pith. sign in

arxiv: 2605.08550 · v3 · pith:UDTFEBXBnew · submitted 2026-05-08 · 💻 cs.LG · stat.ML

A Call to Lagrangian Action: Learning Population Mechanics from Temporal Snapshots

Vincent Guan , Lazar Atanackovic , Kirill Neklyudov This is my paper

Pith reviewed 2026-05-20 22:15 UTC · model grok-4.3

classification 💻 cs.LG stat.ML
keywords Wasserstein Lagrangian Mechanicspopulation dynamicssecond-order dynamicslearning from marginalsHamiltonian equationsgradient flowsforecastinginterpolation
0
0 comments X

The pith

Population dynamics are learned as second-order Lagrangian mechanics from observed marginal snapshots alone.

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

Current models of population dynamics rely on Wasserstein gradient flows that minimize free energy and therefore miss periodic behaviors. The authors instead model the dynamics as the minimizer of a damped Wasserstein Lagrangian action. From this variational principle they derive the corresponding Hamiltonian equations of motion, creating a unified class of second-order dynamics that includes classical mechanics, quantum mechanics, and gradient flows. They introduce an algorithm, WLM, that recovers these dynamics directly from sequences of population marginals without being told the form of the Lagrangian. The resulting model can both forecast future marginals and interpolate between observed ones, outperforming prior methods on vortex motion, embryonic development, and flocking.

Core claim

Wasserstein Lagrangian Mechanics formalizes second-order population dynamics as the stationary points of a damped Wasserstein Lagrangian action; the associated Hamiltonian equations of motion are derived without reference to a specific Lagrangian, and an algorithm is given that learns the dynamics from temporal marginal snapshots alone.

What carries the argument

The damped Wasserstein Lagrangian action, whose minimization produces Hamiltonian equations of motion for the evolving population measure.

If this is right

  • The learned dynamics can forecast future population states from current marginals.
  • Unobserved intermediate states can be interpolated without assuming a fixed Lagrangian form.
  • Classical mechanics, quantum mechanics, and gradient flows become special cases of the same variational principle.
  • Forecast accuracy improves over gradient-flow and flow-matching baselines on vortex, embryonic, and flocking data.

Where Pith is reading between the lines

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

  • The same variational structure could be tested on non-biological populations such as traffic or financial flows.
  • Learned actions might be inspected to discover unknown interaction rules in experimental data.
  • Hybrid models could combine the learned Lagrangian with known physical terms for partially observed systems.

Load-bearing premise

The observed population evolution is exactly the motion that minimizes a damped Wasserstein Lagrangian action.

What would settle it

On a held-out sequence of marginal snapshots drawn from a known periodic system such as a vortex, the learned trajectories deviate systematically from the true future states while a hand-specified Lagrangian model recovers them accurately.

Figures

Figures reproduced from arXiv: 2605.08550 by Kirill Neklyudov, Lazar Atanackovic, Vincent Guan.

