Pith. sign in

REVIEW 3 major objections 4 minor 56 references

Distribution Steering via Sliced Optimal Transport Control

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Sliced optimal transport can be realized as feedback laws that steer a system's law to a prescribed target using only one-dimensional projections.

desk verdict Solid sliced-OT control framework with a real gap: Theorem 3 doesn't cover the atomic targets used in its own numerics, but the Gaussian and linear-system results stand. read the letter →

arxiv 2608.12828 v1 pith:NKNQBFX3 submitted 2026-08-13 math.OC cs.LGcs.SYeess.SYstat.ML

Kaito Ito , Anqi Dong This is my paper · ORCID
classification math.OCcs.LGcs.SYeess.SYstat.ML MSC 49Q2293B0593C05
keywords slicedoptimaltransportdistributionsteeringfeedbackcontrolWassersteindistanceGaussianpreservationcontinuityequationreachabilityGramianrandomizedsampling
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that distribution steering—driving the probability law of a dynamical system from one prescribed distribution to another—can be carried out with sliced optimal transport, using only one-dimensional projections of the current and target laws. The key move is to read each projected optimal transport map as a directional terminal condition, lift its minimum-energy realization back to the ambient state space, and average over all projection directions. For single-integrator dynamics the resulting averaged feedback makes the sliced Wasserstein distance to the target non-increasing; for Gaussian endpoint laws the feedback is affine, preserves Gaussianity, and under a uniform covariance lower-bound condition drives the mean and covariance to their target values. The paper also proves that the randomized single-direction iteration converges to this averaged flow with expected squared Wasserstein error $O(h)$ as the sampling period vanishes, and extends the construction to linear systems through reachability-normalized coordinates and local controllability Gramians. A reader should care because this offers a sample-friendly, feedback-realizable alternative to full-dimensional optimal transport for steering distributions.

What carries the argument

The load-bearing object is the sliced discrepancy field $g_t(x)=\int_{S^{n-1}}(\theta^{\top}x-T^{\theta}_t(\theta^{\top}x))\,\theta\,\sigma(d\theta)$, computed from one-dimensional monotone optimal transport maps $T^{\theta}_t$ between projected current and projected target laws. Its averaged square $D(t)=\int\|g_t(x)\|^2\,\rho_t(dx)$ appears in the dissipation identity, and its directional versions define the randomized update. In the Gaussian case the field specializes to the matrix function $H(\Sigma)=\int_{S^{n-1}}\alpha(\Sigma,\theta)\theta\theta^{\top}\sigma(d\theta)-\frac{1}{n}I$, where $\alpha(\Sigma,\theta)=\sqrt{\theta^{\top}\Sigma_1\theta/\theta^{\top}\Sigma\theta}$; the covariance equation $\dot\Sigma=K\Sigma+\Sigma K^{\top}$ with $K=\lambda H$ preserves Gaussianity and controls terminal convergence. A law-dependent gain $\lambda_{\mathrm{SW}}(t)=n/(\chi(t)(1-t))$ compensates directional cancellation and produces linear decay of the sliced Wasserstein distance. For linear systems, reachability-normalized coordinates $z=LF_t x$ and local controllability Gramians $G_k$ translate the sliced velocity into an exact finite-step input.

What would settle it

Run the averaged sliced feedback on two Gaussians in $\mathbb{R}^2$ with initial covariance $10^{-6}I$ and target covariance $I$, using $\lambda(t)=(1-t)^{-1}$, and monitor the smallest eigenvalue of $\Sigma(t)$; if it dips below any fixed $\alpha>0$ before $t=1$ and the sliced Wasserstein distance stops following the exponential bound (19), the uniform lower-bound hypothesis is the active obstruction. Independently, on the three-component Gaussian mixture of Section 7, estimate $\mathbb{E}[W_2^2(\rho^h_k,\rho_{t_k})]$ for shrinking $h$; if the ratio to $h$ diverges, Assumption 2 does not hold for that case and the $O(h)$ convergence theorem does not apply.

