Pith. sign in

REVIEW 4 major objections 5 minor 49 references

Neural Kolmogorov Equations: Parallelizable Learning of Stochastic Dynamics under General Noise

T0 review · 4 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read Neural Kolmogorov Equations learn stochastic dynamics by modeling the evolution of probability densities rather than individual trajectories, enabling parallel-in-time training and recovery of general Lévy-type noise, including coupled Gaus

desk verdict A solid, original training scheme for neural SDEs via projected Kolmogorov operators, weakened only by overclaimed noise identifiability and unreported threshold parameters. read the letter →

arxiv 2607.19173 v1 pith:XLCGLW5T submitted 2026-07-21 cs.LG

Arthur Bizzi , Olga Fink This is my paper · ORCID
classification cs.LG MSC 60H1060J7465C3068T07
keywords NeuralstochasticdifferentialequationsKolmogorovForwardEquationLévyprocessesoperatorsplittingGaussianmixturessimulation-freetrainingjumpLagrangianGalerkinmethods
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 proposes Neural Kolmogorov Equations (NKEs), a method that learns stochastic differential equations by modeling how probability densities evolve under the Kolmogorov Forward equation. The key claim is that the generator of this equation—advection, diffusion, and jump terms—can be directly parameterized by neural networks, allowing the model to infer the structure of the driving noise without parametric assumptions. By projecting density evolution onto Gaussian mixtures and using operator splitting, NKEs avoid autoregressive trajectory simulation and become parallelizable in time. If correct, the method provides a unified treatment of Gaussian and jump noise with competitive accuracy on the paper's benchmarks, including systems with coupled noise and discontinuous jumps.

What carries the argument

The central mechanism is a Lagrangian Galerkin projection onto K Gaussian mixture components, which reduces the infinite-dimensional Kolmogorov forward generator to an ODE on the parameters (weights, means, covariances). The generator splits as A + D + J: advection transports means, diffusion spreads covariances, and the jump semigroup convolves the mixture with a learned conditional Gaussian mixture. Strang splitting (e^{Ls} ≈ e^{As/2} e^{Js} e^{Ds} e^{As/2}) is used to train each operator separately, making the training loss parallelizable in time and second-order accurate in the absence of jumps.

What would settle it

Simulate a compound-Poisson jump-diffusion where the jump scale equals the diffusion increment scale (e.g., jump standard deviation equals sqrt(dt) times the diffusion coefficient), train NKEs, and compare the learned drift, diffusion, and jump mixture to the ground truth. If the recovered parameters deviate systematically or the generated trajectory Wasserstein distance does not beat an Euler-Maruyama baseline, the jump-thresholding decomposition is the culprit.

Watch

Extended reading notes

Core claim

The core discovery is that the Kolmogorov Forward generator, decomposed into advection, diffusion, and jump components, can be learned as a neural operator on the finite-dimensional space of Gaussian mixtures, turning SDE identification into a deterministic regression problem on density moments. Advection is recovered from the mean dynamics of localized Gaussian clusters, diffusion from their covariance growth, and the jump semigroup from thresholded jump increments via conditional Gaussian mixture deconvolution. This operator-level formulation avoids the need for autoregressive simulation and naturally supports non-diagonal diffusion tensors and general jump laws.

Load-bearing premise

The method's separation of jumps from diffusion relies on jumps being visibly larger than ordinary random fluctuations at the observation time step; if a jump is about the same size as a typical diffusion increment, the split cannot be identified and both the learned jump semigroup and the advection–diffusion operator become biased.

Editorial extensions

If this is right

  • NKEs can learn coupled, non-diagonal diffusion tensors without maximum-likelihood estimation, which is not possible with standard EM-based neural SDE training.
  • Training is parallelizable across time steps because each loss term depends only on local clusters and finite differences, eliminating the need for autoregressive rollout.
  • The same framework handles Gaussian and jump noise in one generative model, learning the jump distribution itself rather than assuming a fixed parametric family.
  • On the paper's benchmarks, NKEs achieve lower mean, covariance, and Wasserstein errors than Euler-Maruyama matching, trajectory flow matching, and SDE-GAN baselines, while keeping training times comparable.

