Displacement-Sparse Neural Optimal Transport

Peter Chen; Qingpeng Zhang; Yue Xie

REVIEW 4 major objections 6 minor 60 references

Displacement-Sparse Neural Optimal Transport

T0 review · 4 major / 6 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Adding a smoothed ℓ0 penalty directly to the dual-potential objective of a neural OT solver produces transport maps whose displacement vector has far fewer active coordinates, while keeping the transported measure close to the target.

desk verdict A genuinely new biased-minimax formulation for displacement-sparse neural OT with a clever adaptive λ schedule, but the headline sparsity numbers may be partly a 10^-2 threshold artifact and the theory does not cover the actual penalty used. read the letter →

arxiv 2502.01889 v2 pith:PJNWLHPO submitted 2025-02-03 cs.LG cs.AI

Peter Chen , Yue Xie , Qingpeng Zhang This is my paper · ORCID

classification cs.LGcs.AI MSC 49Q2268T0790C08

keywords displacement-sparseoptimaltransportneuralinputconvexnetworksmoothedℓ0normadaptiveregularizationsimulatedannealingsingle-cellperturbationinterpretability

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

Optimal transport moves one distribution onto another; neural solvers scale to high-dimensional biology but learn dense maps that move every gene or feature, which is hard to interpret. This paper claims that sparsity can be imposed on these maps by adding a penalty $\$\lambda$ \int \tau(\nabla g(y)-y)\,dQ$ directly to the ICNN minimax objective, so the displacement vector $\nabla g(y)-y$ is encouraged to have few nonzero entries. The paper introduces a smoothed $\ell_0$ regularizer and an adaptive simulated-annealing schedule for the penalty strength $\$\lambda$$. On a synthetic drug-perturbation benchmark with 100 truly affected genes out of 3000, the adaptive method recovers average displacement dimension 107 instead of 2480 for unregularized ICNN OT, while keeping gene-overlap accuracy above 95%. This matters because interpretable, low-dimensional transport maps could make single-cell perturbation analysis more tractable without abandoning neural solvers.

What carries the argument

The load-bearing object is the dual potential $g$ from the ICNN minimax formulation, whose gradient $\nabla g(y)$ recovers the transport map; the displacement vector is $\nabla g(y)-y$. The paper's mechanism is to penalize that displacement vector inside the objective (Eq. 7) rather than through an elastic-cost proximal operator, which is what allows arbitrary penalties—including non-convex ones—to be used. The concrete regularizer is the smoothed $\ell_0$ norm $\|z\|_{0,\xi}=\sum_i (1-\exp(-$z_i^{2}$/(2\$xi^{2}$)))$, a smooth surrogate for counting nonzero coordinates. The adaptive component is a simulated-annealing controller that adjusts $\$\lambda$$ according to an evaluation function balancing sparsity and feasibility in low dimension, or according to a direct dimensionality constraint $l$ in high dimension.

What would settle it

Run the synthetic $k=100$ benchmark with the truly perturbed gene set shuffled after each restart; if the method still reports low displacement dimension but gene-overlap accuracy drops to chance, the claim that it recovers sparse structure rather than merely shrinking displacements is falsified. A second check: at a fixed high $\$\lambda$$, measure whether the displacement dimension can be pushed below the true $k$ without the Wasserstein divergence exploding—the theory predicts error growing linearly in $\$\lambda$$, so a dramatic superlinear blow-up would contradict it.

Watch

Extended reading notes

Core claim

The central claim is that displacement sparsity can be imposed on neural OT maps by regularizing the dual objective rather than the cost: adding $\$\lambda$ \int_{\mathbb{R}^d} \tau(\nabla g(y)-y)\,dQ$ to the minimax objective (7) and annealing $\$\lambda$$ during training yields maps whose displacement vector activates only a few coordinates while still transporting source to target. The paper further claims that a smoothed $\ell_0$ penalty beats the $\ell_1$-based alternative in this neural setting, and that adaptive control of $\$\lambda$$ is necessary to avoid early suboptimal convergence in large-scale training. Theoretically, it proves that the biased objective returns to the unbiased one as $\$\lambda$ \to 0$, that convex penalties produce sparser maps as $\$\lambda$$ grows, and that the map error is bounded by the optimality gaps plus $2\$\lambda$ M_\tau$, so the main cost of sparsity is controlled by $\$\lambda$$. Empirically, on the synthesized scRNA task with ground-truth dimension $k=100$, the adaptive method reaches dimension 107 versus 2480 for unregularized ICNN OT and 120 for the best constant $\$\lambda$$, with gene-overlap accuracy above 95%; on four real 4i drug datasets it reduces displacement dimensionality while keeping Wasserstein divergence comparable.

Load-bearing premise

The proofs assume the neural-network training comes very close to solving the optimal-transport problem before the sparsity penalty matters, and the main experiments use a smoothed-$\ell_0$ penalty for which the convergence and sparsity theorems are not proved because it is not convex.

Editorial extensions

If this is right

Neural OT solvers can now produce displacement-sparse maps with arbitrary sparsity penalties, not only the convex penalties that admit proximal operators.
The smoothed $\ell_0$ regularizer gives a direct handle on the number of active displacement coordinates, which is valuable when perturbed and noise genes are not separable by magnitude.
Adaptive $\lambda$ via simulated annealing removes manual tuning in high-dimensional training and reaches displacement dimensions closer to known ground truth than any fixed $\lambda$.
The map-error bound implies that once the ICNN training gap is small, users can trade a controlled amount of map accuracy for interpretability by choosing $\lambda$.