Watch

Extended reading notes

Core claim

At each sampling instant, choose a direction $\theta\in S^{n-1}$, compute the monotone optimal transport map $T^{\theta}_t$ between the projected current law and the projected target law, and solve a scalar minimum-energy problem that sends $\theta^{\top}x$ to $T^{\theta}_t(\theta^{\top}x)$ over the remaining horizon. Lifting that scalar control back to $\mathbb{R}^n$ gives a single-direction controller; averaging it over all directions with the spherical measure yields the deterministic averaged sliced feedback $v(t,x)=-\lambda(t)g_t(x)$, where $g_t(x)$ is the sliced discrepancy field. The paper establishes that, along this feedback and the continuity equation, the sliced Wasserstein distance obeys $\frac{d}{dt}\mathrm{SW}_2^2(\rho_t,\rho_1)=-2\lambda(t)\int\|g_t\|^2\,d\rho_t$, so it never increases. For Gaussian endpoints the field is affine, the flow preserves Gaussianity, and with $\int_0^1\lambda=\infty$ plus $\Sigma(t)\succeq\alpha I$ the mean and covariance converge to the prescribed terminal values, giving exponential decay of the sliced Wasserstein distance. The randomized sampled-direction iteration is shown to converge to the averaged sliced flow in expectation with squared Wasserstein error $O(h)$ as the sampling period $h\to 0$.

Load-bearing premise

The load-bearing premise is that the direction-averaged force field stays Lipschitz stable along both the random and averaged flows and that the Gaussian covariance remains uniformly bounded away from zero on the whole horizon; the paper states these as assumptions without verifying them for concrete distributions, and the terminal-convergence and $O(h)$-rate claims collapse if either fails.

Editorial extensions

If this is right

  • The sliced controller reduces the per-iteration cost of distribution steering from solving an $N\times N$ assignment problem to sorting $N_{\mathrm{dir}}$ one-dimensional projections: $O(K N_{\mathrm{dir}} N\log N)$ operations and $O(N_{\mathrm{dir}} N)$ memory over the full horizon.
  • Whenever the Gaussian terminal-convergence conditions hold, the same affine feedback steers the mean and covariance of any law with finite second moment to the prescribed values, not only Gaussian laws.
  • For uniformly fully actuated linear systems, the physical control energy is sandwiched between $E_z/q_+$ and $E_z/q_-$, where $E_z$ is the kinetic energy of the transformed sliced flow and $q_-$, $q_+$ are uniform actuation bounds.
  • For general controllable systems, applying the local-Gramian input (50) reproduces the virtual sliced iteration exactly at the sampling instants, so discretization error in the law appears only through the choice of sliced update map, not through the linear dynamics.
  • With the gain $\lambda_{\mathrm{SW}}$, the sliced Wasserstein distance decays exactly linearly, $r(t)=(1-t)r(0)$, and the total control energy obeys $n\,\mathrm{SW}_2^2(\rho_0,\rho_1)\le E_u(0,1)\le (n/\chi)\,\mathrm{SW}_2^2(\rho_0,\rho_1)$ whenever $\chi(t)\ge\chi>0$.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An implication the paper leaves implicit is that the covariance lower-bound condition in Theorem 1 is not implied by Proposition 2's strict positivity; one could test Gaussian examples with a near-degenerate initial covariance, where the exponential-decay guarantee may fail even though the mean still converges.
  • The $O(h)$ convergence proof treats projection directions as i.i.d.; using stratified or antipodal direction sampling should reduce the centered fluctuation variance and might remove the $1/n$ factor in the sliced-Wasserstein bound, a testable modification of the algorithm.
  • For the finite-step realization, shrinking the partition can make the local Gramians ill-conditioned even when they stay positive definite; the paper explicitly notes no finite-energy continuous-time limit should be expected, which suggests a practical trade-off between step count and numerical conditioning that is not quantified.
  • One could extend the same 'directional terminal condition' reading to nonlinear or stochastic dynamics by re-solving the projected endpoint problem in a moving horizon; the paper lists this direction implicitly in its conclusion.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper develops a finite-horizon feedback framework for steering the probability law of a linear dynamical system between prescribed endpoint distributions, using only one-dimensional optimal transport maps along random projections. For the single-integrator dynamics, averaging the directional controllers over the sphere yields a deterministic sliced feedback. Proposition 1 derives a dissipation identity for the sliced Wasserstein distance; Proposition 2 shows that this averaged feedback is affine and preserves Gaussianity; Theorem 1 gives terminal convergence under a uniform covariance lower bound; Theorem 2 constructs a law-dependent gain with linear decay and explicit energy bounds; and Theorem 3 proves O(h) convergence of the randomized controller to the averaged flow under Assumptions 1 and 2. Sections 5 and 6 extend the construction to fully actuated and general controllable linear systems, and Section 7 presents numerical experiments for Gaussian mixtures and image color transfer.