Reading between the lines

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

  • Editorial inference: If the density-operator view is correct, the method could be extended to non-uniform sampling by learning the jump generator directly in a basis that supports the generator, although the paper notes this is out of reach for its Gaussian mixture representation.
  • Editorial inference: The jump-thresholding assumption implies a stress test: performance should degrade gracefully as jump sizes shrink toward the diffusion scale; identifying that boundary would define the method's practical application envelope.
  • Editorial inference: The same Lagrangian projection idea could be applied to other parametric families (e.g., Student-t or skewed mixtures) to handle fat tails or state-dependent jump distributions with less ill-posed projection.
  • Editorial inference: Because the projection for K>1 Gaussian mixtures is non-unique, state-dependent jump learning may be the first place to fail in higher dimensions; regularizing the mixture parameters across the state space would be a natural extension.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes Neural Kolmogorov Equations (NKEs), a method for learning stochastic dynamics from discretely observed trajectories by fitting the Kolmogorov forward generator rather than simulating SDE paths. The generator is projected onto a Gaussian-mixture manifold, yielding ODEs for the evolution of mixture means and covariances under advection and diffusion, plus a finite-time jump semigroup modeled by a mixture-density network. Training uses a Strang-splitting objective, with a MAD-based threshold detector to separate jump increments from advection–diffusion increments. Experiments on Lorenz, Black–Scholes, double-well jump-diffusion, and molecular-dynamics coarse-graining benchmarks compare NKEs with Euler–Maruyama, Trajectory Flow Matching, and SDE-GAN baselines on mean, covariance, and Wasserstein errors.

Significance. The core idea is attractive and potentially useful: formulating neural-SDE learning as moment evolution on probability densities is a principled way to avoid autoregressive trajectory simulation and to handle coupled (non-diagonal) Gaussian noise. The basic advection and diffusion moment identities (Eqs. 13–15 and 17–19) are correct, and Appendix C gives plausible, even if sketched, error bounds for short-time Gaussianity, small-jump diffusion approximation, and jump freezing. If the central claims were fully supported, the work would be a meaningful step beyond simulation-free Gaussian-noise SDE training. However, the paper's headline claim of learning general Lévy-type forcing directly is currently not supported: the jump/diffusion separation rests on an unreported threshold sensitivity, and the small-jump/diffusion equivalence (Prop. 4) implies an identifiability limitation that is not addressed. The experiments are also not consistently faster than the parallel EM baseline.

major comments (4)
  1. [§5.2.2, Eq. (31)] The jump detector threshold depends on κsens, but κsens is never reported and no sensitivity analysis is given. This is load-bearing because the entire two-stage training procedure first separates jump increments from advection–diffusion increments using this detector. For pure-Gaussian diffusion increments, a threshold at κsens=1 flags roughly 32% of increments as jumps; larger κsens will miss true jumps comparable to the diffusion scale. The paper must report the chosen κsens, justify it, and show that the learned diffusion and jump models are stable over a range of κsens. Without this, the benchmark numbers in Tables 1–4 are conditional on an undisclosed hyperparameter.
  2. [Appendix C.2, Prop. 4 and Eq. (20)] Proposition 4 shows that small jumps act, to O(ε^3), like an effective advection–diffusion with drift β=∫_{|y|<ε} y h(dy) and covariance γ=∫ yy^T h(dy). At any fixed observation scale, many different pairs (g, h^{<ε}) produce indistinguishable short-time semigroups, so the learned Lévy triplet is not identifiable from discretely sampled trajectories. This directly undermines the abstract's claim that NKEs 'learn general Lévy-type stochastic forcing directly.' The manuscript should either limit the claim to the identifiable equivalence class, or present a concrete identifiability analysis and a test showing that varying jump sizes around the diffusion scale does not materially change the recovered advection–diffusion operator.
  3. [Appendix B.3 and §7] The paper explicitly acknowledges that the projection onto Gaussian mixtures is non-unique and often ill-posed for K>1, and that this 'leads to discontinuities and poor convergence when learning complex jump patterns.' Yet the jump-learning step uses a Gaussian-mixture MDN with K_J components, i.e., K>1 projection. The manuscript does not report K_J, the component-pruning thresholds, or the covariance regularization mentioned in Appendix B.3.1. As a result, the jump semigroup in the double-well experiment (Table 3) is not a well-defined object. The authors need to specify the projection regularization and demonstrate that the learned jump law is stable under changes in K_J and initialization.
  4. [Tables 1–4, training time columns] The paper claims 'improved training efficiency,' but the reported training times do not support this against the Euler–Maruyama baseline: NKE is slower than EM in Table 1 (42.4 vs 33.9), Table 3 (1545 vs 1152), and Table 4 (323.1 vs 206.3); only in Table 2 is it faster (4.1 vs 35.7). If the claim is meant only relative to SDE-GAN, it should be stated as such. The efficiency argument should be reframed as parallel-in-time capability rather than wall-clock speed, or the comparison should be made on equal footing with the same hardware and implementation maturity.
minor comments (5)
  1. [Abstract and §7] The abstract's 'learn general Lévy-type stochastic forcing directly' is inconsistent with the limitations stated in §7 and with the identifiability issue raised in Appendix C.2. The wording should be softened or accompanied by the required identifiability conditions.
  2. [Appendix A] Typo: 'provie' should be 'provide' in the first paragraph of Appendix A.
  3. [Appendix E] Typo: 'Trajecotry Flow Matching' should be 'Trajectory Flow Matching'.
  4. [§5.4, Eq. (42)] The jump loss is written as a negative log-likelihood of target points under the convolved mixture. It would help to state explicitly that this is a conditional maximum-likelihood objective and to describe how the MDN's output covariances are constrained to be positive definite.
  5. [§4.4, Eq. (21)] The notation H^s for the finite-time jump distribution is introduced, but it is not immediately clear that H^s is a probability measure on jump sizes for the interval [t,t+s] and that the no-jump mass is included. A one-sentence clarification would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: NKE losses are direct moment-matching regressions on empirical data, and the identifiability limitations are explicitly acknowledged.

