Trajectory Generator Matching for Time Series

C. Wald; G. Steidl; J. Chemseddine; P. Hagemann; T. Jahn

REVIEW 2 major objections 5 minor 33 references

Trajectory Generator Matching for Time Series

T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper constructs explicit drift-diffusion and jump generators with the same Gaussian marginals, enabling simulation-free training of conditional time-series models that handle jumps and irregular sampling.

desk verdict A solid generator-matching paper with a genuinely useful closed-form KL for Gaussian jump kernels, but the joint-distribution guarantee rests on an unquantified Lipschitz assumption and a model family the experiments never stress. read the letter →

arxiv 2505.23215 v1 pith:EIPODJP5 submitted 2025-05-29 math.NA cs.LGcs.NA

T. Jahn , J. Chemseddine , P. Hagemann , C. Wald , G. Steidl This is my paper · ORCID

classification math.NAcs.LGcs.NA MSC 60J2560J6060J7560H1062M1068T07

keywords generatormatchingtimeseriesgenerationstochasticdifferentialequationsjumpprocessesirregularsamplingKullback-LeiblerdivergenceMarkovtrajectoryflow

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

The paper develops a generative model for time-continuous stochastic processes from irregularly sampled observations, targeting settings like high-frequency finance and clinical vital signs. Its central move is to construct explicit Markov-process bridges between consecutive observations: a linear-drift SDE and a pure-jump process that both have the same Gaussian marginals at every intermediate time. This makes conditional time-series generation into a simulation-free regression problem, and parameterizing the jump kernel by a scaled Gaussian yields a closed-form KL loss. A memory window plus an autoregressive sampling scheme then approximates the joint distribution of the observed process rather than only its one-step marginals. The approach works with jumps and irregular grids, which previous trajectory flow matching handled poorly.

What carries the argument

The machinery is the pair of generators in Proposition 2, together with the KL identity for Gaussian jump kernels. The drift-diffusion generator $L^{\mathrm{Diff}}_t f(x)=\nabla f(x)^T u_t(x)+\frac{\eta^2}{2}\Delta f(x)$ and the jump generator $L^{\mathrm{Jump}}_t f(x)=\int(f(y)-f(x))q_t(y,x)\,dy$ each solve the Kolmogorov forward equation for the same Gaussian density $p_t$, which is what makes a Markov process with exactly $N(m_t,\tau_t)$ marginals available in closed form. The jump rate uses the positive and negative parts of the same function $\xi_t$, so the two generators are two sides of one identity; Proposition 5 turns the KL between $q_t(\cdot,x)$ and a learned $\lambda N(\mu,\sigma^2 I)$ into a formula involving only the mean and covariance of the truncated-Gaussian measure $J_t$, and Proposition 6 supplies those moments in closed form. This is what removes binning and simulation from training.

What would settle it

Generate a dataset from a transition kernel that is a mixture of two well-separated Gaussians, or a jump process with two distinct jump sizes, train the method with the single-Gaussian jump ansatz, and compare the conditional densities sampled by Algorithm 2 with the true ones under Wasserstein-2; if the distance does not shrink with more training data, the restricted generator family is the reason. A second check: estimate the Lipschitz constant $K$ in Remark 7 from the data and evaluate $(1+20K^2)^n$ for the trajectory length $n$; if the bound exceeds the observed approximation error by many orders, the stated joint guarantee is too loose to explain the empirical success.

