pith. sign in

arxiv: 2606.04265 · v1 · pith:TUNTT374new · submitted 2026-06-02 · 🧮 math.OC · cs.LG· cs.NA· math.NA

Nonlocal Mean Field Schr\"{o}dinger Bridge with Learned Interactions

Daisuke Inoue , Mathieu Lauri\`ere , Dante Kalise This is my paper

Pith reviewed 2026-06-28 08:31 UTC · model grok-4.3

classification 🧮 math.OC cs.LGcs.NAmath.NA
keywords mean-field Schrödinger bridgenonlocal interactionsneural network surrogatesalternating algorithmGrönwall stability boundsinteracting particle systemsoptimal transportmean-field control
0
0 comments X

The pith

Neural network surrogates turn the quadratic-cost mean-field Schrödinger bridge into a linear-cost algorithm.

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

The paper shows that nonlocal interaction terms in the mean-field Schrödinger bridge, which normally require quadratic work per step in the number of particles, can be replaced by neural network approximations. This substitution yields a four-stage alternating algorithm that runs in linear time at inference while still connecting given initial and terminal distributions at minimum energy. Stability bounds of Grönwall type are derived to control how the approximation error affects the generated particle trajectories. Experiments on navigation and opinion-dynamics problems confirm that the learned trajectories closely match those obtained from exact evaluation of the interactions.

Core claim

By training neural networks to stand in for the nonlocal interaction functionals inside the mean-field Schrödinger bridge, the authors obtain an alternating optimization procedure whose per-step cost scales linearly rather than quadratically with population size; Grönwall-type estimates then bound the distance between the surrogate trajectories and the exact mean-field solution, and numerical tests on navigation and opinion dynamics reproduce the exact trajectories while shortening training time.

What carries the argument

The four-stage alternating algorithm that interleaves control optimization, neural-network surrogate updates for the nonlocal terms, and particle propagation.

If this is right

  • Per-step computational cost at inference drops from quadratic to linear in population size.
  • Grönwall-type bounds quantify the propagation of surrogate error into the final trajectories.
  • The method reproduces trajectories obtained by exact evaluation on navigation and opinion-dynamics tasks.
  • Overall training time is reduced compared with direct evaluation of the nonlocal terms.

Where Pith is reading between the lines

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

  • The same surrogate strategy could be applied to other mean-field control problems whose interaction terms are currently too expensive to evaluate directly.
  • The stability bounds supply a practical certificate that could be checked after training to decide whether a given surrogate is usable for a target population size.
  • Because the cost reduction is achieved at inference, the approach opens the door to real-time sampling or control of very large interacting particle systems once the networks are trained.

Load-bearing premise

The neural network surrogates can be trained to approximate the nonlocal interaction functionals with enough accuracy that the generated trajectories remain close to the exact mean-field solution.

What would settle it

Run the navigation or opinion-dynamics experiments with the learned surrogates and measure whether the produced trajectories deviate from the analytically evaluated ones by more than the amount allowed by the derived Grönwall bounds.

Figures

Figures reproduced from arXiv: 2606.04265 by Daisuke Inoue, Dante Kalise, Mathieu Lauri\`ere.

Figure 1
Figure 1. Figure 1: GMM: trajectory comparison (forward left, backward right). The learned surrogate repro [PITH_FULL_IMAGE:figures/full_fig_p015_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: GMM: total loss evaluated with analytical [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: GMM: interaction matching loss L N int. 200 400 600 800 1000 N 25 50 75 100 125 150 175 200 Time per iteration (s) Analytic Learned [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: GMM: wall-clock time per outer iteration vs number of agents [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: GMM: proximity diagnostic A(t) across values of w. Each panel shows the case w = 0 and several nonzero values of w. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: V-neck experiment with Gaussian convolution in the running cost: trajectory comparison [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: V-neck: total loss evaluated with analytical [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: V-neck: interaction matching loss L N int. 200 400 600 800 1000 N 200 400 600 800 1000 1200 1400 1600 Time per iteration (s) Analytic Learned [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: V-neck: wall-clock time per outer iteration vs number of agents [PITH_FULL_IMAGE:figures/full_fig_p019_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: V-neck: proximity diagnostic A(t) across values of w at N = 500. Each panel shows the case w = 0 and several nonzero values of w. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Opinion dynamics: PCA snapshots of forward and backward trajectories in [PITH_FULL_IMAGE:figures/full_fig_p021_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Opinion dynamics: total loss evaluated with analytical drift vs wall-clock time. [PITH_FULL_IMAGE:figures/full_fig_p021_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Opinion dynamics: interaction matching loss [PITH_FULL_IMAGE:figures/full_fig_p022_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Opinion dynamics: wall-clock time per outer iteration vs number of agents [PITH_FULL_IMAGE:figures/full_fig_p022_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Opinion dynamics: proximity diagnostic A(t) across values of w at N = 400. Each panel includes the no-interaction baseline. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_15.png] view at source ↗
read the original abstract