Reading between the lines

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

The same penalty-in-the-dual trick should extend to other structural priors on the transport map, such as group sparsity over gene modules or monotonicity constraints, since it only requires evaluating a term involving $\nabla g$.
The adaptive-$\lambda$ schedule suggests a general curriculum for regularized minimax training beyond optimal transport: anneal regularization intensity to avoid premature convergence to trivial equilibria.
Because the theory covers only convex penalties, a natural testable extension is to prove convergence for pseudoconvex penalties like smoothed $\ell_0$, or to replace it with a convex surrogate that still approximates the $\ell_0$ count.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Chen, Xie, Zhang propose a biased minimax formulation for ICNN-based OT that lets you add sparsity penalties directly to the dual objective, plus a simulated-annealing controller for λ. That is a real step beyond Cuturi et al. 2023, which requires proximal operators and exact Sinkhorn, and prior general-cost neural OT work that needed convexity. The smoothed ℓ0 regularizer is a sensible choice, and the gene-overlap accuracy in Table 3 is independent evidence that the selected displacements point at the right coordinates. The adaptive λ framework is practically motivated and the multiple-run trajectories in D.1 give it some credibility.

The soft spots, in proportion: the headline dimension claim is fragile. Dimension is defined, in D.3, by counting coordinates with |Δ_i| > 10^-2 via np.isclose. The smoothed ℓ0 norm has zero gradient at zero and never forces exact zeros, so a coordinate with a 5e-3 displacement is scored as 'inactive.' That means the 107-vs-2480 comparison may substantially overstate the gain in true support sparsity. The gene-overlap results soften this—if small displacements sit on the true genes, the method is still doing something right—but they don't rescue the absolute dimensionality numbers, and the comparison to Cuturi et al. (which produces genuine zeros) is not apples-to-apples.

The theory is honest but narrow. Theorems 4.4 and 4.5 are proven only for convex τ, and D.4 explicitly says the smoothed ℓ0 is pseudoconvex and excluded. That is an unusual and welcome disclosure, but it means the convergence and monotonicity results don't cover the actual regularizer used in the experiments. The proof of Theorem 4.3 has a genuine gap: the limsup is passed into V_P,Q(f_λ, g_0) without extracting a subsequence along which f_λ converges. That can likely be fixed, but as written it is not a complete proof. Theorem 4.6's bound is standard, conditional on optimality gaps.

No code is provided, which is a real limitation for a method paper whose practical appeal is a heuristic controller. The 4i results lack error bars in Figure 6, though the text says ten runs were averaged.

Who this is for: people doing neural OT for single-cell perturbation who care about interpretable maps. It is not a reshaping of OT theory, but it is a plausible, useful methodological contribution. I'd send it to review—the threshold artifact and the proof gap are fixable in revision, and the core idea deserves referee time. I would not cite the dimension numbers without checking what they mean; I might cite the formulation.

Referee Report

4 major / 6 minor

Summary. The paper proposes a regularized minimax formulation for neural optimal transport in which the standard ICNN dual objective is augmented by a penalty λ∫τ(∇g(y)−y)dQ on the displacement vector. This is intended to generalize the elastic-cost displacement-sparse OT of Cuturi et al. to neural solvers and to allow non-convex penalties such as a smoothed ℓ0 norm. The paper proves consistency as λ→0, convergence of regularized solutions to the unregularized solution set, monotonicity of sparsity for convex τ, and a map-error bound depending on optimality gaps and λ. It also introduces a simulated-annealing heuristic to adapt λ, with separate strategies for low- and high-dimensional tasks. Experiments on a synthetic scRNA-like task (d=3000, k=100) report displacement dimension 107 with adaptive λ versus 2480 for unregularized ICNN OT, and over 95% overlap with ground-truth genes; additional experiments on real 4i perturbation data compare the smoothed ℓ0 penalty to ℓ1 and to Cuturi et al.

Significance. If the results hold, the paper addresses a real limitation of neural OT solvers: learned maps can be high-dimensional and difficult to interpret, while exact OT approaches to displacement sparsity require proximal operators that exclude non-convex penalties. The proposed formulation is intuitive and the convex-penalty theory is a useful first step. The synthetic setup with known ground-truth dimension k and gene-overlap evaluation is a good experimental design, and the adaptive-λ scheme is practically relevant for users who need a target displacement dimension. The paper is generally clearly written and does a service by explicitly flagging in Appendix D.4 that its main theoretical guarantees only cover convex penalties and do not cover the smoothed ℓ0 penalty used in the experiments.

major comments (4)