Figure 1
Figure 1. Figure 1: Wasserstein gradient flows describe first-order popu￾lation dynamics that minimize the free energy F[ρt]. We pro￾pose Wasserstein Lagrangian mechanics (WLM), which describe a richer class of damped second-order dynamics, based on the population-level potential energy U[ρt]. Given the same quadratic functional, gradient flows dissipate until equilibrium, while WLM produces oscillating dynamics if γ = 0, and… view at source ↗
Figure 2
Figure 2. Figure 2: Learning population mechanics with WLM: In (a), we illustrate the principle of least Wasserstein action (3): given observed marginals p0 and p1, the true interpolants form a minimal action curve in the space of densities, with respect to a population-level Lagrangian action. Alternative curves of densities have higher action and are drawn in red. Wasserstein least action induces Hamiltonian mechanics on th… view at source ↗
Figure 3
Figure 3. Figure 3: We visualize Proposition 2.4, which shows that Wasser￾stein gradient flows admit two characterizations under WLM. The implication is from the principle of superposition (Am￾brosio et al., 2008, Theorem 8.2.1). It follows that the over￾damped characterization produces gradient flow dynamics for t > 0, no matter what v0 is initialized as. Then, to prove the second description, we note that since a gradient f… view at source ↗
Figure 4
Figure 4. Figure 4: WLM’s predictions (×) for unseen interpolants ( ) are visualized for the (a) Ocean vortex (in spatial coordinates) and (b) Embryoid body (in UMAP coordinates) datasets. Marginals that are used to train WLM’s mechanics model are plotted in gray [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Learning Boids: We plot the ground truth Boids (first row) against the population dynamics learned by our method WLM (second row), and the gradient flow baselines JKONET∗ (third row) and NN-APPEX (fourth row). All methods were trained on 50 marginals of size 1000 within t ∈ [0, 24.5]. The rollouts of each method are simulated from time 0 (with 5000 samples drawn) until the additional forecast horizon [25.0… view at source ↗
Figure 6
Figure 6. Figure 6: WLM Learned Friction (γ) curve for each SDEs over 100, 000 training epochs. Minimum learned friction at 100k epochs is γ = 5.28 (Styblinski-Tang paired setting), which collapses to the equilibrium distribution almost instantaneously. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Learnable friction on the Embryoid Body (EB) dataset for different holdout times. 128 256 512 1024 2048 full Particles per batch 0.68 0.70 0.72 0.74 0.76 0.78 W (lower is better) EB: Mini-batch Size vs. Performance & Runtime OT-CFM (0.790) WLF-UOT (0.738) OT-MFM (0.713) WLM (Ours) Time per epoch (ms) 100 200 300 400 500 600 Time per epoch (ms) [PITH_FULL_IMAGE:figures/full_fig_p020_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Runtime vs. performance when performing mini-batching with WLM averaged over 3 random seeds. C.4. Boids We implement the Boids algorithm using a Vicsek-style interacting particle system based on the classic Boids algorithm. By default, we simulate N = 1000 agents in R 2 adhering to local interaction rules, which is essentially the three Boids interaction rules with a boundary condition. In particular, at e… view at source ↗
Figure 9
Figure 9. Figure 9: Predicting Boids on unseen dynamics: Qualitative comparison of ground truth Boids dynamics (top row of each panel) versus the predicted WLM dynamics (bottom row of each panel) for three unseen Gaussian mixture initial distributions. WLM was trained on 50 frames whose population was a centered Gaussian (see [PITH_FULL_IMAGE:figures/full_fig_p022_9.png] view at source ↗
read the original abstract

The population dynamics of molecules, cells, and organisms are governed by a number of unknown forces. In the last decade, population dynamics have predominantly been modeled with Wasserstein gradient flows. However, since gradient flows minimize free energy, they fail to capture important dynamical properties, such as periodicity. In this work, we propose a change in perspective by considering dynamics that minimize a population-level action under a damped Wasserstein Lagrangian. By deriving the corresponding Hamiltonian equations of motion, we formalize Wasserstein Lagrangian Mechanics, a structured class of second-order dynamics that encompasses classical mechanics, quantum mechanics, and gradient flows. We then propose WLM as the first algorithm that learns these second-order dynamics from observed marginals, without specifying the Lagrangian. By directly learning the population mechanics, WLM can both forecast and interpolate unseen marginals, and outperforms existing gradient flow and flow matching methods across a wide range of dynamics, including vortex dynamics, embryonic development, and flocking.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 3 minor

Summary. The paper introduces Wasserstein Lagrangian Mechanics (WLM) as a class of second-order dynamics obtained by minimizing a damped Wasserstein Lagrangian action. It derives the associated Hamiltonian equations of motion, showing that this class encompasses classical mechanics, quantum mechanics, and gradient flows. The authors then present the WLM algorithm, which learns the underlying mechanics directly from sequences of observed marginal distributions without requiring an explicit Lagrangian, and apply it to forecast and interpolate future or intermediate marginals while outperforming gradient-flow and flow-matching baselines on vortex dynamics, embryonic development, and flocking examples.