The Schr\"odinger Bridge Problem constructs a stochastic process that connects an initial distribution to a terminal distribution with minimum energy. This work considers its mean-field extension, the Mean-Field Schr\"odinger Bridge, for interacting particle systems. With nonlocal interactions, evaluating the resulting particle-dependent distributional terms can scale quadratically with the population size, which makes large-scale problems intractable. We address this bottleneck by approximating the nonlocal interactions with neural network surrogates. The resulting four-stage alternating algorithm reduces the per-step cost from quadratic to linear in the population size at inference. We also derive Gr\"onwall-type stability bounds that show how surrogate errors propagate to the generated trajectories. In numerical experiments on navigation and opinion-dynamics tasks, the proposed method reproduces trajectories obtained with analytical evaluation and reduces training time.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The manuscript extends the mean-field Schrödinger bridge problem to interacting particle systems with nonlocal interactions. It approximates the nonlocal terms via neural-network surrogates, yielding a four-stage alternating algorithm whose per-step inference cost scales linearly rather than quadratically with population size. Gronwall-type stability bounds are derived to control propagation of surrogate approximation error into the generated trajectories. Numerical experiments on navigation and opinion-dynamics tasks report that the method reproduces trajectories obtained by exact analytical evaluation while also reducing training time.

Significance. If the derived Gronwall bounds are non-vacuous and the learned surrogates achieve the accuracy needed for trajectory fidelity, the work removes a central computational obstacle in large-scale mean-field optimal transport and control. The combination of complexity reduction, explicit stability analysis, and empirical validation against exact solutions would make previously intractable nonlocal mean-field problems accessible, with direct relevance to multi-agent systems and statistical physics.

major comments (2)
  1. [§4.2] §4.2, Algorithm 1 and the surrounding text: the four-stage alternation is presented as converging to the mean-field bridge, yet no contraction mapping or monotonicity argument is supplied for the coupled updates of the forward/backward processes and the surrogate parameters; without this, the claim that the algorithm solves the original problem rests on empirical observation alone.
  2. [§5.3] §5.3, the Gronwall bound (presumably Eq. (12) or (13)): the derivation assumes the surrogate error is bounded uniformly in time and independent of the particle measure, but the training procedure (described in §4.3) updates the network on samples drawn from the current iterate of the measure; this dependence is not folded into the error-propagation estimate, so the bound may not directly control the closed-loop trajectory error.
minor comments (2)
  1. [§5.1] The abstract and §5.1 state that training time is reduced, yet no wall-clock timings, iteration counts, or hardware specifications are supplied to quantify the improvement relative to the quadratic baseline.
  2. Notation for the nonlocal interaction kernel K and its neural surrogate K_θ is introduced without an explicit statement of the function space in which the approximation error is measured (e.g., L^∞ or Wasserstein).

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback. We address the two major comments point by point below.

read point-by-point responses

  1. Referee: [§4.2] §4.2, Algorithm 1 and the surrounding text: the four-stage alternation is presented as converging to the mean-field bridge, yet no contraction mapping or monotonicity argument is supplied for the coupled updates of the forward/backward processes and the surrogate parameters; without this, the claim that the algorithm solves the original problem rests on empirical observation alone.