[Appendix D.3, Table 1] The reported 'dimension' of the displacement is defined via np.isclose(.,0) with a threshold of 10^-2 (Appendix D.3), while the smoothed ℓ0 penalty used in the headline experiments is smooth at zero (σ=1.0, Appendix E) and therefore does not produce exact zeros. Consequently, the central comparison '107 vs 2480' may reflect that unregularized ICNN displacements lie just above the threshold and regularized ones just below it, rather than genuine support sparsity. Please report the full distribution of |Δ_i|, perform a threshold-sensitivity analysis (e.g., 10^-3, 10^-2, 5×10^-2), and report how many coordinates are exactly zero, so the interpretability claim is not an artifact of the cutoff.
[Section 4 and Appendix D.4] Theorems 4.4 and 4.5 are proved only for convex regularizers, and Appendix D.4 explicitly states that the smoothed ℓ0-norm 'does not benefit from the theoretical guarantee.' Since the main experiments use smoothed ℓ0, the theory in Section 4 does not actually cover the method that carries the empirical claims. Either extend the theory to pseudoconvex penalties (or provide an approximation argument), or clearly restate the contribution as a heuristic method whose theory is limited to the convex case.
[Appendix B, proof of Theorem 4.3] In the proof of Theorem 4.3, after equation (13) the argument writes limsup_{λ→0} J_λ ≤ V_{P,Q}(f_λ,g_0) and then substitutes f=f_λ into inequality (11). But f_λ depends on λ, so passing the limit into V_{P,Q}(f_λ,g_0) requires extracting a convergent subsequence (for example by Arzelà-Ascoli under Assumption 3.1) and using continuity of V_{P,Q}. As written this is a gap, and it affects the consistency claim for non-convex τ as well as the limiting argument generally.
[Section 4.1, Algorithm 4, and Section 5.1] In the high-dimensional mode, Algorithm 4 directly increases λ until the measured dimension falls below the constraint l, then decreases λ while keeping the dimension below l. Thus the claim that 'adaptive-λ achieves a displacement dimensionality closer to the ground truth' (Section 5.1, Table 1) is partly by construction, because the algorithm is optimizing against that very dimension. This is not by itself a flaw, but the evaluation should be reframed as a feasibility-under-constraint result, and the constant-λ baseline should be compared after the same thresholding procedure is applied.

minor comments (6)

[Section 5.1] There is a corrupted block of text containing '/uni00000013/...' tokens in the paragraph after the experimental setup; this should be removed or replaced with the intended figure caption or description.
[Table 1] The header 'Cuturi et al. 23Oursλ' is missing a space and should read 'Cuturi et al. [23]' and 'Ours (λ)'.
[Section 3.3 vs Appendix D.4] The paper calls the smoothed ℓ0-norm 'non-convex' in Section 3.3 but 'pseudoconvex' in Appendix D.4; please clarify the terminology and be consistent about which property is being invoked.
[Appendix B] The proof of Lemma B.2 is labeled 'Lemma A.2' in the text, and the proof of Theorem 4.3 refers to 'Theorem A.1' when the stated minimax theorem is in fact Theorem B.1; internal cross-references should be corrected.
[Appendix B, proof of Theorem 4.6] In the proof of Theorem 4.6, the notation 's(f*)' and 'λs(g′)' is confusing; the penalty function s is applied to dual potentials g, so the expressions should be written consistently as s(g*) and λs(g′).
[Figure 5 caption] The caption says 'The average displacement dimensionality is shown' but does not define the dimension metric; please specify in the caption that dimensionality is computed with the 10^-2 threshold described in Appendix D.3.

Circularity Check

1 steps flagged · score 6.0 of 10

The reported adaptive-λ dimensionality advantage is partly by construction: Algorithm 4 is given the target dimension l (set to ground truth 100) and drives λ until the dimension meets it.

fitted input called prediction [Section 4.1 (High-Dimensional Tasks), Algorithm 4, and Figure 7 caption; results in Tables 1–2]
"In contrast, for high-dimensional tasks, we directly constrain the map to match the user’s desired dimensionality, while searching for the most feasible solution under this constraint. ... while DimT ≥ l and Tem > Temmin do λ′ ← SIMULATED ANNEALING PROPOSAL (T, λ) # Increase λ for sparsity ... DimT ← COMPUTE DIM(T). ... Dimensionality constraint is set to l = 100 and smoothed ℓ0 penalty is used."

The synthetic experiment's central claim is that adaptive-λ reaches displacement dimension 107 versus 2480 for ICNN OT and closer to the ground-truth k=100 than constant-λ. But Algorithm 4 takes l as an input and increases λ until DimT < l, then decreases λ only while DimT < l; in the experiment l is set to the ground-truth k=100. Thus the final dimension is not an independent prediction: it is the constraint the algorithm is engineered to satisfy. The comparison to constant-λ baselines is therefore not a test of the method's ability to recover dimensionality; it is a comparison between a constrained solver and unconstrained solvers.

full rationale

The paper's main theoretical framework (Eqs. 6–7, Theorems 4.3–4.6) is a standard penalized minimax construction: adding λ∫τ(∇g(y)−y)dQ to the dual objective and proving that the optimum approaches the unregularized optimum as λ→0 is a direct consequence of the penalty's boundedness and is not circular. The monotonicity result in Theorem 4.5 is a standard consequence of increasing a penalty strength rather than an independent prediction. The high-dimensional adaptive scheme is the one genuinely circular element: because Algorithm 4 explicitly takes the target dimensionality l as input and in Figure 7 l is set to the ground-truth k=100, the reported final dimension (107) is a fitted target, not a discovered value. Constant-λ comparisons in Tables 1–2 do not receive this constraint, so the 'adaptive-λ outperforms' claim overstates what was predicted. Independent content remains: the smoothed-ℓ0 regularizer is compared fairly against ℓ1 on neural OT, the selected-gene overlap with ground truth is an external check not used as a training constraint, and the real 4i experiments impose no dimension target. The threshold-based dimension metric (np.isclose at 10^-2, Appendix D.3) is a validity concern because the smoothed ℓ0 penalty is smooth at zero and never forces exact zeros, so apparent sparsity may be partly threshold artifact; this should be scored as a correctness/robustness risk, not as circularity. The paper's own limitation (Appendix D.4) concedes that Theorems 4.4–4.5 hold only for convex penalties while experiments use the non-convex smoothed ℓ0 norm, which weakens the 'theoretically grounded' claim but is again a scope mismatch rather than a circular derivation. Overall, partial circularity: one headline empirical claim is by construction, while the formulation, regularizer comparison, and overlap validation are self-contained.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central method depends on the user-chosen λ, a, l, and σ; the theoretical statements add structural restrictions on the measures and function classes that are not verified in practice. No new physical entities are introduced.