full rationale

I found no circular step. The derivation chain is: KFE semigroup -> Lagrangian Gaussian-mixture projection -> moment equations -> neural parameterizations -> moment-matching losses -> held-out evaluation. Each learned operator is fit to empirical moments or target clouds that do not encode the model output as an assumption. The MAD thresholding in Eq. 31 and the small-jump diffusion approximation in Prop. 4 are identifiability/modeling choices, not circular definitions; the paper explicitly states that undetected small jumps are modeled as effective diffusion ('In practice, any jumps too small to be detected will be modelled as effective diffusion, as discussed'), so there is no hidden equivalence between input and prediction. The limitations section also concedes that NKEs learn a fixed-step semigroup and that Gaussian-mixture projections for K>1 are non-unique; these reduce scope and correctness guarantees but do not make the derivation circular. References are to external classical results (Levy-Khintchine, Trotter/Strang splitting, MAD) with no self-citation chain, and the benchmarks use held-out test trajectories, so fitted parameters are not presented as predictions of the same data they were fit to.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The paper's contributions rest on a careful chain of approximations: the Gaussian-mixture projection, the MAD jump separation, and the finite-time jump semigroup. These are mathematical/statistical assumptions rather than new physical entities. The main unstated knobs are the jump-detection sensitivity and the number of nearest neighbors/clusters, which are not reported in the provided text.

free parameters (3)
  • κsens (jump detector sensitivity)
    Controls the MAD-based jump detector (Eq. 31); determines which increments are labeled jump-contaminated and thus which data train the advection-diffusion vs jump modules. No value or tuning procedure is reported.
  • κnn (nearest-neighbor cluster size)
    Number of nearest neighbors forming each Gaussian cluster (Eq. 29). Sets the scale of the local Gaussian approximation. Value not reported.
  • K_J (number of mixture components in jump MDN)
    Number of Gaussian components in the learned finite-time jump distribution (Eq. 41). Chosen by hand; not reported in the available text.
assumptions (6)
  • standard math Lévy–Khintchine representation / Lévy-Itô decomposition of the SDE
    Used in Eq. (1) and Eqs. (2)-(5) to write the SDE and the KFE with advection, diffusion, and jump terms.
  • domain assumption Assumptions (A1)-(A3): drift/diffusion C^1 and uniformly Lipschitz; jump measures have bounded moments
    Needed for well-posedness of the SDE and for the exchange of expectations in the moment derivations.
  • domain assumption Short-time Gaussian closure: e^{(A+D)s}G_{μ,Σ} ≈ G_{μ+sΔμ,Σ+sΔΣ}
    Justified in Appendix C.1 under C^3 bounds on f and a; underpins the entire Gaussian-mixture projection and the advection/diffusion losses.
  • domain assumption Jump-freezing approximation at the Gaussian center (Eq. 23)
    Assumes H^s is Lipschitz in space and that the packet is sufficiently concentrated; Proposition 5 gives L1 error O(sqrt(tr Σ)).
  • domain assumption MAD jump detectability (Eq. 31)
    Assumes jump-induced increments are outliers relative to diffusion increments at the observation scale; this is qualitative and not proven.
  • domain assumption Strang splitting (Eq. 27) is accurate for the operator ordering
    The paper notes it is second-order only in the absence of jumps; with jumps the ordering may be only first-order, affecting the loss design.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Neural Kolmogorov Equations: Parallelizable Learning of Stochastic Dynamics under General Noise." pith.science (2026). https://pith.science/paper/XLCGLW5T