Watch

Extended reading notes

Core claim

For fixed endpoints x0 and x1, the paper proves that the Gaussian interpolation family $N(m_t, \tau_t)$ with $m_t=(1-t)x_0+tx_1$ and $\tau_t=\eta^2 t(1-t)+\rho^2$ is generated exactly by two different Markov processes. The drift-diffusion generator has drift $u_t = x_1 - x_0 - (x - m_t)\eta^2 t/\tau_t$ and diffusion $\eta$; the jump generator has rate $q_t(y,x)=\lambda_t(x)J_t(y)$, where $\lambda_t(x)=\max(0,-\xi_t(x))$ and $J_t$ is the renormalized positive part of $\xi_t$ times the Gaussian density. Both satisfy the Kolmogorov forward equation, so both produce the same marginals; setting $\rho>0$ replaces the singular Brownian bridge of the earlier approach with a stabilized one. Because the losses in Proposition 4 are Bregman divergences, training the drift and jump networks against these explicit generators yields conditional generators, and Proposition 5 gives the KL between the target jump rate and a learned Gaussian jump kernel in closed form. Algorithm 2 glues these per-interval processes using a memory window, which by Remark 7 approximates the full joint distribution under a Lipschitz condition.

Load-bearing premise

The method assumes each one-step conditional distribution can be matched by a linear-drift SDE with fixed diffusion plus a single Gaussian jump kernel using a fixed memory window, and for the joint-distribution guarantee it also assumes the true conditional laws are Lipschitz with a known constant; real transitions that are multimodal, heavy-tailed, or long-range dependent fall outside this and come with no error bound.

Editorial extensions

If this is right

Because both generators are known in closed form for every interval, training never has to simulate the SDE or jump process; sampling is deferred to Algorithm 2.
Taking a convex combination of the learned drift-diffusion and jump generators produces another valid generator, so one model can interpolate between purely continuous and purely jumpy dynamics.
The closed-form KL loss for Gaussian jump kernels removes the need for binning the jump rate, making the loss cheap and differentiable.
With a fixed memory window, the autoregressive sampler approximates the full joint distribution of the observations, not only the one-step marginals, up to an error governed by the Wasserstein-Lipschitz constant of the conditional laws.
The stabilized bridge with $\rho>0$ removes the $t=1$ singularity of the previous method, so training and inference do not require an ad hoc noise floor.

Reading between the lines

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

Not claimed by the paper: the closed-form KL identity only needs the first two moments of $J_t$, so replacing the single Gaussian jump kernel by a Gaussian mixture or a normalizing flow should keep the loss structure intact.
Not claimed by the paper: marking events with an extra coordinate, such as trade sizes or alarm types, should let the same generator pair model marked point processes; this is a natural extension of the jump generator.
Not claimed by the paper: part of the reported gain over the earlier method may come from predicting velocity rather than denoising; a controlled comparison holding the bridge fixed would separate the two choices.
Not claimed by the paper: because the joint error bound grows as $(1+20K^2)^n$, the practical guarantee on long trajectories depends on $K$ being small; fitting $K$ from data would show whether the bound is informative.

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.

Desk Editor's Note

read the letter

The paper is worth a serious look. What is actually new: a rho-stabilized Brownian-bridge generator that removes the singular drift in Zhang et al.'s trajectory flow matching, a jump-kernel parameterization by scaled Gaussians with an analytic KL divergence (instead of binning as in Holderrieth et al.), and a clean conditional, irregular-grid autoregressive framing. I re-checked the key identities in d=1: the drift-diffusion KFE computation, the jump generator, and the KL formula in Proposition 5 are all correct. The Markov superposition trick also appears to work as advertised. The experiment section is thin but honest: two synthetic datasets, one baseline, MMD as the only metric. The improvement over TFM in the subsampled regime is real, and the superposition often helps. That said, there are three soft spots. First, Proposition 5 states the sigma-minimizer as trace(Sigma_J), which is only right in d=1. The correct value is sqrt(trace(Sigma_J)/d); harmless for the experiments but a clear slip. Second, the joint-distribution guarantee in Remark 7 / Appendix C is too weak to carry the paper's central claim. The chaining bound grows like (1+20K^2)^n, K is never estimated, and for n=50-100 it is vacuous unless K is tiny. No per-step approximation error is quantified. The authors should either prove a tighter bound or soften the claim to "approximates the joint distribution under unverified but plausible regularity." Third, the restricted model family (single Gaussian jump kernel, fixed diffusion, finite memory window) can systematically miss multi-modal or heavy-tailed transitions, and the two benchmarks do not stress it. The Black-Scholes process is essentially log-normal; the synthetic trend+noise is unimodal. So the empirical evidence does not establish robustness for the kind of financial or clinical series the paper motivates. Overall, the core math is sound and the closed-form KL is a practical contribution. I recommend sending to peer review, with a request to fix the d>1 slip and to either add a realistic error bound or temper the joint-distribution claim. This paper is for researchers working on simulation-free generative modeling of continuous-time or irregularly sampled processes. I'd bring it to reading group and cite the stabilized bridge and KL formula.