free parameters (4)

λ (sparsity intensity) = varies; e.g., 5e-4 to 2 in experiments
Controls trade-off between sparsity and feasibility; per-experiment values are scanned or tuned via simulated annealing (Algorithm 1/4), and the headline results depend on this choice.
a (tradeoff weight in Eval) = a ∈ [0,1] (user-specified)
Balances sparsity Spa against feasibility Res in the low-dimensional heuristic; the choice of a determines which λ is selected (Eq. 8).
l (dimensionality constraint) = l=100 in synthetic experiment
High-dimensional tasks directly constrain the displacement dimensionality to user specification l; the reported 'dim(Δ(x))≈107' is measured against this chosen target.
σ (Gaussian width in smoothed ℓ0) = σ=1.0 in Figure 5 experiments
Width of the Gaussian approximation of the ℓ0 norm; chosen by hand in Appendix E.

assumptions (4)

domain assumption Assumption 3.1: P,Q admit densities; S(P), S(Q) are uniformly bounded convex subsets; ICNN(Rd) ⊂ S(P), ICNN(Rd) ⊂ S(Q).
Used to guarantee existence of optimal dual potentials and that the transport map is recovered as ∇g0 (Prop 3.2); required for all theoretical results.
domain assumption P and Q have bounded support; functions in S(P), S(Q) are L-Lipschitz (Lemma 4.1).
Needed to bound the sparsity penalty s(g) by Mτ and to apply Arzelà-Ascoli in Theorem 4.4; the paper argues bounded support is achievable by bounding ICNN parameters (Remark 4.2).
domain assumption The dual potential for the unbiased problem f is α-strongly convex (Theorem 4.6).
The map-error bound ∥∇g−∇g0∥² ≤ (2/α)(ε1+ε2+2λMτ) uses strong convexity of f; ICNN training does not enforce strong convexity.
ad hoc to paper Optimality gaps ε1, ε2 converge to 0 (bounded by generalization error δ1 and optimization error δ2).
The paper asserts this without proof to conclude that the main error term comes from λ; it assumes ICNN training converges to the global minimax optimum, which is not guaranteed.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Displacement-Sparse Neural Optimal Transport." pith.science (2026). https://pith.science/paper/PJNWLHPO

@misc{pith2026250201889,
  author       = {Pith},
  title        = {Pith review of: Displacement-Sparse Neural Optimal Transport},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PJNWLHPO}},
  note         = {Machine review of arXiv:2502.01889}
}

abstract

Optimal transport (OT) aims to find a map $T$ that transports mass from one probability measure to another while minimizing a cost function. Recently, neural OT solvers have gained popularity in high dimensional biological applications such as drug perturbation, due to their superior computational and memory efficiency compared to traditional exact Sinkhorn solvers. However, the overly complex high dimensional maps learned by neural OT solvers often suffer from poor interpretability. Prior work addressed this issue in the context of exact OT solvers by introducing \emph{displacement-sparse maps} via designed elastic cost, but such method failed to be applied to neural OT settings. In this work, we propose an intuitive and theoretically grounded approach to learning \emph{displacement-sparse maps} within neural OT solvers. Building on our new formulation, we introduce a novel smoothed $\ell_0$ regularizer that outperforms the $\ell_1$ based alternative from prior work. Leveraging Input Convex Neural Network's flexibility, we further develop a heuristic framework for adaptively controlling sparsity intensity, an approach uniquely enabled by the neural OT paradigm. We demonstrate the necessity of this adaptive framework in large-scale, high-dimensional training, showing not only improved accuracy but also practical ease of use for downstream applications.

Figures

Figures reproduced from arXiv: 2502.01889 by the authors.

**Figure 2.** Aggregate results for τstvs at different sparsity intensity levels. A comparison between the mapped source elements T(X) and the target elements Y is shown. where at higher sparsity intensity level, fewer points can be transported to the target measure, and the displacement vectors become shorter due to the stronger sparsity penalty. Finding a dynamic balance for this trade-off is practically infeasible for exact Si… view at source ↗

**Figure 3.** Input Convex Neural Network (ICNN) Structure. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (7 more)

**Figure 4.** Figure 4: Left: Two classic eight-Gaussian examples are presented, where the source measure is located at the center, and eight target measures are generated from Gaussian distributions. These examples illustrate the trade-off between the sparsity of the map and its feasibility …

**Figure 5.** Figure 5: Synthesized sc-RNA perturbation dataset with n = 4000, d = 3000, and k = 100. The average displacement dimensionality is shown. We used the exact OT solver (ℓ1) to help construct the synthetic dataset, ensuring that the ground-truth displacement dimensionality is at…

**Figure 6.** Figure 6: The results of the 4i perturbation are presented. The average of ten runs is shown for each [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