Significance. If the derivations are correct and the learning procedure recovers identifiable second-order structure, the work would meaningfully extend population-dynamics modeling beyond first-order Wasserstein gradient flows, enabling capture of periodic and inertial phenomena that are currently inaccessible. The unification of disparate mechanics under a single action-minimization principle and the data-driven recovery of the Lagrangian from marginal snapshots alone would constitute a substantive contribution to both theoretical dynamical systems and machine-learning applications in biology and physics.

major comments (3)
  1. [Section 3] Section 3 (Derivation of Hamiltonian equations): The central claim that observed marginal trajectories uniquely determine a Lagrangian (and hence the second-order dynamics) is not supported by an identifiability theorem. Distinct choices of kinetic or interaction terms in the Lagrangian can produce identical measure-valued paths once damping is introduced, yet the manuscript provides no argument or regularization that selects a canonical representative.
  2. [Section 5] Section 5 (WLM algorithm and learning procedure): The optimization objective is stated to recover the true mechanics from marginal snapshots alone, but the loss does not include any term that enforces uniqueness across the equivalence class of Lagrangians. Consequently, the learned model may converge to an arbitrary member of that class rather than the underlying physical mechanism.
  3. [Section 6] Section 6 (Experimental evaluation): The reported outperformance on vortex, embryonic, and flocking tasks is presented without controls that test identifiability (e.g., recovery of known ground-truth Lagrangians or sensitivity to different initializations). Without such checks, it remains unclear whether the gains arise from correctly recovering second-order structure or from fitting an equivalent but non-unique model.
minor comments (3)
  1. [Section 3] The notation for the damped Wasserstein Lagrangian and the precise form of the damping term should be introduced with an explicit equation early in Section 3 to avoid ambiguity when the Hamiltonian is later derived.
  2. Several references to prior Wasserstein gradient-flow literature are missing or cited only in passing; a short related-work paragraph would help situate the contribution.
  3. Figure captions for the qualitative trajectory visualizations should include the time indices of the displayed marginals and the quantitative error metric used for comparison.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed comments. We address each major comment point by point below, indicating where we agree and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Section 3] Section 3 (Derivation of Hamiltonian equations): The central claim that observed marginal trajectories uniquely determine a Lagrangian (and hence the second-order dynamics) is not supported by an identifiability theorem. Distinct choices of kinetic or interaction terms in the Lagrangian can produce identical measure-valued paths once damping is introduced, yet the manuscript provides no argument or regularization that selects a canonical representative.

    Authors: We thank the referee for this observation. The manuscript derives the Hamiltonian equations from the damped Wasserstein Lagrangian action and shows that this framework unifies classical mechanics, quantum mechanics, and gradient flows, but it does not include a formal identifiability theorem asserting uniqueness of the Lagrangian from marginal trajectories alone. Different Lagrangians can indeed produce equivalent paths under damping. We will add a clarifying paragraph in Section 3 acknowledging the equivalence class of Lagrangians and explaining that the damping term, together with the data-driven optimization, selects a representative whose induced dynamics match the observed marginal evolution. revision: partial

  2. Referee: [Section 5] Section 5 (WLM algorithm and learning procedure): The optimization objective is stated to recover the true mechanics from marginal snapshots alone, but the loss does not include any term that enforces uniqueness across the equivalence class of Lagrangians. Consequently, the learned model may converge to an arbitrary member of that class rather than the underlying physical mechanism.