Referee Report

2 major / 5 minor

Summary. The paper proposes a generative model for time series based on generator matching. For each interval between consecutive observation times, the authors construct explicit Markov generators whose marginals interpolate between the observed data points: a drift-diffusion generator with a Gaussian bridge law N(m_t, τ_t) and a pure-jump generator with a related jump kernel. Training losses are derived from oracle regressions to these analytically known generators, and the jump kernel is parameterized by a scaled Gaussian so that the KL divergence in the loss has a closed form. The learned conditional generators are chained autoregressively to sample new trajectories, including on irregular time grids. Experiments on a synthetic trend-plus-noise process and a simulated Black-Scholes process compare the SDE, jump, and Markov-superposition variants against trajectory flow matching, reporting MMD on held-out test trajectories.

Significance. The paper contains several genuinely useful theoretical building blocks: Proposition 2 gives explicit, singularity-free generator formulas for Gaussian bridge marginals that can be verified directly against the Kolmogorov forward equation; Proposition 5 provides a closed-form KL divergence for Gaussian jump kernels, avoiding expensive binning; and the empirical comparison includes a held-out MMD metric and an ablation of the Markov superposition parameter. The main weakness is that the central claim of approximation of the full joint distribution rests on unverified assumptions (a Lipschitz constant K and a per-step error ε), and the validation is limited to two unimodal or mildly non-Gaussian benchmarks that do not stress the restricted model family. The theoretical result is a conditional statement rather than an unconditional guarantee, and the introduction's phrasing overstates what is proved.

major comments (2)

[Section 5, Remark 7, Appendix C] The joint-distribution guarantee for Algorithm 2 relies on Proposition 9, which requires a per-step approximation error ε and a Wasserstein-2 Lipschitz constant K for both the target conditional laws and the learned kernels. No numerical estimate of K is given, and the bound (1+20K^2)^n W2^2(P_X0,P_Y0) + 4ε Σ_{i=0}^{n-1}(1+20K^2)^i is vacuous for the reported horizons n=50–100 unless K is extremely small. Moreover, condition (i) of Proposition 9 is assumed, not established, for the restricted model class. The introduction's statement that including memory 'proves' that the learned time series approximates the true joint distribution is therefore stronger than what the theorem actually shows; the claim should be qualified or supported by numerical estimates of K and ε.
[Section 4 and Section 6.2] The training losses (4) and (5) characterize the exact conditional generator only when the minimizer is unrestricted. In the implemented method, the drift is restricted to a linear function, the diffusion is fixed isotropic η, the jump kernel is a single Gaussian λN(μ,σ^2 I_d), and conditioning is limited to a finite memory window m. For processes with multi-modal, heavy-tailed, or long-range-dependent transitions, the per-step conditional marginal will be systematically biased, and no error bound is provided for such misspecification. The synthetic benchmarks (trend-plus-noise and Black-Scholes) have unimodal or mildly non-Gaussian transitions, so the reported MMD values do not demonstrate robustness to this restriction. The paper should either include an experiment with a process that violates the Gaussian/linear assumptions or state this limitation more prominently in the abstract and conclusions.

minor comments (5)