**Figure 7.** Figure 7: Three consecutive runs of high-dimensional dynamic adjustment of [PITH_FULL_IMAGE:figures/full_fig_p021_7.png]

**Figure 8.** Figure 8: Examples of displacement vectors for the first 300 genes on the first cells in the [PITH_FULL_IMAGE:figures/full_fig_p022_8.png]

**Figure 9.** Figure 9: Examples of displacement vectors for first 100 genes on [PITH_FULL_IMAGE:figures/full_fig_p022_9.png]

**Figure 10.** Figure 10: Different trajectories of λ for the ℓ1 penalty are shown. The experiment is conducted on the dataset with a ground truth dimension of k = 100. “Smoothed ℓ0” represents the training trajectory of the previously used Smoothed ℓ0 with λ = 0.005. 22 [PITH_FULL_IMAGE:figu…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 48 canonical work pages

[1]

Optimizing functionals on the space of probabilities with input convex neural networks

David Alvarez-Melis, Yair Schiff, and Youssef Mroueh. Optimizing functionals on the space of probabilities with input convex neural networks. Transactions on Machine Learning Research,

work page
[2]

Zico Kolter

Brandon Amos, Lei Xu, and J. Zico Kolter. Input convex neural networks. In Proceedings of the 34th International Conference on Machine Learning, volume 70 of Proceedings of Machine Learning Research, pages 146–155. PMLR, 2017

work page 2017
[3]

Sparsistency for inverse optimal transport

Francisco Andrade, Gabriel Peyré, and Clarice Poon. Sparsistency for inverse optimal transport. arXiv preprint arXiv:2310.05461, 2024

arXiv 2024
[4]

Wasserstein generative ad- versarial networks

Martin Arjovsky, Soumith Chintala, and Léon Bottou. Wasserstein generative ad- versarial networks. In Doina Precup and Yee Whye Teh, editors, Proceedings of the 34th International Conference on Machine Learning , volume 70 of Proceedings of Machine Learning Research , pages 214–223. PMLR, 06–11 Aug 2017. URL https://proceedings.mlr.press/v70/arjovsky17a.html

work page 2017
[5]

Neural optimal transport with general cost functionals

Arip Asadulaev, Alexander Korotin, Vage Egiazarian, Petr Mokrov, and Evgeny Burnaev. Neural optimal transport with general cost functionals. In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=gIiz7tBtYZ

work page 2024
[6]

Sparse regularized optimal transport with deformed q-entropy

Han Bao and Shinsaku Sakaue. Sparse regularized optimal transport with deformed q-entropy. Entropy, 24(11):1634, 2022. doi: 10.3390/e24111634

work page doi:10.3390/e24111634 2022
[7]

Smooth and sparse optimal transport

Mathieu Blondel, Vivien Seguy, and Antoine Rolet. Smooth and sparse optimal transport. In Proceedings of the 21st International Conference on Artificial Intelligence and Statistics (AISTATS), volume 84 of Proceedings of Machine Learning Research, pages 880–889. PMLR,

work page
[8]

Sliced and radon wasser- stein barycenters of measures

Nicolas Bonneel, Julien Rabin, Gabriel Peyré, and Hanspeter Pfister. Sliced and radon wasser- stein barycenters of measures. Journal of Mathematical Imaging and Vision , 51(1):22–45, 2015

work page 2015

Show all 60 references

[9]

Polar factorization and monotone rearrangement of vector-valued functions

Yann Brenier. Polar factorization and monotone rearrangement of vector-valued functions. Communications on Pure and Applied Mathematics, 44(4):375–417, 1991

1991
[10]

Supervised training of conditional monge maps

Charlotte Bunne, Kilian Fatras, Rémi Flamary, Marco Cuturi, and Andreas Krause. Supervised training of conditional monge maps. In Gustau Camps-Valls, Francisco J. R. Ruiz, and Isabel Valera, editors, Proceedings of The 25th International Conference on Artificial Intelligence a...
[11]

Andreas Krause, and Gunnar Rätsch

Charlotte Bunne, Stefan Stark, Gabriele Gut, Kjong-Van Lehmann, Lucas Pelkmans, R. Andreas Krause, and Gunnar Rätsch. Datasets of learning single-cell perturbation responses using neural optimal transport, 2023

2023
[12]

Stark, Gregor Gut, et al

Cédric Bunne, Sebastian G. Stark, Gregor Gut, et al. Learning single-cell perturbation responses using neural optimal transport. Nature Methods , 20:1759–1768, 2023. doi: 10.1038/s41592-023-01969-x

2023 doi
[13]

Nazar Buzun, Maksim Bobrin, and Dmitry V . Dylov. Enot: Expec- tile regularization for fast and accurate training of neural optimal trans- port. In Advances in Neural Information Processing Systems , 2024. URL https://proceedings.neurips.cc/paper_files/paper/2024/hash/d885c74a...