Significance. The paper is a genuinely useful contribution to distribution steering: it replaces full-dimensional transport maps with projected one-dimensional maps, provides a clean dissipation identity, gives an exact affine Gaussian flow with explicit decay and energy bounds, and develops realizations over linear dynamics. The appendix proofs are detailed, and the paper is honest about some limitations, notably in Section 6.2 where it explicitly states that partition refinement of the finite-step realization need not converge. The main caveat is that the central randomized-convergence theorem rests on a stability assumption that is not verified for any non-Gaussian class or for the atomic empirical targets actually used in the numerical experiments; a revision should either verify that assumption in a relevant class or restrict the statement of Theorem 3 accordingly.

major comments (3)
  1. [Section 4, Assumption 2 and Theorem 3]
  2. [Section 3.4, Theorem 1 and Proposition 2]
  3. [Section 6, Proposition 6 and Algorithm 1]
minor comments (4)
  1. [Section 7.1]
  2. [Section 7.2, Figure 6]
  3. [Notation, Definition 1]
  4. [Corollary 1]

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: core results are derived from OT definitions and explicit design choices; the unverified Lipschitz assumption is a gap, not a circular reduction.

full rationale

The derivation chain is self-contained. Proposition 1 obtains the variation formula d/dt SW2^2 = 2∫g_t·v from the one-dimensional Wasserstein derivative averaged over projections, not from the conclusion being proved; the averaged feedback v = -λ g_t then gives the dissipation identity by direct substitution. Gaussian preservation follows by computing g_t for Gaussian laws and solving the resulting affine mean/covariance equations. Theorem 1 is a Gronwall/coercivity argument, and Theorem 2 is an explicit control-design construction: λ_SW is defined as r^2/(D(1-t)) so the linear decay r(t)=(1-t)r(0) follows by algebra, but the paper presents this as a gain design rather than as an empirical prediction. Theorem 3 is conditional on Assumption 2, which is postulated rather than verified and may fail for the atomic target laws used in the numerical experiments; however, an unproved or false regularity hypothesis is an analytic gap, not a circular reduction of the conclusion to the input. The only self-citation, [ID26], is cited as a Gaussian-only precursor, and no load-bearing proof step reduces to it. No fitted parameter is renamed as a prediction, and no ansatz is smuggled in via citation: the averaged field is derived from the definition of sliced optimal transport and the minimum-energy realization of projected endpoint conditions.

Assumptions & free parameters 3 free parameters · 7 assumptions · 0 invented entities

The paper introduces no new physical entities. Its central claims rest on standard optimal transport background, explicit regularity assumptions, and two stability assumptions that are stated but not verified for concrete distributions. The main free parameters are the gain, the sampling grid, and the finite-step step size, none of which are fitted to data.

free parameters (3)
  • Gain function lambda(t) = (1-t)^-1, lambda_SW(t)=r(t)^2/(D(t)(1-t)), or another user-chosen nonnegative function
    The gain is a design knob controlling decay rate and control energy. Theorem 1 requires a divergent integral of lambda, and Theorem 2 defines lambda_SW from the current flow, so the central rates depend on this choice.
  • Sampling period h and partition {t_k} = h -> 0 in Theorem 3; logarithmically refined grid and 60 uniform steps in the numerical examples
    The randomized controller is analyzed in the limit h to 0, and the finite-step realization in Section 6 depends on a user-selected partition. The numerical examples choose specific grids.
  • Step size alpha_k in the finite-step sliced update = Unspecified in Algorithm 1
    The update map in Section 6.1 uses a step size alpha_k chosen by the user. The paper does not give a convergence rate for the virtual sliced iteration, only exact finite-step displacement matching.
assumptions (7)
  • domain assumption Initial law is absolutely continuous and the target law has finite second moment, rho0 in P_{2,ac}, rho1 in P_2.
    This is the standing well-posedness setting for one-dimensional optimal transport maps and finite sliced Wasserstein distances.
  • domain assumption The controlled density satisfies the continuity equation in the weak sense with finite kinetic energy on compact subintervals.
    Used in Proposition 1 and throughout the averaged-flow analysis to justify absolute continuity of the sliced Wasserstein distance.
  • domain assumption Assumption 1: a Lagrangian averaged flow exists on [0, t-bar].
    This is an existence assumption for the comparison flow used in Theorem 3. It is stated but not proven for any specific distribution class.
  • domain assumption Assumption 2: Lipschitz stability of the averaged sliced field along the iterative and averaged flows.
    This is the load-bearing condition for the randomized-to-average convergence theorem, and it is not verified for any concrete dynamics.
  • domain assumption Theorem 1 assumes Sigma(t) >= alpha I uniformly on [0,1).
    The paper proves strict positivity of Sigma(t) but not a uniform lower bound; this extra condition is needed for the exponential decay constant and terminal convergence.
  • domain assumption Lemma 1 assumes D(t)>0 whenever r(t)>0.
    This rules out exact cancellation of directional sliced displacements before the target is reached. The paper proves it for Gaussian flows but not in general.
  • domain assumption Assumption 3: B(t) continuous and full row rank for fully actuated systems; Assumption 4: local controllability on each partition interval.
    Both are needed for the reachability-normalized realization and the finite-step realization in Sections 5 and 6.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Distribution Steering via Sliced Optimal Transport Control." pith.science (2026). https://pith.science/paper/NKNQBFX3

