresongroupsofdiffeomorphisms. Journal of the Institute of Mathematics of Jussieu, 16(4):745–785, 2017

    work page 2017

  7. [7]

    Lectures on young measure theory and its applications in economics

    Eik J Balder. Lectures on young measure theory and its applications in economics. 1998

    work page 1998

  8. [8]

    Springer Science & Business Media, 2012

    Viorel Barbu and Teodor Precupanu.Convexity and optimization in Banach spaces. Springer Science & Business Media, 2012. 26

    work page 2012

  9. [9]

    Young measures, superposition and transport.Indiana University mathematics journal, pages 247–275, 2008

    Patrick Bernard. Young measures, superposition and transport.Indiana University mathematics journal, pages 247–275, 2008

    work page 2008

  10. [10]

    Number 164

    Vladimir Igorevich Bogachev.Differentiable measures and the Malliavin calculus. Number 164. American Mathematical Soc., 2010

    work page 2010

  11. [11]

    Convex analytic methods in markov decision processes

    Vivek S Borkar. Convex analytic methods in markov decision processes. InHandbook of Markov Decision Processes: Methods and Applications, pages 347–375. Springer, 2002

    work page 2002

  12. [12]

    Semiconcavity results for optimal control problems admitting no singular minimizing controls

    P Cannarsa and L Rifford. Semiconcavity results for optimal control problems admitting no singular minimizing controls. InAnnales de l’IHP Analyse non linéaire, volume 25, pages 773–802, 2008

    work page 2008

  13. [13]

    Optimal transport over a linear dynamical system.IEEE Transactions on Automatic Control, 62(5):2137–2152, 2016

    Yongxin Chen, Tryphon T Georgiou, and Michele Pavon. Optimal transport over a linear dynamical system.IEEE Transactions on Automatic Control, 62(5):2137–2152, 2016

    work page 2016

  14. [14]

    Benamou-brenier and kantorovich are equivalent on sub-riemannian manifolds with no abnormal geodesics.arXiv preprint arXiv:2507.20959, 2025

    Giovanna Citti, Mattia Galeotti, and Andrea Pinamonti. Benamou-brenier and kantorovich are equivalent on sub-riemannian manifolds with no abnormal geodesics.arXiv preprint arXiv:2507.20959, 2025

    work page arXiv 2025

  15. [15]

    A blob method for mean field control with terminal constraints.ESAIM: Control, Optimisation and Calculus of Variations, 31:20, 2025

    Katy Craig, Karthik Elamvazhuthi, and Harlin Lee. A blob method for mean field control with terminal constraints.ESAIM: Control, Optimisation and Calculus of Variations, 31:20, 2025

    work page 2025

  16. [16]

    Domingo-Enrich, M

    Carles Domingo-Enrich, Michal Drozdzal, Brian Karrer, and Ricky TQ Chen. Adjoint matching: Fine-tuning flow and diffusion generative models with memoryless stochastic optimal control. arXiv preprint arXiv:2409.08861, 2024

    work page arXiv 2024

  17. [17]

    Approximate and exact controllability of the continuity equation with a localized vector field.SIAM Journal on Control and Optimization, 57(2):1284–1311, 2019

    Michel Duprez, Morgan Morancey, and Francesco Rossi. Approximate and exact controllability of the continuity equation with a localized vector field.SIAM Journal on Control and Optimization, 57(2):1284–1311, 2019

    work page 2019

  18. [18]

    On the equivalence between static and dynamic optimal transport governed by linear control systems.arXiv preprint arXiv:2505.17570, 2025

    Amit Einav, Yue Jiang, and Alpár R Mészáros. On the equivalence between static and dynamic optimal transport governed by linear control systems.arXiv preprint arXiv:2505.17570, 2025

    work page arXiv 2025

  19. [19]

    Benamou-brenier formulation of optimal transport for nonlinear control systems on rd.arXiv preprint arXiv:2407.16088, 2024

    Karthik Elamvazhuthi. Benamou-brenier formulation of optimal transport for nonlinear control systems on rd.arXiv preprint arXiv:2407.16088, 2024

    work page arXiv 2024

  20. [20]

    Score matching diffusion based feedback control and planning of nonlinear systems.arXiv preprint arXiv:2504.09836, 2025

    Karthik Elamvazhuthi, Darshan Gadginmath, and Fabio Pasqualetti. Score matching diffusion based feedback control and planning of nonlinear systems.arXiv preprint arXiv:2504.09836, 2025

    work page arXiv 2025

  21. [21]

    Optimal transport over nonlinear systems via infinitesimal generators on graphs.Journal of Computational Dynamics, 5(1&2):1–32, 2018

    Karthik Elamvazhuthi and Piyush Grover. Optimal transport over nonlinear systems via infinitesimal generators on graphs.Journal of Computational Dynamics, 5(1&2):1–32, 2018

    work page 2018

  22. [22]

    Optimal transport of linear systems over equilibrium measures.Automatica, 175:112222, 2025

    Karthik Elamvazhuthi and Matt Jacobs. Optimal transport of linear systems over equilibrium measures.Automatica, 175:112222, 2025

    work page 2025

  23. [23]

    Dynamical optimal transport of nonlinear control-affine systems.Journal of Computational Dynamics, 10(4):425–449, 2023

    Karthik Elamvazhuthi, Siting Liu, Wuchen Li, and Stanley Osher. Dynamical optimal transport of nonlinear control-affine systems.Journal of Computational Dynamics, 10(4):425–449, 2023

    work page 2023

  24. [24]

    Mass transportation on sub-riemannian manifolds.Geometric and functional analysis, 20(1):124–159, 2010