[Proposition 5] The stated minimizer of F_{t,x} is (λ_t(x), μ_J, trace Σ_J), but the stationarity condition ∂σ F = 0 gives σ^2 = trace(Σ_J)/d, so the correct minimizer is σ = sqrt(trace(Σ_J)/d). In d=1 this equals σ_J, not σ_J^2 as suggested by the printed formula; the statement should be corrected.
[Proposition 2(ii)] The definition of q_t contains a typographical error: 'q_t(y,x) =:= λ_t(x) J_t(y)' should be 'q_t(y,x) = λ_t(x) J_t(y)'.
[Appendix B, proof of Proposition 6] The replacement formula for I_k on the interval t ∈ [t_0, (t_0+t_1)/2) includes a factor '2 (k−1)!!' that appears incorrect for the unnormalized Gaussian integrals used elsewhere in the proof; please verify whether the correct expression should be (k−1)!! − I_k for even k and −I_k for odd k (possibly with normalization by sqrt(2π)).
[Section 6.2, Tables 1 and 2] The column headers in Tables 1 and 2 are ambiguous: the text refers to 'different numbers of subsampling times' and 'subsampling rates' interchangeably, while figure captions use percentages such as '50% subsampling rate'. Please clarify whether the entries 5, 10, 25, 50, 100 are percentages or absolute counts of retained time steps.
[Algorithm 2] The notation 'ALG1(x_i, L^{ξ_i,t_{i+1}}_t)' is not defined in the pseudocode; presumably it means one step of Algorithm 1 applied with the given generator, but this should be stated explicitly.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; the derivation chain is self-contained and predictive claims are checked against held-out data.

full rationale

No circular steps found. Proposition 2 constructs the drift-diffusion and jump generators explicitly and verifies the Kolmogorov forward equation in Appendix B, so the Gaussian-bridge marginals are a proven property of the construction rather than a fitted or relabeled quantity. The losses in Section 4 regress neural networks onto the analytic targets u^{x0,x1}_t and q^{x0,x1}_t; Proposition 4 identifies the regression minima with the averaged generators, and Proposition 5 supplies the closed-form KL projection, neither step reintroducing the target as an input. Evaluation uses held-out test trajectories with MMD; only the Markov-superposition weight alpha is chosen on validation data, while rho, eta, and memory length m are fixed hyperparameters. No fitted value is relabeled as a prediction. The joint-distribution guarantee in Remark 7 and Appendix C is conditional on an unquantified per-step error epsilon and Lipschitz constant K; this is a correctness limitation rather than circularity, and the paper itself flags the Gaussian-jump restriction and one-dimensional experiments as limitations. A separate correctness slip: Proposition 5's minimizer for d>1 should be sigma = sqrt(trace Sigma_J / d), since the derivative yields d sigma^2 = trace Sigma_J; the paper's 'trace Sigma_J' is dimensionally inconsistent and the proof omits the factor d, though the experiments use d=1. Self-citations ([29], [21]) appear only as general flow-matching references and in a limitations remark, so they carry no load; the score of 2 reflects only the presence of these minor non-load-bearing self-citations.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central construction rests on generator/KFE machinery and on the mixture theorem from [14], plus an unverified Lipschitz condition for the joint-distribution claim and a fixed-memory Markov assumption in the experiments. Free parameters rho, eta, alpha, and m are chosen by hand and affect the reported numbers; none are fitted constants relabeled as predictions. No new physical entities are introduced.

free parameters (4)

rho (stabilization noise scale) = rho^2 = 0.03 in all experiments
Controls both the endpoint blur P0 = N(x0, rho^2), an intrinsic bias of the method, and the regularity of the drift since tau_t >= rho^2. Chosen by hand; enters Propositions 2, 5, and 6.
eta (diffusion coefficient) = eta^2 = 0.3 in experiments
Fixed diffusion of the SDE bridge and a parameter of the jump-kernel formulas; not learned or estimated, it determines the interpolation variance tau_t together with rho.
alpha (Markov superposition weight) = varies, e.g., 1.0, 0.85, 0.9, 0.8, 0.3, 0.35; tuned on validation for the best seed
Convex weight between the learned SDE and jump generators in L^t,alpha. Tuned per dataset and subsampling rate on the validation set, then fixed across seeds.
memory length m = m = 10 (synthetic dataset), m = 20 (Black-Scholes dataset)
Truncation of the conditioning history; chosen by hand with an informal 'main dependence on recent times' justification. The approximation theory of Remark 7 covers the full-history case, not the truncated one.