2024
[14]

Geometric framework for 3d cell segmentation correction, 2025

Peter Chen, Bryan Chang, Olivia Annette Creasey, Julie Beth Sneddon, and Yin- ing Liu. Geometric framework for 3d cell segmentation correction, 2025. URL https://arxiv.org/abs/2502.01890

2025
[15]

Fast and scalable wasserstein- 1 neural optimal transport solver for single-cell perturbation prediction

Yanshuo Chen, Zhengmian Hu, Wei Chen, and Heng Huang. Fast and scalable wasserstein- 1 neural optimal transport solver for single-cell perturbation prediction. arXiv preprint arXiv:2411.00614, 2024. 10

2024 arXiv
[16]

Sinkhorn distances: Lightspeed computation of optimal transport

Marco Cuturi. Sinkhorn distances: Lightspeed computation of optimal transport. In C.J. Burges, L. Bottou, M. Welling, Z. Ghahramani, and K.Q. Weinberger, editors, Advances in Neural Information Processing Systems , volume 26. Curran Associates, Inc., 2013. URL https://proceedi...

2013
[17]

Optimal transport tools (ott): A jax toolbox for all things wasserstein

Marco Cuturi, Laetitia Meng-Papaxanthos, Yingtao Tian, Charlotte Bunne, Geoff Davis, and Olivier Teboul. Optimal transport tools (ott): A jax toolbox for all things wasserstein. arXiv preprint arXiv:2201.12324, 2022

2022 arXiv
[18]

Monge, bregman and occam: Interpretable optimal transport in high-dimensions with feature-sparse maps

Marco Cuturi, Michal Klein, and Pierre Ablin. Monge, bregman and occam: Interpretable optimal transport in high-dimensions with feature-sparse maps. In Proceedings of the 40th In- ternational Conference on Machine Learning, volume 202 of Proceedings of Machine Learning Researc...

2023
[19]

Noble, and Ritambhara Singh

Pinar Demetci, Ryan Santorella, Bjorn Sandstede, William S. Noble, and Ritambhara Singh. Scot: single-cell multi-omics alignment with optimal transport. Journal of Computational Biology, 29:3–18, 2022. doi: 10.1089/cmb.2021.0295

2022
[20]

Schwartz

Nelson Dunford and Jacob T. Schwartz. Linear Operators, Volume 1 . Pure and Applied Mathematics. Wiley-Interscience, New York, 1958. Part 1: General Theory

1958
[21]

Scalable computation of monge maps with general costs

Jianqiao Fan, Shu Liu, Shaojun Ma, and Huamin Zhao. Scalable computation of monge maps with general costs. In International Conference on Learning Representations (ICLR), 2022. URL https://openreview.net/forum?id=rEnGR3VdDW5

2022
[22]

Scalable computations of wasser- stein barycenter via input convex neural networks

Jiaojiao Fan, Amirhossein Taghvaei, and Yongxin Chen. Scalable computations of wasser- stein barycenter via input convex neural networks. In Marina Meila and Tong Zhang, editors, Proceedings of the 38th International Conference on Machine Learning, volume 139 of Pro- ceedings ...

2021
[23]

Sparsity of quadratically regularized optimal transport: Scalar case

Alberto González-Sanz and Marcel Nutz. Sparsity of quadratically regularized optimal transport: Scalar case. arXiv preprint arXiv:2410.03353, 2024

2024
[24]

Improved training of Wasserstein GANs

Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron C Courville. Improved training of Wasserstein GANs. In Advances in Neural Information Processing Systems, volume 30, pages 5767–5777. Curran Associates, Inc., 2017

2017
[25]

Optimal transport improves cell–cell simi- larity inference in single-cell omics data

Geert-Jan Huizing, Gabriel Peyré, and Laura Cantini. Optimal transport improves cell–cell simi- larity inference in single-cell omics data. Bioinformatics, 38(8):2169–2177, 2022. doi: 10.1093/ bioinformatics/btac084. URL https://doi.org/10.1093/bioinformatics/btac084

2022 doi
[26]

Mapping cells through time and space with moscot

Dominik Klein, Giovanni Palla, Marius Lange, et al. Mapping cells through time and space with moscot. Nature, 625(8001):123–130, 2024. doi: 10.1038/s41586-024-08453-2

2024 doi
[27]

Learning elastic costs to shape monge displacements

Michal Klein, Aram-Alexandre Pooladian, Pierre Ablin, Eugene Ndiaye, Jonathan Niles- Weed, and marco cuturi. Learning elastic costs to shape monge displacements. In The Thirty-eighth Annual Conference on Neural Information Processing Systems , 2024. URL https://openreview.net/...

2024
[28]

Wasserstein-2 generative networks

Alexander Korotin, Vage Egiazarian, Arip Asadulaev, Alexander Safin, and Evgeny Burnaev. Wasserstein-2 generative networks. In International Conference on Learning Representations,
[29]

Do neural optimal transport solvers work? a continuous wasserstein-2 bench- mark

Alexander Korotin, Lingxiao Li, Aude Genevay, Justin Solomon, Alexander Filippov, and Evgeny Burnaev. Do neural optimal transport solvers work? a continuous wasserstein-2 bench- mark. In Advances in Neural Information Processing Systems, NIPS ’21. Curran Associates Inc., 2021

2021
[30]

Sparsity-constrained optimal transport

Tianlin Liu, Joan Puigcerver, and Mathieu Blondel. Sparsity-constrained optimal transport. In Proceedings of the Eleventh International Conference on Learning Representations (ICLR) ,
[31]

Blumberg

Yining Liu, Yinuo Jin, Elham Azizi, and Andrew J. Blumberg. Cellstitch: 3d cellular anisotropic image segmentation via optimal transport. BMC Bioinformatics, 24:480, 2023. doi: 10.1186/ s12859-023-05608-2. URL https://doi.org/10.1186/s12859-023-05608-2