    Alessio Figalli and Ludovic Rifford. Mass transportation on sub-riemannian manifolds.Geometric and functional analysis, 20(1):124–159, 2010

    work page 2010

  25. [25]

    Springer Science & Business Media, 2012

    Wendell H Fleming and Raymond W Rishel.Deterministic and stochastic optimal control, volume 1. Springer Science & Business Media, 2012

    work page 2012

  26. [26]

    Score matching for sub-riemannian bridge sampling.arXiv preprint arXiv:2404.15258, 2024

    Erlend Grong, Karen Habermann, and Stefan Sommer. Score matching for sub-riemannian bridge sampling.arXiv preprint arXiv:2404.15258, 2024

    work page arXiv 2024

  27. [27]

    Mimicking the one-dimensional marginal distributions of processes having an itô differential.Probability theory and related fields, 71(4):501–516, 1986

    István Gyöngy. Mimicking the one-dimensional marginal distributions of processes having an itô differential.Probability theory and related fields, 71(4):501–516, 1986. 27

    work page 1986

  28. [28]

    The linear programming approach to deterministic optimal control problems.Applicationes Mathematicae, 24(1):17–33, 1996

    Daniel Hernández-Hernández, Onésimo Hernández-Lerma, and Michael Taksar. The linear programming approach to deterministic optimal control problems.Applicationes Mathematicae, 24(1):17–33, 1996

    work page 1996

  29. [29]

    Mass transportation with lq cost functions

    Ahed Hindawi, J-B Pomet, and Ludovic Rifford. Mass transportation with lq cost functions. Acta applicandae mathematicae, 113(2):215–229, 2011

    work page 2011

  30. [30]

    Entropic model predictive optimal transport over dynamical systems.Automatica, 152:110980, 2023

    Kaito Ito and Kenji Kashima. Entropic model predictive optimal transport over dynamical systems.Automatica, 152:110980, 2023

    work page 2023

  31. [31]

    Springer, 2014

    Frédéric Jean.Control of nonholonomic systems: from sub-Riemannian geometry to motion planning. Springer, 2014

    work page 2014

  32. [32]

    Nonlinear optimal control via occupation measures and lmi-relaxations.SIAM journal on control and optimization, 47(4):1643–1666, 2008

    Jean B Lasserre, Didier Henrion, Christophe Prieur, and Emmanuel Trélat. Nonlinear optimal control via occupation measures and lmi-relaxations.SIAM journal on control and optimization, 47(4):1643–1666, 2008

    work page 2008

  33. [33]

    Flow matching for generative modeling

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le. Flow matching for generative modeling. InThe Eleventh International Conference on Learning Representations

    work page

  34. [34]

    PhD thesis, PhD thesis, Universitá degli Studi di Pavia, 2006

    Stefano Lisini.Absolutely continuous curves in Wasserstein spaces with applications to continuity equation and to nonlinear diffusion equations. PhD thesis, PhD thesis, Universitá degli Studi di Pavia, 2006. 4.2. 1, 2006

    work page 2006

  35. [35]

    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

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  36. [36]

    Convex optimiza- tion of nonlinear feedback controllers via occupation measures.The International Journal of Robotics Research, 33(9):1209–1230, 2014

    Anirudha Majumdar, Ram Vasudevan, Mark M Tobenkin, and Russ Tedrake. Convex optimiza- tion of nonlinear feedback controllers via occupation measures.The International Journal of Robotics Research, 33(9):1209–1230, 2014

    work page 2014

  37. [37]

    Flow matching for stochastic linear control systems

    Yuhang Mei, Mohammad Al-Jarrah, Amirhossein Taghvaei, and Yongxin Chen. Flow matching for stochastic linear control systems. In7th Annual Learning for Dynamics\& Control Conference, pages 484–496. PMLR, 2025

    work page 2025

  38. [38]

    A time-reversal control synthesis for steering the state of stochastic systems.arXiv preprint arXiv:2504.00238, 2025

    Yuhang Mei, Amirhossein Taghvaei, and Ali Pakniyat. A time-reversal control synthesis for steering the state of stochastic systems.arXiv preprint arXiv:2504.00238, 2025

    work page arXiv 2025

  39. [39]

    Ot-flow: Fast and accurate continuous normalizing flows via optimal transport

    Derek Onken, Samy Wu Fung, Xingjian Li, and Lars Ruthotto. Ot-flow: Fast and accurate continuous normalizing flows via optimal transport. InProceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 9223–9232, 2021

    work page 2021

  40. [40]

    Morse-sard type results in sub-riemannian geometry

    Ludovic Rifford and Emmanuel Trélat. Morse-sard type results in sub-riemannian geometry. Mathematische Annalen, 332(1):145–159, 2005

    work page 2005

  41. [41]

    On the stabilization problem for nonholonomic distribu- tions.Journal of the European Mathematical Society, 11(2):223–255, 2009

    Ludovic Rifford and Emmanuel Trélat. On the stabilization problem for nonholonomic distribu- tions.Journal of the European Mathematical Society, 11(2):223–255, 2009

    work page 2009

  42. [42]

    Optimal transport for applied mathematicians

    Filippo Santambrogio. Optimal transport for applied mathematicians. 2015

    work page 2015

  43. [43]

    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

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  44. [44]

    Springer, 2008

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

    work page 2008

  45. [45]

    Duality-based dynamical optimal transport of discrete time systems.arXiv preprint arXiv:2410.09801, 2024

    Dongjun Wu and Anders Rantzer. Duality-based dynamical optimal transport of discrete time systems.arXiv preprint arXiv:2410.09801, 2024. 28

    work page arXiv 2024