    Authors: We agree that no contraction mapping or monotonicity argument is supplied for convergence of the four-stage alternation. The algorithm is constructed by alternating between the forward and backward processes and the surrogate updates, following the structure of the mean-field Schrödinger bridge; its ability to recover the correct trajectories is shown empirically on the navigation and opinion-dynamics examples. In revision we will add an explicit statement that global convergence is observed numerically rather than proven theoretically and list this as an open direction. revision: partial

  2. Referee: [§5.3] §5.3, the Gronwall bound (presumably Eq. (12) or (13)): the derivation assumes the surrogate error is bounded uniformly in time and independent of the particle measure, but the training procedure (described in §4.3) updates the network on samples drawn from the current iterate of the measure; this dependence is not folded into the error-propagation estimate, so the bound may not directly control the closed-loop trajectory error.

    Authors: The Gronwall bounds are derived under the standing assumption that the surrogate error remains uniformly bounded in time. The online training procedure does draw samples from the current measure iterate, so the error bound is not shown to be preserved in closed loop. We will revise §5.3 to state the assumption more precisely and to note that a fully coupled analysis of the training-measure interaction lies beyond the present stability estimate. revision: partial

Circularity Check

0 steps flagged

No significant circularity

full rationale

The derivation introduces neural network surrogates to approximate nonlocal interaction terms, yielding an alternating four-stage algorithm whose per-step cost scales linearly rather than quadratically with population size. Separate Grönwall-type bounds are derived to bound surrogate error propagation into trajectories; these bounds rest on standard differential inequality techniques and do not presuppose the learned surrogates. Experiments compare generated trajectories against analytically evaluable cases, providing an external check. No equation reduces a claimed prediction to a fitted quantity by construction, no uniqueness result is imported via self-citation, and no ansatz is smuggled through prior work. The central claims therefore remain independent of their own inputs.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the mean-field limit for interacting particles and the existence of sufficiently accurate neural approximations whose errors are controlled by Gronwall estimates; no explicit free parameters beyond learned network weights are stated.

free parameters (1)
  • neural network parameters
    Weights of the surrogate networks are fitted during the alternating training procedure to approximate nonlocal interaction terms.
axioms (1)
  • domain assumption The mean-field limit holds and the nonlocal interaction functional admits a neural approximation with controllable error.
    Invoked when replacing exact quadratic-cost terms with surrogates and when applying Gronwall bounds to the resulting trajectories.

pith-pipeline@v0.9.1-grok · 5673 in / 1242 out tokens · 20722 ms · 2026-06-28T08:31:19.197149+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

55 extracted references · 36 canonical work pages

  1. [1]

    Schr¨ odinger, ¨ uber die Umkehrung der Naturgesetze, Angewandte Chemie 44 (30) (1931) 636– 636.doi:10.1002/ange.19310443014

    E. Schr¨ odinger, ¨ uber die Umkehrung der Naturgesetze, Angewandte Chemie 44 (30) (1931) 636– 636.doi:10.1002/ange.19310443014

    work page doi:10.1002/ange.19310443014 1931

  2. [2]

    C. L´ eonard, A survey of the Schr¨ odinger problem and some of its connections with optimal transport, Discrete and Continuous Dynamical Systems 34 (4) (2014) 1533–1574.doi:10.3934/ dcds.2014.34.1533

    2014

  3. [3]

    Y. Chen, T. T. Georgiou, M. Pavon, Stochastic Control Liaisons: Richard Sinkhorn Meets Gas- pard Monge on a Schr¨ odinger Bridge, SIAM Review 63 (2) (2021) 249–313.doi:10.1137/ 20M1339982

    2021

  4. [4]

    Y. Chen, T. T. Georgiou, M. Pavon, Optimal Steering of a Linear Stochastic System to a Final Probability Distribution, Part I, IEEE Transactions on Automatic Control 61 (5) (2016) 1158– 1169.doi:10.1109/TAC.2015.2457784

    work page doi:10.1109/tac.2015.2457784 2016

  5. [5]

    Y. Chen, T. T. Georgiou, M. Pavon, Optimal Steering of a Linear Stochastic System to a Final Probability Distribution, Part II, IEEE Transactions on Automatic Control 61 (5) (2016) 1170– 1180.doi:10.1109/TAC.2015.2457791

    work page doi:10.1109/tac.2015.2457791 2016

  6. [6]