2023 doi
[32]

Group sparse optimal transport for sparse process flexibility design

Dixin Luo, Tingting Yu, and Hongteng Xu. Group sparse optimal transport for sparse process flexibility design. In Proceedings of the 32nd International Joint Conference on Artificial Intelligence (IJCAI-23), pages 6121–6129, 2023

2023
[33]

Optimal transport mapping via input convex neural networks

Ashok Makkuva, Amirhossein Taghvaei, Sewoong Oh, and Jason Lee. Optimal transport mapping via input convex neural networks. In Proceedings of the 37th International Conference on Machine Learning , volume 119 of Proceedings of Machine Learning Research , pages 6672–6681. PMLR, 2020

2020
[34]

Hosein Mohimani, Massoud Babaie-Zadeh, and Christian Jutten

G. Hosein Mohimani, Massoud Babaie-Zadeh, and Christian Jutten. Fast sparse representation based on smoothed l0 norm. In Mike E. Davies, Christopher J. James, Samer A. Abdallah, and Mark D. Plumbley, editors, Independent Component Analysis and Signal Separation , pages 389–396...

2007
[35]

Large-scale wasserstein gradient flows

Petr Mokrov, Alexander Korotin, Lingxiao Li, Aude Genevay, Justin Solomon, and Evgeny Burnaev. Large-scale wasserstein gradient flows. In A. Beygelzimer, Y . Dauphin, P. Liang, and J. Wortman Vaughan, editors,Advances in Neural Information Processing Systems, 2021. URL https:/...

2021
[36]

Mémoire sur la théorie des déblais et des remblais

Gaspard Monge. Mémoire sur la théorie des déblais et des remblais. Histoire de l’Académie Royale des Sciences, pages 666–704, 1781
[37]

Quadratically regularized optimal transport: Existence and multiplicity of potentials

Marcel Nutz. Quadratically regularized optimal transport: Existence and multiplicity of potentials. arXiv preprint arXiv:2404.06847, 2024

2024 arXiv
[38]

Tyrrell Rockafellar and Roger J.-B

R. Tyrrell Rockafellar and Roger J.-B. Wets. Variational Analysis, volume 317 of Grundlehren der mathematischen Wissenschaften: A Series of Comprehensive Studies in Mathemat- ics. Springer, Berlin, Heidelberg, 1998. ISBN 978-3-642-64980-6. doi: 10.1007/ 978-3-642-02431-3

1998
[39]

Schiebinger, J

G. Schiebinger, J. Shu, M. Tabaka, B. Cleary, V . Subramanian, A. Solomon, J. Gould, S. Liu, S. Lin, P. Berube, L. Lee, J. Chen, J. Brumbaugh, P. Rigollet, K. Hochedlinger, R. Jaenisch, A. Regev, and E. S. Lander. Optimal-transport analysis of single-cell gene expression ident...

2019 doi
[40]

A shrinkage- thresholding metropolis adjusted langevin algorithm for bayesian variable selection

Amandine Schreck, Gersende Fort, Sylvain Le Corff, and Eric Moulines. A shrinkage- thresholding metropolis adjusted langevin algorithm for bayesian variable selection. IEEE Journal of Selected Topics in Signal Processing, 10(2):366–375, 2015

2015
[41]

On general minimax theorems

Maurice Sion. On general minimax theorems. Pacific Journal of Mathematics, 8(1):171–176,
[42]

2-wasserstein approximation via restricted convex potentials with application to improved training for gans, 2019

Amirhossein Taghvaei and Amin Jalali. 2-wasserstein approximation via restricted convex potentials with application to improved training for gans, 2019. URL https://arxiv.org/abs/1902.07197

2019 arXiv
[43]

Optimal transport with f-divergence regularization and generalized sinkhorn algorithm

Dávid Terjék and Diego González-Sánchez. Optimal transport with f-divergence regularization and generalized sinkhorn algorithm. In Proceedings of the 25th International Conference on Artificial Intelligence and Statistics (AISTATS), pages 5135–5165. PMLR, 2022

2022
[44]

Multistageot: Multistage optimal transport infers trajectories from a snapshot of single-cell data.arXiv preprint arXiv:2502.05241, 2025

Magnus Tronstad, Johan Karlsson, and Joakim S Dahlin. Multistageot: Multistage optimal transport infers trajectories from a snapshot of single-cell data.arXiv preprint arXiv:2502.05241, 2025

2025 arXiv
[45]

The monge gap: A regularizer to learn all transport maps

Théo Uscidda and Marco Cuturi. The monge gap: A regularizer to learn all transport maps. In Proceedings of the 40th International Conference on Machine Learning, pages 34709–34733. PMLR, 2023. URL https://proceedings.mlr.press/v202/uscidda23a.html. 12

2023
[46]

Topics in Optimal Transportation, volume 58

Cédric Villani. Topics in Optimal Transportation, volume 58. American Mathematical Society, 2003

2003
[47]

Optimal Transport: Old and New , volume 338 of Grundlehren der mathematischen Wissenschaften

Cédric Villani. Optimal Transport: Old and New , volume 338 of Grundlehren der mathematischen Wissenschaften . Springer-Verlag, Berlin, Heidelberg,
[48]

Cmot: Cross-modality optimal transport for multimodal inference

Daifeng Wang et al. Cmot: Cross-modality optimal transport for multimodal inference. Genome Biology, 24(1):163, 2023. doi: 10.1186/s13059-023-02989-8