    Authors: We agree that the current loss, which measures discrepancy between predicted and observed marginals under the learned Hamiltonian flow, does not contain an explicit uniqueness regularizer. In practice the procedure recovers dynamics that accurately forecast and interpolate the data across the reported tasks. To address the concern, we will introduce a mild regularization term favoring simpler kinetic and interaction potentials and will discuss the practical consequences of non-uniqueness in the revised Section 5. revision: yes

  3. Referee: [Section 6] Section 6 (Experimental evaluation): The reported outperformance on vortex, embryonic, and flocking tasks is presented without controls that test identifiability (e.g., recovery of known ground-truth Lagrangians or sensitivity to different initializations). Without such checks, it remains unclear whether the gains arise from correctly recovering second-order structure or from fitting an equivalent but non-unique model.

    Authors: The experiments demonstrate consistent outperformance over gradient-flow and flow-matching baselines on tasks exhibiting inertia and periodicity. We acknowledge that dedicated identifiability controls were not included. We will add synthetic experiments with known ground-truth Lagrangians, quantify recovery error, and report performance across multiple random initializations in the revised experimental section. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's core derivation applies the standard variational principle of minimizing a damped Wasserstein Lagrangian action to obtain Hamiltonian equations of motion in the space of measures. This step is a direct, non-circular extension of classical Lagrangian mechanics to the Wasserstein setting and does not reduce to its own inputs by definition or by fitting. The subsequent claim that WLM learns second-order dynamics from marginal snapshots alone is presented as an algorithmic contribution rather than a tautological prediction; no equations or self-citations in the abstract or described claims force the output to equal the input by construction. The representation is explicitly positioned as an encompassing class (including gradient flows) rather than a uniqueness theorem that would require external verification. This is the common honest case of a self-contained modeling framework with independent grounding in variational mechanics.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim relies on the domain assumption of action minimization in Wasserstein space and introduces a new entity for the mechanics framework.

axioms (1)
  • domain assumption Population dynamics minimize a population-level action under a damped Wasserstein Lagrangian.
    This is the foundational assumption stated in the abstract for deriving the mechanics.
invented entities (1)
  • Wasserstein Lagrangian Mechanics (WLM) no independent evidence
    purpose: To provide a structured class of second-order dynamics for population mechanics.
    Introduced as a new formalism in the paper.

pith-pipeline@v0.9.0 · 5698 in / 1453 out tokens · 43239 ms · 2026-05-20T22:15:55.275706+00:00 · methodology

Review history (3 revisions) →

discussion (0)

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

Lean theorems connected to this paper

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

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

Reference graph

Works this paper leans on

260 extracted references · 260 canonical work pages · 7 internal anchors

  1. [1]

    Advances in Neural Information Processing Systems , volume=

    Variational inference via Wasserstein gradient flows , author=. Advances in Neural Information Processing Systems , volume=

    work page

  2. [2]

    Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences , volume=

    Large deviations and gradient flows , author=. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences , volume=. 2013 , publisher=

    work page 2013

  3. [3]

    Symplectic ODE-Net: Learning Hamiltonian Dy- namics with Control

    Symplectic ode-net: Learning hamiltonian dynamics with control , author=. arXiv preprint arXiv:1909.12077 , year=

    work page arXiv 1909

  4. [4]

    Handbook of Numerical Analysis , volume=

    Lagrangian schemes for Wasserstein gradient flows , author=. Handbook of Numerical Analysis , volume=. 2021 , publisher=

    work page 2021

  5. [5]

    arXiv preprint arXiv:2410.07550 , year=

    Conditional Lagrangian Wasserstein Flow for Time Series Imputation , author=. arXiv preprint arXiv:2410.07550 , year=

    work page arXiv

  6. [6]

    arXiv preprint arXiv:2505.11823 , year=

    Variational Regularized Unbalanced Optimal Transport: Single Network, Least Action , author=. arXiv preprint arXiv:2505.11823 , year=

    work page arXiv

  7. [7]

    Krishnapriyan

    Action-Minimization Meets Generative Modeling: Efficient Transition Path Sampling with the Onsager-Machlup Functional , author=. arXiv preprint arXiv:2504.18506 , year=

    work page arXiv

  8. [8]