    Y. Chen, T. T. Georgiou, M. Pavon, Optimal Steering of a Linear Stochastic System to a Final Probability Distribution—Part III, IEEE Transactions on Automatic Control 63 (9) (2018) 3112– 3118.doi:10.1109/TAC.2018.2791362

    work page doi:10.1109/tac.2018.2791362 2018

  7. [7]

    K. F. Caluya, A. Halder, Wasserstein Proximal Algorithms for the Schr¨ odinger Bridge Problem: Density Control With Nonlinear Drift, IEEE Transactions on Automatic Control 67 (3) (2022) 1163–1178.doi:10.1109/TAC.2021.3060704. 28

    work page doi:10.1109/tac.2021.3060704 2022

  8. [8]

    K. F. Caluya, A. Halder, Reflected Schr¨ odinger Bridge: Density Control with Path Constraints, in: 2021 American Control Conference (ACC), 2021, pp. 1137–1142.doi:10.23919/ACC50511. 2021.9482813

    work page doi:10.23919/acc50511 2021

  9. [9]

    Vargas, P

    F. Vargas, P. Thodoroff, A. Lamacraft, N. Lawrence, Solving Schr¨ odinger Bridges via Maximum Likelihood, Entropy 23 (9) (2021) 1134.doi:10.3390/e23091134

    work page doi:10.3390/e23091134 2021

  10. [10]

    Chen, G.-H

    T. Chen, G.-H. Liu, E. Theodorou, Likelihood Training of Schr¨ odinger Bridge using Forward- Backward SDEs Theory, in: International Conference on Learning Representations, 2022

    2022

  11. [11]

    De Bortoli, J

    V. De Bortoli, J. Thornton, J. Heng, A. Doucet, Diffusion Schr¨ odinger Bridge with Applications to Score-Based Generative Modeling, in: Advances in Neural Information Processing Systems, Vol. 34, Curran Associates, Inc., 2021, pp. 17695–17709

    2021

  12. [12]

    G.-H. Liu, A. Vahdat, D.-A. Huang, E. Theodorou, W. Nie, A. Anandkumar, I 2SB: Image-to- Image Schr¨ odinger Bridge, in: Proceedings of the 40th International Conference on Machine Learning, PMLR, 2023, pp. 22042–22062

    2023

  13. [13]

    Y. Shi, V. De Bortoli, A. Campbell, A. Doucet, Diffusion Schr¨ odinger Bridge Matching, in: Thirty-Seventh Conference on Neural Information Processing Systems, 2023

    2023

  14. [14]

    Zheng, Q

    T. Zheng, Q. Han, H. Lin, Transporting Robotic Swarms Via Mean-Field Feedback Control, IEEE Transactions on Automatic Control 67 (8) (2022) 4170–4177.doi:10.1109/TAC.2021.3108672

    work page doi:10.1109/tac.2021.3108672 2022

  15. [15]

    K. Cui, M. Li, C. Fabian, H. Koeppl, Scalable Task-Driven Robotic Swarm Control via Collision Avoidance and Learning Mean-Field Control, in: 2023 IEEE International Conference on Robotics and Automation (ICRA), 2023, pp. 1192–1199.doi:10.1109/ICRA48891.2023.10161498

    work page doi:10.1109/icra48891.2023.10161498 2023

  16. [16]

    Rapakoulias, A

    G. Rapakoulias, A. Reza Pedram, P. Tsiotras, Steering Large Agent Populations Using Mean- Field Schr¨ odinger Bridges With Gaussian Mixture Models, IEEE Control Systems Letters 9 (2025) 1760–1765.doi:10.1109/LCSYS.2025.3581859

    work page doi:10.1109/lcsys.2025.3581859 2025

  17. [17]

    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, E. S. Lander, Optimal-Transport Analysis of Single-Cell Gene Expression Identifies Developmental Trajectories in Reprogramming, Cell 176 (4) (2019) 928–943.e22.do...