@misc{pith2026260812828,
  author       = {Pith},
  title        = {Pith review of: Distribution Steering via Sliced Optimal Transport Control},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NKNQBFX3}},
  note         = {Machine review of arXiv:2608.12828}
}
read the original abstract

Distribution steering seeks feedback laws that drive the state law of a dynamical system between prescribed initial and terminal distributions. Optimal transport provides a natural geometric approach, but its implementation generally requires a transport map or coupling in the full state space. Sliced optimal transport avoids this full-dimensional construction through one-dimensional projections. Yet, the resulting projected maps specify only directional displacements and do not by themselves prescribe a realizable feedback law. To this end, we develop a finite-horizon control framework based on sliced optimal transport. At each sampling instant, a projected optimal transport map defines a directional terminal condition, whose minimum-energy realization yields a randomized single-direction controller. Averaging over projection directions gives a deterministic sliced feedback. For the single-integrator dynamics, the averaged feedback makes the sliced Wasserstein distance to the target non-increasing. For Gaussian endpoint laws, it is affine, preserves Gaussianity, and steers the mean and covariance to their prescribed terminal values. We further identify a law-dependent gain that yields linear decay of the sliced Wasserstein distance together with an explicit characterization of the control energy. We also prove that the randomized controller converges to the averaged sliced flow as the sampling period vanishes. Finally, we extend the construction to linear dynamical systems. Reachability-normalized coordinates allow instantaneous realization of the sliced velocity for uniformly fully actuated systems, while local controllability Gramians provide exact finite-step realization for general controllable systems. Numerical examples illustrate the resulting distributional flows.

Figures

Figures reproduced from arXiv: 2608.12828 by the authors.

Figure 1
Figure 1. Randomized sliced steering between two Gaussian mixtures. Thin curves show [PITH_FULL_IMAGE:figures/full_fig_p023_1.png] view at source ↗
Figure 2
Figure 2. Empirical sliced Wasserstein distance to the target over time, estimated using [PITH_FULL_IMAGE:figures/full_fig_p023_2.png] view at source ↗
Figure 3
Figure 3. Initial and target paintings and their empirical color distributions in the ( [PITH_FULL_IMAGE:figures/full_fig_p024_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Color-distribution steering through fully actuated linear dynamics. Snapshots are [PITH_FULL_IMAGE:figures/full_fig_p025_4.png]
Figure 5
Figure 5. Figure 5: Accumulated physical input energy and the corresponding lower and upper bounds. [PITH_FULL_IMAGE:figures/full_fig_p025_5.png]
Figure 6
Figure 6. Figure 6: Runtime and memory scaling of sliced and exact optimal transport. Solid lines [PITH_FULL_IMAGE:figures/full_fig_p026_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 44 canonical work pages

  1. [1]

    Proceedings of the 39th Annual Conference on Learning Theory , pages=

    Convergence Rates for Distribution Matching with Sliced Optimal Transport , author=. Proceedings of the 39th Annual Conference on Learning Theory , pages=. 2026 , publisher=

    work page 2026

  2. [2]

    arXiv preprint arXiv:2307.05705 , year=

    Measure transfer via stochastic slicing and matching , author=. arXiv preprint arXiv:2307.05705 , year=

    arXiv

  3. [3]

    Ito, Kaito and Dong, Anqi , booktitle=. Sliced. 2026 , organization=

    work page 2026

  4. [4]

    Ordinary

    Teschl, Gerald , year=. Ordinary

    work page

  5. [5]

    Duistermaat, Johannes Jisse and Kolk, Johan A.C. , year=

    work page

  6. [6]