    Neural lagrangian schr

    Koshizuka, Takeshi and Sato, Issei , journal=. Neural lagrangian schr

    work page

  9. [9]

    Advances in Neural Information Processing Systems , volume=

    Unsupervised learning of lagrangian dynamics from images for prediction and control , author=. Advances in Neural Information Processing Systems , volume=

    work page

  10. [10]

    Numerische Mathematik , volume=

    A computational fluid mechanics solution to the Monge-Kantorovich mass transfer problem , author=. Numerische Mathematik , volume=. 2000 , publisher=

    work page 2000

  11. [11]

    SIAM journal on mathematical analysis , volume=

    The variational formulation of the Fokker--Planck equation , author=. SIAM journal on mathematical analysis , volume=. 1998 , publisher=

    work page 1998

  12. [12]

    Partial differential equations and applications , pages=

    An extended variational principle , author=. Partial differential equations and applications , pages=. 2017 , publisher=

    work page 2017

  13. [13]

    Journal of mathematics of Kyoto University , volume=

    The Onsager-Machlup function for diffusion processes , author=. Journal of mathematics of Kyoto University , volume=. 1982 , publisher=

    work page 1982

  14. [14]

    Econometrica: Journal of the Econometric Society , pages=

    The dimensionality of the aliasing problem in models with rational spectral densities , author=. Econometrica: Journal of the Econometric Society , pages=. 1983 , publisher=

    work page 1983

  15. [15]

    Statistical inference for stochastic processes , volume=

    Identification and inference for multivariate cointegrated and ergodic Gaussian diffusions , author=. Statistical inference for stochastic processes , volume=. 2004 , publisher=

    work page 2004

  16. [16]

    Econometric Theory , volume=

    Identifying restrictions for finite parameter continuous time models with discrete time data , author=. Econometric Theory , volume=. 2017 , publisher=

    work page 2017

  17. [17]

    Stochastic analysis and applications , volume=

    Construction of equivalent stochastic differential equation models , author=. Stochastic analysis and applications , volume=. 2008 , publisher=

    work page 2008

  18. [18]

    IEEE Transactions on Signal Processing , volume=

    Identifying latent stochastic differential equations , author=. IEEE Transactions on Signal Processing , volume=. 2021 , publisher=

    work page 2021

  19. [19]

    Advances in Neural Information Processing Systems , volume=

    Generator Identification for Linear SDEs with Additive and Multiplicative Noise , author=. Advances in Neural Information Processing Systems , volume=

    work page

  20. [20]

    Journal of the Royal Society Interface , volume=

    Identifiability analysis for stochastic differential equation models in systems biology , author=. Journal of the Royal Society Interface , volume=. 2020 , publisher=

    work page 2020

  21. [21]

    Identifiability of SDEs for reaction networks

    Identifiability of SDEs for reaction networks , author=. arXiv preprint arXiv:2505.07638 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  22. [22]

    Cell Systems , volume=

    Noise distorts the epigenetic landscape and shapes cell-fate decisions , author=. Cell Systems , volume=. 2022 , publisher=

    work page 2022

  23. [23]

    Proceedings of the National Academy of Sciences , volume=

    Fundamental limits on dynamic inference from single-cell snapshots , author=. Proceedings of the National Academy of Sciences , volume=. 2018 , publisher=

    work page 2018

  24. [24]

    2025 , school=

    On Qualitative Experimental Design for PDE Parameter Identification Inverse Problems , author=. 2025 , school=

    work page 2025

  25. [25]

    Journal of Mathematical Biology , volume=

    Structural identifiability of linear-in-parameter parabolic PDEs through auxiliary elliptic operators , author=. Journal of Mathematical Biology , volume=. 2025 , publisher=

    work page 2025

  26. [26]