assumptions (5)

domain assumption Generator characterization of Markov processes via the KFE and Algorithm 1's Euler-type approximation of jump-diffusion processes (cited to [14, Appendix B]).
The construction maps generators to processes through this machinery without proof in this paper.
domain assumption Mixture and Bregman theorems: L^{x0}_t = integral L^{x0,x1}_t d alpha_x(x1) generates a process with marginals P^{x0}_t, and the argmin characterizations (4)-(5) hold (cited to [14, Propositions 1 and 2]).
Carries the marginal-matching guarantee of the trained generators; quoted, not reproven.
ad hoc to paper Lipschitz continuity of the conditional distributions and of the approximating Markov kernels in the conditioning variables with respect to the Wasserstein-2 metric, with constant K (Appendix C, Proposition 9 conditions ii and iii).
Unverified and specific to this paper; needed for the joint-distribution error bound, which grows like (1+20K^2)^n in the number of time points.
domain assumption Fixed-memory Markov property: the process dependence is captured by the last m observations (Section 6.2).
Truncating history is a heuristic for real data; the experiments assume it, while the theory does not cover m smaller than the full history.
standard math Gaussian moment identities and error-function recursion used in Proposition 6 (standard calculus).
Uncontroversial background for the closed-form mean and variance of J_t.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Trajectory Generator Matching for Time Series." pith.science (2026). https://pith.science/paper/EIPODJP5

@misc{pith2026250523215,
  author       = {Pith},
  title        = {Pith review of: Trajectory Generator Matching for Time Series},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EIPODJP5}},
  note         = {Machine review of arXiv:2505.23215}
}

read the original abstract

Accurately modeling time-continuous stochastic processes from irregular observations remains a significant challenge. In this paper, we leverage ideas from generative modeling of image data to push the boundary of time series generation. For this, we find new generators of SDEs and jump processes, inspired by trajectory flow matching, that have the marginal distributions of the time series of interest. Specifically, we can handle discontinuities of the underlying processes by parameterizing the jump kernel densities by scaled Gaussians that allow for closed form formulas of the corresponding Kullback-Leibler divergence in the loss. Unlike most other approaches, we are able to handle irregularly sampled time series.

Figures

Figures reproduced from arXiv: 2505.23215 by the authors.

**Figure 2.** Results on the synthetic dataset using a [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

**Figure 3.** Results on the synthetic Black-Scholes dataset using a [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Figures from the paper (10 more)

**Figure 4.** Figure 4: Ground truth data [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]

**Figure 5.** Figure 5: Results on the synthetic dataset using a [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]

**Figure 6.** Figure 6: Results on the synthetic dataset using a [PITH_FULL_IMAGE:figures/full_fig_p019_6.png]

**Figure 7.** Figure 7: Results on the synthetic dataset using a [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]

**Figure 8.** Figure 8: Results on the synthetic dataset using a [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]

**Figure 9.** Figure 9: Ground truth data 19 [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]

**Figure 10.** Figure 10: Results on the synthetic Black-Scholes dataset using a [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]

**Figure 11.** Figure 11: Results on the synthetic Black-Scholes dataset using a [PITH_FULL_IMAGE:figures/full_fig_p020_11.png]

**Figure 12.** Figure 12: Results on the synthetic Black-Scholes dataset using a [PITH_FULL_IMAGE:figures/full_fig_p020_12.png]

**Figure 13.** Figure 13: Results on the synthetic Black-Scholes dataset using a [PITH_FULL_IMAGE:figures/full_fig_p020_13.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 23 canonical work pages

[1]

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

arXiv 2023
[2]

Ambrosio, N

L. Ambrosio, N. Gigli, and G. Savaré.Gradient flows: in metric spaces and in the space of probability measures. Springer Science & Business Media, 2008

2008
[3]

Benton, Y

J. Benton, Y . Shi, V . De Bortoli, G. Deligiannidis, and A. Doucet. From denoising diffusions to denoising markov models.Journal of the Royal Statistical Society Series B: Statistical Methodology, 86(2):286–301, 2024

work page 2024
[4]