    2019

  18. [18]

    A. Tong, J. Huang, G. Wolf, D. V. Van Dijk, S. Krishnaswamy, TrajectoryNet: A Dynamic Opti- mal Transport Network for Modeling Cellular Dynamics, in: Proceedings of the 37th International Conference on Machine Learning, PMLR, 2020, pp. 9526–9536

    2020

  19. [19]

    Zhang, Z

    Z. Zhang, Z. Wang, Y. Sun, T. Li, P. Zhou, Modeling Cell Dynamics and Interactions with Unbalanced Mean Field Schr¨ odinger Bridge, in: The Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

    2025

  20. [20]

    Sznitman,Topics in propagation of chaos, in École d’Été de Probabilités de Saint-Flour XIX—1989, vol

    A.-S. Sznitman, Topics in propagation of chaos, in: D. L. Burkholder, E. Pardoux, A.-S. Sznitman, P.-L. Hennequin (Eds.), Ecole d’Et´ e de Probabilit´ es de Saint-Flour XIX — 1989, Springer, Berlin, Heidelberg, 1991, pp. 165–251.doi:10.1007/BFb0085169

    work page doi:10.1007/bfb0085169 1989

  21. [21]

    Lacker, A general characterization of the mean field limit for stochastic differential games, Probability Theory and Related Fields 165 (2016) 581–648.doi:10.1007/s00440-015-0641-9

    D. Lacker, A general characterization of the mean field limit for stochastic differential games, Probability Theory and Related Fields 165 (2016) 581–648.doi:10.1007/s00440-015-0641-9

    work page doi:10.1007/s00440-015-0641-9 2016

  22. [22]

    Carmona, F

    R. Carmona, F. Delarue, Probabilistic Theory of Mean Field Games with Applications I: Mean Field FBSDEs, Control, and Games, Vol. 83 of Probability Theory and Stochastic Modelling, Springer International Publishing, Cham, 2018.doi:10.1007/978-3-319-58920-6

    work page doi:10.1007/978-3-319-58920-6 2018

  23. [23]

    Backhoff, G

    J. Backhoff, G. Conforti, I. Gentil, C. L´ eonard, The mean field Schr¨ odinger problem: Ergodic behavior, entropy estimates and functional inequalities, Probability Theory and Related Fields 178 (1) (2020) 475–530.doi:10.1007/s00440-020-00977-8. 29

    work page doi:10.1007/s00440-020-00977-8 2020

  24. [24]

    Hern´ andez, L

    C. Hern´ andez, L. Tangpi, Propagation of Chaos for Mean Field Schr¨ odinger Problems, SIAM Journal on Control and Optimization 63 (1) (2025) 112–150.doi:10.1137/23M1566716

    work page doi:10.1137/23m1566716 2025

  25. [25]

    H. Pham, X. Wei, Bellman equation and viscosity solutions for mean-field stochastic control problem, ESAIM: Control, Optimisation and Calculus of Variations 24 (1) (2018) 437–461.doi: 10.1051/cocv/2017019

    work page doi:10.1051/cocv/2017019 2018

  26. [26]

    G.-H. Liu, T. Chen, O. So, E. Theodorou, Deep Generalized Schr¨ odinger Bridge, Advances in Neural Information Processing Systems 35 (2022) 9374–9388

    2022

  27. [27]

    J. Han, A. Jentzen, W. E, Solving High-Dimensional Partial Differential Equations using Deep Learning, Proceedings of the National Academy of Sciences 115 (34) (2018) 8505–8510.doi: 10.1073/pnas.1718942115

    work page doi:10.1073/pnas.1718942115 2018

  28. [28]

    Ruthotto, S

    L. Ruthotto, S. J. Osher, W. Li, L. Nurbekyan, S. W. Fung, A machine learning framework for solving high-dimensional mean field game and mean field control problems, Proceedings of the National Academy of Sciences 117 (17) (2020) 9183–9193.doi:10.1073/pnas.1922204117

    work page doi:10.1073/pnas.1922204117 2020

  29. [29]

    J. Han, R. Hu, J. Long, Learning High-Dimensional McKean–Vlasov Forward-Backward Stochas- tic Differential Equations with General Distribution Dependence, SIAM Journal on Numerical Analysis 62 (1) (2024) 1–24.doi:10.1137/22M151861X

    work page doi:10.1137/22m151861x 2024

  30. [30]