    Mathematical biosciences , volume=

    On structural identifiability , author=. Mathematical biosciences , volume=. 1970 , publisher=

    work page 1970

  27. [27]

    PloS one , volume=

    On finding and using identifiable parameter combinations in nonlinear dynamic systems biology models and COMBOS: a novel web implementation , author=. PloS one , volume=. 2014 , publisher=

    work page 2014

  28. [28]

    Computer methods and programs in biomedicine , volume=

    DAISY: A new software tool to test global identifiability of biological and physiological systems , author=. Computer methods and programs in biomedicine , volume=. 2007 , publisher=

    work page 2007

  29. [29]

    Bioinformatics , volume=

    GenSSI: a software toolbox for structural identifiability analysis of biological models , author=. Bioinformatics , volume=. 2011 , publisher=

    work page 2011

  30. [30]

    Bioinformatics , volume=

    Structural and practical identifiability analysis of partially observed dynamical models by exploiting the profile likelihood , author=. Bioinformatics , volume=. 2009 , publisher=

    work page 2009

  31. [31]

    SIAM Journal on Applied Dynamical Systems , volume=

    Identifiability of linear and linear-in-parameters dynamical systems from a single trajectory , author=. SIAM Journal on Applied Dynamical Systems , volume=. 2014 , publisher=

    work page 2014

  32. [32]

    Bioinformatics , volume=

    Benchmarking tools for a priori identifiability analysis , author=. Bioinformatics , volume=. 2023 , publisher=

    work page 2023

  33. [33]

    arXiv preprint arXiv:2409.06879 , year=

    Joint trajectory and network inference via reference fitting , author=. arXiv preprint arXiv:2409.06879 , year=

    work page arXiv

  34. [34]

    International Conference on Machine Learning , pages=

    Learning population-level diffusions with generative RNNs , author=. International Conference on Machine Learning , pages=. 2016 , organization=

    work page 2016

  35. [35]

    Flow Matching for Generative Modeling

    Flow matching for generative modeling , author=. arXiv preprint arXiv:2210.02747 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  36. [36]

    International conference on machine learning , pages=

    Trajectorynet: A dynamic optimal transport network for modeling cellular dynamics , author=. International conference on machine learning , pages=. 2020 , organization=

    work page 2020

  37. [37]

    Simulation-free schr

    Tong, Alexander and Malkin, Nikolay and Fatras, Kilian and Atanackovic, Lazar and Zhang, Yanlei and Huguet, Guillaume and Wolf, Guy and Bengio, Yoshua , journal=. Simulation-free schr

    work page

  38. [38]

    arXiv preprint arXiv:2002.02798 , volume=

    How to train your neural ode , author=. arXiv preprint arXiv:2002.02798 , volume=

    work page arXiv 2002

  39. [39]

    Improving and generalizing flow-based generative models with minibatch optimal transport

    Improving and generalizing flow-based generative models with minibatch optimal transport , author=. arXiv preprint arXiv:2302.00482 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  40. [40]

    Advances in Neural Information Processing Systems , volume=

    Metric flow matching for smooth interpolations on the data manifold , author=. Advances in Neural Information Processing Systems , volume=

    work page

  41. [41]

    arXiv preprint arXiv:2408.14608 , year=

    Meta flow matching: Integrating vector fields on the wasserstein manifold , author=. arXiv preprint arXiv:2408.14608 , year=

    work page arXiv

  42. [42]

    arXiv preprint arXiv:2510.26645 , year=

    Curly flow matching for learning non-gradient field dynamics , author=. arXiv preprint arXiv:2510.26645 , year=

    work page arXiv

  43. [43]

    Nature communications , volume=

    Generative modeling of single-cell time series with PRESCIENT enables prediction of cell trajectories with interventions , author=. Nature communications , volume=. 2021 , publisher=

    work page 2021

  44. [44]