@misc{pith2026260719173,
  author       = {Pith},
  title        = {Pith review of: Neural Kolmogorov Equations: Parallelizable Learning of Stochastic Dynamics under General Noise},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XLCGLW5T}},
  note         = {Machine review of arXiv:2607.19173}
}
read the original abstract

Neural stochastic differential equations (SDEs) have emerged as powerful tools for learning noisy or stochastic dynamics directly from data; however, existing approaches largely assume uncoupled and continuous noise, limiting their applicability to realistic stochastic drivers, and often scale poorly in time, requiring expensive autoregressive training. To address these limitations, we propose Neural Kolmogorov Equations (NKEs), a deterministic, infinite-dimensional reformulation of Neural SDEs based on the Kolmogorov Forward equation, transforming the learning problem from modelling individual stochastic trajectories to modelling the evolution of probability densities. NKEs learn general L\'evy-type stochastic forcing directly through the operator structure of the KFE, and enable parallel-in-time training via a Lagrangian Galerkin projection and operator splitting. We evaluate NKEs on several stochastic benchmarks, including systems with coupled noise and jump processes, and verify that NKEs provide flexible models that accurately recover deterministic and stochastic dynamics with competitive predictive accuracy and improved training efficiency. Code and pretrained models will be released.

Figures

Figures reproduced from arXiv: 2607.19173 by the authors.

Figure 1
Figure 1. A stochastic harmonic oscillator. The Kolmogorov Equations treat individual particle [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The forward Kolmogorov semigroup e L on a Gaussian. Advection transports its mean and stretches its covariance, diffusion spreads its covariance, and the jump semigroup splits it. 3 Problem formulation We consider fully general continuous-time Markov dynamics with infinitely divisible increments; by the Lévy–Khintchine theorem, they take the following canonical representation as a Lévy-Itô SDE: dxt = f(xt) dt + g(xt… view at source ↗
Figure 3
Figure 3. Neural Kolmogorov Equations. NKEs operate as deterministic Neural ODEs on the finite [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Advection-diffusion learning. Euler-Maruyama methods attempt to directly match the [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Jump learning. NKEs cluster jump origins, then model their destinations with a Gaussian [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Results for the Lorenz system. (a) Ground truth. (b) Euler-Maruyama. (c) Kolmogorov [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Results for the Black-Scholes system. (a) Ground truth. (b) Kolmogorov. (c) Euler-Maruyama [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Results for the Double well system. (a) Ground truth. (b) Euler-Maruyama. (c) Kolmogorov [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Results for the rarefied coarse-graining system (positions only). [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 4 linked inside Pith

  1. [1]

    Ricky T. Q. Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary differential equations. InAdvances in Neural Information Processing Systems, volume 31, 2018

    2018

  2. [2]

    Universal differential equations for scientific machine learning.arXiv preprint arXiv:2001.04385, 2020

    Christopher Rackauckas, Yingbo Ma, Julius Martensen, Collin Warner, Kirill Zubov, Rohit Supekar, Dominic Skinner, Ali Ramadhan, and Alan Edelman. Universal differential equations for scientific machine learning.arXiv preprint arXiv:2001.04385, 2020

    arXiv 2001

  3. [3]

    Cohen, Christoph Reisinger, and Sheng Wang

    Samuel N. Cohen, Christoph Reisinger, and Sheng Wang. Arbitrage-free neural-SDE market models.Applied Mathematical Finance, 30(1):1–46, 2023