    R. Hu, B. Jin, M. Lauri` ere, J. Zhang, Deep Signature Approach for McKean-Vlasov FBSDEs in a Random Environment (Nov. 2025).arXiv:2511.09112,doi:10.48550/arXiv.2511.09112

    work page doi:10.48550/arxiv.2511.09112 2025

  31. [31]

    Dayanikli, M

    G. Dayanikli, M. Lauri` ere, J. Zhang, Deep Learning for Population-Dependent Controls in Mean Field Control Problems with Common Noise, in: Proceedings of the 23rd International Confer- ence on Autonomous Agents and Multiagent Systems, AAMAS ’24, International Foundation for Autonomous Agents and Multiagent Systems, Richland, SC, 2024, pp. 2231–2233

    2024

  32. [32]

    Z. Cao, K. Das, N. Langren´ e, M. Lauri` ere, Scalable Method for Mean Field Control with Ker- nel Interactions via Random Fourier Features (Jan. 2026).arXiv:2601.01175,doi:10.48550/ arXiv.2601.01175

    Pith/arXiv arXiv 2026

  33. [33]

    Dai Pra, A stochastic control approach to reciprocal diffusion processes, Applied Mathematics and Optimization 23 (1) (1991) 313–329.doi:10.1007/BF01442404

    P. Dai Pra, A stochastic control approach to reciprocal diffusion processes, Applied Mathematics and Optimization 23 (1) (1991) 313–329.doi:10.1007/BF01442404

    work page doi:10.1007/bf01442404 1991

  34. [34]

    Cuturi, Sinkhorn Distances: Lightspeed Computation of Optimal Transport, in: Advances in Neural Information Processing Systems, Vol

    M. Cuturi, Sinkhorn Distances: Lightspeed Computation of Optimal Transport, in: Advances in Neural Information Processing Systems, Vol. 26, Curran Associates, Inc., 2013, pp. 2292–2300

    2013

  35. [35]

    Peyr´ e, M

    G. Peyr´ e, M. Cuturi, Computational Optimal Transport, Foundations and Trends®in Machine Learning 11 (5-6) (2019) 355–607

    2019

  36. [36]

    Gushchin, S

    N. Gushchin, S. Kholkin, E. Burnaev, A. Korotin, Light and Optimal Schr¨ odinger Bridge Match- ing, in: Proceedings of the 41st International Conference on Machine Learning, Vol. 235 of ICML’24, JMLR.org, Vienna, Austria, 2024, pp. 17100–17122

    2024

  37. [37]

    F. Lu, M. Zhong, S. Tang, M. Maggioni, Nonparametric inference of interaction laws in systems of agents from trajectory data, Proceedings of the National Academy of Sciences 116 (29) (2019) 14424–14433.doi:10.1073/pnas.1822012116

    work page doi:10.1073/pnas.1822012116 2019

  38. [38]

    F. Lu, M. Maggioni, S. Tang, Learning interaction kernels in heterogeneous systems of agents from multiple trajectories, Journal of Machine Learning Research 22 (32) (2021) 1–67

    2021

  39. [39]

    F. Lu, M. Maggioni, S. Tang, Learning Interaction Kernels in Stochastic Systems of Interacting Particles from Multiple Trajectories, Foundations of Computational Mathematics 22 (4) (2022) 1013–1067.doi:10.1007/s10208-021-09521-z

    work page doi:10.1007/s10208-021-09521-z 2022

  40. [40]

    R. Yao, X. Chen, Y. Yang, Mean-field nonparametric estimation of interacting particle systems, in: Proceedings of Thirty Fifth Conference on Learning Theory, PMLR, 2022, pp. 2242–2275. 30

    2022

  41. [41]

    Q. Lang, F. Lu, Learning Interaction Kernels in Mean-Field Equations of First-Order Systems of Interacting Particles, SIAM Journal on Scientific Computing 44 (1) (2022) A260–A285.doi: 10.1137/20M1377072

    work page doi:10.1137/20m1377072 2022

  42. [42]

    Y. Shi, W. Yang, L. Hong, Extracting interaction kernels for many-particle systems by a two-phase approach, Physics of Fluids 37 (3) (2025) 033331.doi:10.1063/5.0255706

    work page doi:10.1063/5.0255706 2025

  43. [43]