    Rachev, Svetlozar T. and R. Mass. 1998 , publisher=

    work page 1998

  7. [7]

    2003 , publisher=

    Villani, C. 2003 , publisher=

    work page 2003

  8. [8]

    2005 , publisher=

    Ambrosio, Luigi and Gigli, Nicola and Savar. 2005 , publisher=

    work page 2005

Show all 56 references
  1. [9]

    , booktitle=

    Deshpande, Ishan and Hu, Yuan-Ting and Sun, Ruoyu and Pyrros, Ayis and Siddiqui, Nasir and Koyejo, Sanmi and Zhao, Zhizhen and Forsyth, David and Schwing, Alexander G. , booktitle=

  2. [10]

    Brezis, Haim , year=

  3. [11]

    Econometrica , volume=

    Envelope theorems for arbitrary choice sets , author=. Econometrica , volume=. 2002 , publisher=

    2002

  4. [12]

    Journal of the London Mathematical Society , volume=

    Some theorems on distribution functions , author=. Journal of the London Mathematical Society , volume=. 1936 , publisher=

    1936

  5. [13]

    The Annals of Probability , volume=

    When is a probability measure determined by infinitely many projections? , author=. The Annals of Probability , volume=. 1997 , publisher=

    1997

  6. [14]

    Computational optimal transport:

    Peyr. Computational optimal transport:. Foundations and Trends. 2019 , publisher=

    2019

  7. [15]

    and Pavon, Michele , journal=

    Chen, Yongxin and Georgiou, Tryphon T. and Pavon, Michele , journal=. On the relation between optimal transport and. 2016 , publisher=

    2016

  8. [16]

    arXiv preprint arXiv:2508.01243 , year=

    Sliced Optimal Transport Plans , author=. arXiv preprint arXiv:2508.01243 , year=

    arXiv

  9. [17]

    Inverse Problems , volume=

    Sliced optimal transport on the sphere , author=. Inverse Problems , volume=. 2023 , publisher=

    2023

  10. [18]

    ACM Trans

    Sliced optimal transport sampling , author=. ACM Trans. Graph. , volume=

  11. [19]

    Thummerer, Adrian and van der Bijl, Erik and Galapon, Arthur Jr and Kamp, Florian and Savenije, Mark and Muijs, Christina and Aluwini, Shafak and Steenbakkers, Roel JHM and Beuel, Stephanie and Intven, Martijn PW and others , journal=. Synth. 2025 , publisher=

    2025

  12. [20]

    A Multi-Centric Anthropomorphic 3

    Amirian, Mohammadreza and Bach, Michael and Jimenez-del-Toro, Oscar and Aberle, Christoph and Schaer, Roger and Andrearczyk, Vincent and Maestrati, Jean-F. A Multi-Centric Anthropomorphic 3. arXiv preprint arXiv:2507.01539 , year=

    arXiv

  13. [21]

    IEEE Control Systems Letters , volume=

    Entropic Model Predictive Optimal Transport for Underactuated Linear Systems , author=. IEEE Control Systems Letters , volume=. 2023 , publisher=

    2023

  14. [22]

    Long-time asymptotics of the sliced-

    Cozzi, Giacomo and Santambrogio, Filippo , journal=. Long-time asymptotics of the sliced-. 2025 , publisher=

    2025

  15. [23]

    Liutkus, Antoine and Simsekli, Umut and Majewski, Szymon and Durmus, Alain and St. Sliced-. International Conference on Machine Learning , pages=. 2019 , organization=

    2019

  16. [24]

    Towards Understanding Gradient Dynamics of the Sliced-

    Christophe Vauthier and Anna Korba and Quentin M. Towards Understanding Gradient Dynamics of the Sliced-. Forty-second International Conference on Machine Learning , year=

  17. [25]

    and Mayne, David Q

    Rawlings, James B. and Mayne, David Q. and Diehl, Moritz M. , year =

  18. [26]

    Automatica , volume=

    Entropic model predictive optimal transport over dynamical systems , author=. Automatica , volume=. 2023 , publisher=

    2023

  19. [27]

    Computer Vision and Image Understanding , volume=

    Automated colour grading using colour distribution transfer , author=. Computer Vision and Image Understanding , volume=. 2007 , publisher=

    2007

  20. [28]

    Folland, Gerald B. , year=

  21. [29]

    IEEE Transactions on Automatic Control , volume=

    Optimal transport over a linear dynamical system , author=. IEEE Transactions on Automatic Control , volume=. 2017 , publisher=