    International conference on machine learning , pages=

    Action matching: Learning stochastic dynamics from samples , author=. International conference on machine learning , pages=. 2023 , organization=

    work page 2023

  45. [45]

    arXiv preprint arXiv:2507.05107 , year=

    DICE: Discrete inverse continuity equation for learning population dynamics , author=. arXiv preprint arXiv:2507.05107 , year=

    work page arXiv

  46. [46]

    Multi-marginal

    Shen, Yunyi and Berlinghieri, Renato and Broderick, Tamara , journal=. Multi-marginal

    work page

  47. [47]

    Oh SnapMMD! Forecasting Stochastic Dynamics Beyond the Schr

    Berlinghieri, Renato and Shen, Yunyi and Jiang, Jialong and Broderick, Tamara , journal=. Oh SnapMMD! Forecasting Stochastic Dynamics Beyond the Schr

    work page

  48. [48]

    arXiv preprint arXiv:2505.15987 , year=

    Towards Identifiability of Interventional Stochastic Differential Equations , author=. arXiv preprint arXiv:2505.15987 , year=

    work page arXiv

  49. [49]

    arXiv preprint arXiv:2410.22729 , year=

    Identifying Drift, Diffusion, and Causal Structure from Temporal Snapshots , author=. arXiv preprint arXiv:2410.22729 , year=

    work page arXiv

  50. [50]

    The 29th International Conference on Artificial Intelligence and Statistics , year=

    Gradient-flow SDEs have unique transient population dynamics , author=. The 29th International Conference on Artificial Intelligence and Statistics , year=

    work page

  51. [51]

    arXiv preprint arXiv:2406.12616 , year=

    Learning Diffusion at Lightspeed , author=. arXiv preprint arXiv:2406.12616 , year=

    work page arXiv

  52. [52]

    arXiv preprint arXiv:2506.01502 , year=

    Learning of Population Dynamics: Inverse Optimization Meets JKO Scheme , author=. arXiv preprint arXiv:2506.01502 , year=

    work page arXiv

  53. [53]

    arXiv preprint arXiv:2411.01982 , year=

    Learning Controlled Stochastic Differential Equations , author=. arXiv preprint arXiv:2411.01982 , year=

    work page arXiv

  54. [54]

    Trajectory inference via mean-field

    Chizat, L. Trajectory inference via mean-field. Advances in Neural Information Processing Systems , volume=

    work page

  55. [55]

    Cell , volume=

    Optimal-transport analysis of single-cell gene expression identifies developmental trajectories in reprogramming , author=. Cell , volume=. 2019 , publisher=

    work page 2019

  56. [56]

    arXiv preprint arXiv:2102.09204 , year=

    Towards a mathematical theory of trajectory inference , author=. arXiv preprint arXiv:2102.09204 , year=

    work page arXiv

  57. [57]

    2007 , publisher=

    Parameter estimation in stochastic differential equations , author=. 2007 , publisher=

    work page 2007

  58. [58]

    2011 , journal =

    Parameter Estimation for Rough Differential Equations , author =. 2011 , journal =. doi:10.1214/11-AOS893 , urldate =

    work page doi:10.1214/11-aos893 2011

  59. [59]

    17 Generalized method of moments: Econometric applications , series =

    Masao Ogaki , abstract =. 17 Generalized method of moments: Econometric applications , series =. 1993 , booktitle =. doi:https://doi.org/10.1016/S0169-7161(05)80052-5 , url =

    work page doi:10.1016/s0169-7161(05)80052-5 1993

  60. [60]

    Annual Reviews in Control , volume=

    Parameter estimation in stochastic differential equations: an overview , author=. Annual Reviews in Control , volume=. 2000 , publisher=

    work page 2000

  61. [61]

    Journal of Sound and Vibration , volume=

    Stochastic parameter estimation in nonlinear time-delayed vibratory systems with distributed delay , author=. Journal of Sound and Vibration , volume=. 2013 , publisher=

    work page 2013

  62. [62]