    J. Yang, X. Ye, R. Trivedi, H. Xu, H. Zha, Learning Deep Mean Field Games for Modeling Large Population Behavior, in: International Conference on Learning Representations, 2018

    2018

  44. [44]

    L. Ding, W. Li, S. Osher, W. Yin, A Mean Field Game Inverse Problem, Journal of Scientific Computing 92 (1) (2022) 7.doi:10.1007/s10915-022-01825-8

    work page doi:10.1007/s10915-022-01825-8 2022

  45. [45]

    H. Liu, C. Mou, S. Zhang, Inverse problems for mean field games, Inverse Problems 39 (8) (2023) 085003.doi:10.1088/1361-6420/acdd90

    work page doi:10.1088/1361-6420/acdd90 2023

  46. [46]

    Z. Mo, X. Chen, X. Di, E. Iacomini, C. Segala, M. Herty, M. Lauriere, A Game-Theoretic Framework for Generic Second-Order Traffic Flow Models Using Mean Field Games and Ad- versarial Inverse Reinforcement Learning, Transportation Science 58 (6) (2024) 1403–1426. doi:10.1287/trsc.2024.0532

    work page doi:10.1287/trsc.2024.0532 2024

  47. [47]

    Fiedler, M

    C. Fiedler, M. Herty, C. Segala, S. Trimpe, Recent kernel methods for interacting particle systems: First numerical results, European Journal of Applied Mathematics 36 (2) (2025) 464–489.doi: 10.1017/S0956792524000706

    work page doi:10.1017/s0956792524000706 2025

  48. [48]

    L´ eger, W

    F. L´ eger, W. Li, Hopf–Cole transformation via generalized Schr¨ odinger bridge problem, Journal of Differential Equations 274 (2021) 788–827.doi:10.1016/j.jde.2020.10.029

    work page doi:10.1016/j.jde.2020.10.029 2021

  49. [49]

    Chazelle, Q

    B. Chazelle, Q. Jiu, Q. Li, C. Wang, Well-posedness of the limiting equation of a noisy consensus model in opinion dynamics, Journal of Differential Equations 263 (1) (2017) 365–397.doi: 10.1016/j.jde.2017.02.036

    work page doi:10.1016/j.jde.2017.02.036 2017

  50. [50]

    Bernardo, C

    C. Bernardo, C. Altafini, A. Proskurnikov, F. Vasca, Bounded confidence opinion dynamics: A survey, Automatica 159 (2024) 111302.doi:10.1016/j.automatica.2023.111302

    work page doi:10.1016/j.automatica.2023.111302 2024

  51. [51]

    Garnier, G

    J. Garnier, G. Papanicolaou, T.-W. Yang, Consensus Convergence with Stochastic Effects, Viet- nam Journal of Mathematics 45 (1) (2017) 51–75.doi:10.1007/s10013-016-0190-2

    work page doi:10.1007/s10013-016-0190-2 2017

  52. [52]

    Ceragioli, P

    F. Ceragioli, P. Frasca, Continuous-time Discontinuous Equations in Bounded Confidence Opinion Dynamics, IFAC Proceedings Volumes 44 (1) (2011) 1986–1990.doi:10.3182/ 20110828-6-IT-1002.02106

    arXiv 2011

  53. [53]

    Ceragioli, P

    F. Ceragioli, P. Frasca, Continuous and discontinuous opinion dynamics with bounded confidence, Nonlinear Analysis: Real World Applications 13 (3) (2012) 1239–1251.doi:10.1016/j.nonrwa. 2011.10.002

    work page doi:10.1016/j.nonrwa 2012

  54. [54]

    G. Albi, S. Bicego, D. Kalise, Control of high-dimensional collective dynamics by deep neural feedback laws and kinetic modelling, Journal of Computational Physics 539 (2025) 114229.doi: 10.1016/j.jcp.2025.114229

    work page doi:10.1016/j.jcp.2025.114229 2025

  55. [55]

    Bicego, D

    S. Bicego, D. Kalise, G. A. Pavliotis, Computation and control of unstable steady states for mean field multiagent systems, Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 481 (2311) (2025) 20240476.doi:10.1098/rspa.2024.0476. 31

    work page doi:10.1098/rspa.2024.0476 2025