2023 doi
[49]

Improved algorithms for convex-concave min- imax optimization

Yuanhao Wang and Jian Li. Improved algorithms for convex-concave min- imax optimization. In H. Larochelle, M. Ranzato, R. Hadsell, M.F. Bal- can, and H. Lin, editors, Advances in Neural Information Processing Sys- tems, volume 33, pages 4800–4810. Curran Associates, Inc., 2020...
[50]

Improving the improved training of wasserstein GANs

Xiang Wei, Zixia Liu, Liqiang Wang, and Boqing Gong. Improving the improved training of wasserstein GANs. In International Conference on Learning Representations , 2018. URL https://openreview.net/forum?id=SJx9GQb0-

2018
[51]

Zhang, A

S. Zhang, A. Afanassiev, L. Greenstreet, T. Matsumoto, and G. Schiebinger. Optimal transport analysis reveals trajectories in steady-state systems. PLoS Computational Biology , 17(7): e1009466, 2021. doi: 10.1371/journal.pcbi.1009466

2021 doi
[52]

Zhang, Gilles Mordant, Tetsuya Matsumoto, and Geoffrey Schiebinger

Stephen X. Zhang, Gilles Mordant, Tetsuya Matsumoto, and Geoffrey Schiebinger. Manifold learning with sparse regularised optimal transport. arXiv preprint arXiv:2307.09816, 2023

2023 arXiv
[53]

Modeling cell type developmental trajectory using multinomial unbalanced optimal transport

Junhao Zhu, Kevin Zhang, Zhaolei Zhang, and Dehan Kong. Modeling cell type developmental trajectory using multinomial unbalanced optimal transport. arXiv preprint arXiv:2501.03501, 2025. 13 A Further Related Works Neural OT with General Cost. In this section, we further discus...

2025
[60]

Smoothed ℓ0

requires the objective function to be strictly convex w.r.t. to the coupling Π to guarantee that the solution obtained by the minimax formulation is indeed an OT map [ 5, Theorem 2]. But this may necessitate regularization such as negative entropy which leads to further deviat...

2000
[1958]

doi: 10.2140/pjm.1958.8.171

1958 doi
[2009]

doi: 10.1007/978-3-540-71050-9

ISBN 978-3-540-71049-3. doi: 10.1007/978-3-540-71050-9. URL https://link.springer.com/book/10.1007/978-3-540-71050-9

doi
[2018]

URL http://proceedings.mlr.press/v84/blondel18a/blondel18a.pdf
[2021]

URL https://openreview.net/forum?id=bEoxzW_EXsa
[2022]

URL https://openreview.net/forum?id=dpOYN7o8Jm

ISSN 2835-8856. URL https://openreview.net/forum?id=dpOYN7o8Jm
[2023]

URL https://openreview.net/forum?id=yHY9NbQJ5BP. 11

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.

[1] [1]

Optimizing functionals on the space of probabilities with input convex neural networks

David Alvarez-Melis, Yair Schiff, and Youssef Mroueh. Optimizing functionals on the space of probabilities with input convex neural networks. Transactions on Machine Learning Research,

work page

[2] [2]

Zico Kolter

Brandon Amos, Lei Xu, and J. Zico Kolter. Input convex neural networks. In Proceedings of the 34th International Conference on Machine Learning, volume 70 of Proceedings of Machine Learning Research, pages 146–155. PMLR, 2017

work page 2017

[3] [3]

Sparsistency for inverse optimal transport

Francisco Andrade, Gabriel Peyré, and Clarice Poon. Sparsistency for inverse optimal transport. arXiv preprint arXiv:2310.05461, 2024

arXiv 2024

[4] [4]

Wasserstein generative ad- versarial networks

Martin Arjovsky, Soumith Chintala, and Léon Bottou. Wasserstein generative ad- versarial networks. In Doina Precup and Yee Whye Teh, editors, Proceedings of the 34th International Conference on Machine Learning , volume 70 of Proceedings of Machine Learning Research , pages 214–223. PMLR, 06–11 Aug 2017. URL https://proceedings.mlr.press/v70/arjovsky17a.html

work page 2017

[5] [5]

Neural optimal transport with general cost functionals

Arip Asadulaev, Alexander Korotin, Vage Egiazarian, Petr Mokrov, and Evgeny Burnaev. Neural optimal transport with general cost functionals. In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=gIiz7tBtYZ

work page 2024

[6] [6]

Sparse regularized optimal transport with deformed q-entropy

Han Bao and Shinsaku Sakaue. Sparse regularized optimal transport with deformed q-entropy. Entropy, 24(11):1634, 2022. doi: 10.3390/e24111634

work page doi:10.3390/e24111634 2022

[7] [7]

Smooth and sparse optimal transport

Mathieu Blondel, Vivien Seguy, and Antoine Rolet. Smooth and sparse optimal transport. In Proceedings of the 21st International Conference on Artificial Intelligence and Statistics (AISTATS), volume 84 of Proceedings of Machine Learning Research, pages 880–889. PMLR,

work page

[8] [8]

Sliced and radon wasser- stein barycenters of measures

Nicolas Bonneel, Julien Rabin, Gabriel Peyré, and Hanspeter Pfister. Sliced and radon wasser- stein barycenters of measures. Journal of Mathematical Imaging and Vision , 51(1):22–45, 2015

work page 2015