    InterStat , volume=

    Maximum likelihood estimation for short time series with replicated observations: A simulation study , author=. InterStat , volume=

    work page

  63. [63]

    Journal of the American Statistical Association , volume=

    A multiresolution method for parameter estimation of diffusion processes , author=. Journal of the American Statistical Association , volume=. 2012 , publisher=

    work page 2012

  64. [64]

    Automatica , volume=

    Parameter estimation in stochastic grey-box models , author=. Automatica , volume=. 2004 , publisher=

    work page 2004

  65. [65]

    Identification of continuous-time models from sampled data , pages=

    Estimation of continuous-time stochastic system parameters , author=. Identification of continuous-time models from sampled data , pages=. 2008 , publisher=

    work page 2008

  66. [66]

    The Annals of Statistics , volume=

    NONPARAMETRIC STATISTICAL INFERENCE FOR DRIFT VECTOR FIELDS OF MULTI-DIMENSIONAL DIFFUSIONS , author=. The Annals of Statistics , volume=. 2020 , publisher=

    work page 2020

  67. [67]

    arXiv preprint arXiv:2312.08493 , year=

    Deep learning-based estimation of time-dependent parameters in Markov models with application to nonlinear regression and SDEs , author=. arXiv preprint arXiv:2312.08493 , year=

    work page arXiv

  68. [68]

    SIAM Journal on Matrix Analysis and Applications , volume=

    Identifiability in continuous Lyapunov models , author=. SIAM Journal on Matrix Analysis and Applications , volume=. 2023 , publisher=

    work page 2023

  69. [69]

    Bernoulli , volume=

    Graphical modeling of stochastic processes driven by correlated noise , author=. Bernoulli , volume=. 2022 , publisher=

    work page 2022

  70. [70]

    Electronic Journal of Probability , year=

    Causal interpretation of stochastic differential equations , author=. Electronic Journal of Probability , year=

    work page

  71. [71]

    Engineering , volume=

    Data-driven discovery of stochastic differential equations , author=. Engineering , volume=. 2022 , publisher=

    work page 2022

  72. [72]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Graphical models for marked point processes based on local independence , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2008 , publisher=

    work page 2008

  73. [73]

    The Annals of Statistics , volume=

    Markov equivalence of marginalized local independence graphs , author=. The Annals of Statistics , volume=. 2020 , publisher=

    work page 2020

  74. [74]

    arXiv preprint arXiv:2406.01161 , year=

    Dynamic Structural Causal Models , author=. arXiv preprint arXiv:2406.01161 , year=

    work page arXiv

  75. [75]

    Learning the structure of a nonstationary vector autoregression

    Signature Kernel Conditional Independence Tests in Causal Discovery for Stochastic Processes , author=. arXiv preprint arXiv:2402.18477 , year=

    work page arXiv

  76. [76]

    Entropy , volume=

    Kernel-based independence tests for causal structure learning on functional data , author=. Entropy , volume=. 2023 , publisher=

    work page 2023

  77. [77]

    arXiv preprint arXiv:2310.02366 , year=

    Stochastic force inference via density estimation , author=. arXiv preprint arXiv:2310.02366 , year=

    work page arXiv

  78. [78]

    International Conference on Artificial Intelligence and Statistics , pages=

    Causal Modeling with Stationary Diffusions , author=. International Conference on Artificial Intelligence and Statistics , pages=. 2024 , organization=

    work page 2024

  79. [79]

    Journal of Public Administration Research and Theory , volume=

    Causal inference methods: Lessons from applied microeconomics , author=. Journal of Public Administration Research and Theory , volume=. 2019 , publisher=

    work page 2019

  80. [80]

    Frontiers in pharmacology , volume=

    Propensity score methods in health technology assessment: principles, extended applications, and recent advances , author=. Frontiers in pharmacology , volume=. 2019 , publisher=

    work page 2019

Showing first 80 references.