    2017

  22. [30]

    2000 , publisher=

    Benamou, Jean-David and Brenier, Yann , journal=. 2000 , publisher=

    2000

  23. [31]

    2013 , publisher=

    Axler, Sheldon and Bourdon, Paul and Wade, Ramey , volume=. 2013 , publisher=

    2013

  24. [32]

    2008 , publisher=

    Ba. 2008 , publisher=

    2008

  25. [33]

    Danskin, John M. , year=. The

  26. [34]

    2008 , edition=

    Ambrosio, Luigi and Gigli, Nicola and Savar. 2008 , edition=

    2008

  27. [35]

    Wasserstein barycenter and its application to texture mixing , author=. Proc. International Conference on Scale Space and Variational Methods in Computer Vision , pages=. 2011 , organization=

    2011

  28. [36]

    Annual Review of Control, Robotics, and Autonomous Systems , volume=

    Optimal transport in systems and control , author=. Annual Review of Control, Robotics, and Autonomous Systems , volume=. 2021 , publisher=

    2021

  29. [37]

    and Pavon, Michele , journal=

    Chen, Yongxin and Georgiou, Tryphon T. and Pavon, Michele , journal=. Stochastic control liaisons:. 2021 , publisher=

    2021

  30. [38]

    and Tannenbaum, Allen , journal=

    Chen, Yongxin and Georgiou, Tryphon T. and Tannenbaum, Allen , journal=. Optimal transport for. 2019 , publisher=

    2019

  31. [39]

    Deans, Stanley R. , year=. The

  32. [40]

    International Journal of Computer Vision , volume=

    Optimal mass transport for registration and warping , author=. International Journal of Computer Vision , volume=. 2004 , publisher=

    2004

  33. [41]

    Systems & Control Letters , volume=

    Path integral control of partially observed systems via fully observable control approximations , author=. Systems & Control Letters , volume=. 2025 , publisher=

    2025

  34. [42]

    Monge, Gaspard , journal=. M

  35. [43]

    On the translocation of masses , author=. Dokl. Akad. Nauk. USSR (NS) , volume=

  36. [44]

    , journal=

    Dong, Anqi and Stephanovitch, Arthur and Georgiou, Tryphon T. , journal=. Monge--. 2024 , publisher=

    2024

  37. [45]

    Nguyen, Khai and Ho, Nhat , booktitle=. Sliced

  38. [46]

    Sliced and

    Bonneel, Nicolas and Rabin, Julien and Peyr. Sliced and. Journal of Mathematical Imaging and Vision , volume=. 2015 , publisher=

    2015

  39. [47]

    Acta Numerica , volume=

    Distributionally robust optimization , author=. Acta Numerica , volume=. 2025 , publisher=

    2025

  40. [48]

    Villani, C. Optimal. 2009 , publisher=

    2009

  41. [49]

    and Pavon, Michele , journal=

    Chen, Yongxin and Georgiou, Tryphon T. and Pavon, Michele , journal=. Optimal steering of a linear stochastic system to a final probability distribution,. 2016 , publisher=

    2016

  42. [50]

    Iterative

    Benamou, Jean-David and Carlier, Guillaume and Cuturi, Marco and Nenna, Luca and Peyr. Iterative. SIAM Journal on Scientific Computing , volume=. 2015 , publisher=

    2015

  43. [51]

    Fast computation of

    Cuturi, Marco and Doucet, Arnaud , booktitle=. Fast computation of. 2014 , organization=

    2014

  44. [52]

    European Journal of Applied Mathematics , volume=

    Optimal transport through a toll station , author=. European Journal of Applied Mathematics , volume=. 2025 , publisher=

    2025

  45. [53]

    and Tannenbaum, Allen , journal=

    Dong, Anqi and Georgiou, Tryphon T. and Tannenbaum, Allen , journal=. Data. 2025 , publisher=

    2025

  46. [54]

    IEEE Transactions on Automatic Control , volume=

    Maximum entropy density control of discrete-time linear systems with quadratic cost , author=. IEEE Transactions on Automatic Control , volume=. 2025 , publisher=

    2025

  47. [55]

    Convolutional

    Solomon, Justin and De Goes, Fernando and Peyr. Convolutional. ACM Transactions on Graphics (ToG) , volume=. 2015 , publisher=

    2015

  48. [56]

    2013 , school=

    Unidimensional and evolution methods for optimal transportation , author=. 2013 , school=

    2013

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.