Billingsley.Convergence of probability measures

P. Billingsley.Convergence of probability measures. John Wiley & Sons, 2013

2013
[5]

Buehler, B

H. Buehler, B. Horvath, T. Lyons, I. Perez Arribas, and B. Wood. Generating financial markets with signatures.Available at SSRN 3657366, 2020

work page 2020
[6]

Deep Hedging

H. Bühler, L. Gonon, J. Teichmann, and B. Wood. Deep hedging.arXiv preprint arXiv:1802.03042, 2018

work page Pith review arXiv 2018
[7]

Campbell, W

A. Campbell, W. Harvey, C. D. Weilbach, V . D. Bortoli, T. Rainforth, and A. Doucet. Trans- dimensional generative modeling via jump diffusion models. InThirty-seventh Conference on Neural Information Processing Systems, 2023

work page 2023
[8]

R. T. Q. Chen, Y . Rubanova, J. Bettencourt, and D. K. Duvenaud. Neural ordinary differential equations. In S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett, editors,Advances in Neural Information Processing Systems, volume 31. Curran Associates, Inc., 2018

work page 2018

Show all 33 references

[9]

Corso, H

G. Corso, H. Stärk, B. Jing, R. Barzilay, and T. S. Jaakkola. Diffdock: Diffusion steps, twists, and turns for molecular docking. InThe Eleventh International Conference on Learning Representations, 2023

2023
[10]

Feydy, T

J. Feydy, T. Séjourné, F.-X. Vialard, S.-i. Amari, A. Trouve, and G. Peyré. Interpolating between optimal transport and mmd using sinkhorn divergences. InThe 22nd International Conference on Artificial Intelligence and Statistics, pages 2681–2690, 2019

2019
[11]

I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. C. Courville, and Y . Bengio. Generative adversarial nets. InNIPS, pages 2672–2680, 2014

2014
[12]

B. Gülmez. Stock price prediction with optimized deep lstm network with artificial rabbits optimization algorithm.Expert Systems with Applications, 227:120346, 2023

2023
[13]

Herrera, F

C. Herrera, F. Krach, and J. Teichmann. Neural jump ordinary differential equations: Con- sistent continuous-time prediction and filtering. InInternational Conference on Learning Representations (ICLR), 2021

2021
[14]

Holderrieth, M

P. Holderrieth, M. Havasi, J. Yim, N. Shaul, I. Gat, T. Jaakkola, B. Karrer, R. T. Q. Chen, and Y . Lipman. Generator matching: Generative modeling with arbitrary markov processes. InThe Thirteenth International Conference on Learning Representations, 2025

2025
[15]

Jia and A

J. Jia and A. R. Benson. Neural jump stochastic differential equations.Advances in Neural Information Processing Systems, 32, 2019

2019
[16]

Kerrigan, G

G. Kerrigan, G. Migliorini, and P. Smyth. Functional flow matching.The 27th International Conference on AI and Statistics (AISTATS), 2024

2024
[17]

Kidger, J

P. Kidger, J. Foster, X. Li, and T. J. Lyons. Neural SDEs as infinite-dimensional GANs. In International Conference on Machine Learning, pages 5453–5463. PMLR, 2021

2021
[18]

D. P. Kingma and J. Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2017

2017 arXiv
[19]

Lipman, R

Y . Lipman, R. T. Q. Chen, H. Ben-Hamu, M. Nickel, and M. Le. Flow matching for generative modeling. InThe Eleventh International Conference on Learning Representations, 2023. 11

2023
[20]

X. Liu, C. Gong, and qiang liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. InThe Eleventh International Conference on Learning Representations, 2023

2023
[21]

S. T. Martin, A. Gagneux, P. Hagemann, and G. Steidl. PnP-flow: Plug-and-play image restoration with flow matching. InThe Thirteenth International Conference on Learning Representations (ICLR), 2025

2025
[22]

Paszke, S

A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. Te- jani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala. Pytorch: An imperative style, high-p...

2019
[23]

Rombach, A

R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022

2022
[24]

Sohl-Dickstein, E

J. Sohl-Dickstein, E. Weiss, N. Maheswaranathan, and S. Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In F. Bach and D. Blei, editors,Proceedings of the 32nd International Conference on Machine Learning, volume 37 ofProceedings of Machine Learning Re...

2015
[25]

Y . Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, and B. Poole. Score-based generative modeling through stochastic differential equations. InInternational Conference on Learning Representations, 2021

2021
[26]

G. J. Székely and M. L. Rizzo. Energy statistics: A class of statistics based on distances.Journal of Statistical Planning and Inference, 143(8):1249–1272, 2013

2013
[27]

Vaswani, N

A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. u. Kaiser, and I. Polosukhin. Attention is all you need. In I. Guyon, U. V . Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, editors,Advances in Neural Information Processing...

2017
[28]

von Waldenfels

W. von Waldenfels. Fast positive operatoren.Zeitschrift für Wahrscheinlichkeitstheorie und Verwandte Gebiete, 4:159–174, 1965

1965
[29]

Wald and G

C. Wald and G. Steidl. Flow Matching: Markov kernels, stochastic processes and transport plans.ArXiv:2501.16839, 2025

2025 arXiv
[30]

J. M.-T. Wu, Z. Li, N. Herencsar, B. V o, and J. C.-W. Lin. A graph-based cnn-lstm stock price prediction algorithm with leading indicators.Multimedia Systems, 29(3):1751–1770, 2023

2023
[31]

J. Yoon, D. Jarrett, and M. Van der Schaar. Time-series generative adversarial networks. Advances in neural information processing systems, 32, 2019

2019
[32]

Zhang, Y

X. Zhang, Y . Pu, Y . Kawamura, A. Loza, Y . Bengio, D. Shung, and A. Tong. Trajectory flow matching with applications to clinical time series modelling. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

2024
[33]

Zlotchevski and L

A. Zlotchevski and L. Chen. Schrödinger bridge problem for jump diffusions.arXiv preprint arXiv:2411.13765, 2025. 12 A Additional Definitions Let Mac +(Rd) denote the space of absolutely continuous non-negative measures on Rd with respect to the Lebesgue measure. Our loss func...

2025 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

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

arXiv 2023

[2] [2]

Ambrosio, N

L. Ambrosio, N. Gigli, and G. Savaré.Gradient flows: in metric spaces and in the space of probability measures. Springer Science & Business Media, 2008

2008

[3] [3]

Benton, Y

J. Benton, Y . Shi, V . De Bortoli, G. Deligiannidis, and A. Doucet. From denoising diffusions to denoising markov models.Journal of the Royal Statistical Society Series B: Statistical Methodology, 86(2):286–301, 2024

work page 2024

[4] [4]

Billingsley.Convergence of probability measures

P. Billingsley.Convergence of probability measures. John Wiley & Sons, 2013

2013

[5] [5]

Buehler, B

H. Buehler, B. Horvath, T. Lyons, I. Perez Arribas, and B. Wood. Generating financial markets with signatures.Available at SSRN 3657366, 2020

work page 2020

[6] [6]

Deep Hedging

H. Bühler, L. Gonon, J. Teichmann, and B. Wood. Deep hedging.arXiv preprint arXiv:1802.03042, 2018

work page Pith review arXiv 2018

[7] [7]

Campbell, W

A. Campbell, W. Harvey, C. D. Weilbach, V . D. Bortoli, T. Rainforth, and A. Doucet. Trans- dimensional generative modeling via jump diffusion models. InThirty-seventh Conference on Neural Information Processing Systems, 2023

work page 2023

[8] [8]

R. T. Q. Chen, Y . Rubanova, J. Bettencourt, and D. K. Duvenaud. Neural ordinary differential equations. In S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett, editors,Advances in Neural Information Processing Systems, volume 31. Curran Associates, Inc., 2018

work page 2018