    2023

  4. [4]

    FP-Diffusion: Improving score-based diffusion models by enforcing the underlying score Fokker–Planck equation

    Chieh-Hsin Lai, Yuhta Takida, Naoki Murata, Toshimitsu Uesaka, Yuki Mitsufuji, and Stefano Ermon. FP-Diffusion: Improving score-based diffusion models by enforcing the underlying score Fokker–Planck equation. InProceedings of the 40th International Conference on Machine Learning, 2023

    2023

  5. [5]

    Neural controlled differential equations for irregular time series

    Patrick Kidger, James Morrill, James Foster, and Terry Lyons. Neural controlled differential equations for irregular time series. InAdvances in Neural Information Processing Systems, volume 33, 2020

    2020

  6. [6]

    Xuechen Li, Ting-Kam Leonard Wong, Ricky T. Q. Chen, and David Duvenaud. Scalable gradients for stochastic differential equations. InProceedings of the 23rd International Confer- ence on Artificial Intelligence and Statistics, volume 108 ofProceedings of Machine Learning Research, 2020

    2020

  7. [7]

    Neural SDEs as infinite-dimensional GANs

    Patrick Kidger, James Foster, Xuechen Li, Harald Oberhauser, and Terry Lyons. Neural SDEs as infinite-dimensional GANs. InProceedings of the 38th International Conference on Machine Learning, volume 139 ofProceedings of Machine Learning Research, 2021

    2021

  8. [8]

    Will Grathwohl, Ricky T. Q. Chen, Jesse Bettencourt, Ilya Sutskever, and David Duvenaud. FFJORD: Free-form continuous dynamics for scalable reversible generative models. InInterna- tional Conference on Learning Representations, 2019

    2019

Show all 49 references
  1. [9]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems, volume 33, pages 6840–6851, 2020

    2020

  2. [10]

    Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole

    Yang Song, Jascha Sohl-Dickstein, Diederik P. Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. In International Conference on Learning Representations, 2021

    2021

  3. [11]

    Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling. InInternational Conference on Learning Representations, 2023

    2023

  4. [12]

    Kloeden and Eckhard Platen.Numerical Solution of Stochastic Differential Equations

    Peter E. Kloeden and Eckhard Platen.Numerical Solution of Stochastic Differential Equations. Springer, 1992

    1992

  5. [13]

    Trajectory flow matching with applications to clinical time series modelling

    Xi Zhang, Yuan Pu, Yuki Kawamura, Andrew Loza, Yoshua Bengio, Dennis L Shung, and Alexander Tong. Trajectory flow matching with applications to clinical time series modelling. Advances in Neural Information Processing Systems, 37:107198–107224, 2024

    2024

  6. [14]

    Naesseth

    Grigory Bartosh, Dmitry Vetrov, and Christian A. Naesseth. SDE matching: Scalable and simulation-free training of latent stochastic differential equations. InProceedings of the 42nd International Conference on Machine Learning, volume 267 ofProceedings of Machine Learning Rese...

    2025

  7. [15]

    Neural SDE: Stabilizing neural ODE networks with stochastic noise.arXiv preprint arXiv:1906.02355, 2019

    Xuanqing Liu, Tesi Xiao, Si Si, Qin Cao, Sanjiv Kumar, and Cho-Jui Hsieh. Neural SDE: Stabilizing neural ODE networks with stochastic noise.arXiv preprint arXiv:1906.02355, 2019

    1906 arXiv

  8. [16]

    Junteng Jia and Austin R. Benson. Neural jump stochastic differential equations. InAdvances in Neural Information Processing Systems, volume 32, 2019. 13

    2019

  9. [17]

    Neural jump-diffusion temporal point processes

    Shuai Zhang, Chuan Zhou, Yang Aron Liu, Peng Zhang, Xixun Lin, and Zhi-Ming Ma. Neural jump-diffusion temporal point processes. InProceedings of the 41st International Conference on Machine Learning, volume 235 ofProceedings of Machine Learning Research, pages 60541–60557, 2024

    2024

  10. [18]

    Yulia Rubanova, Ricky T. Q. Chen, and David Duvenaud. Latent ODEs for irregularly-sampled time series. InAdvances in Neural Information Processing Systems, volume 32, 2019

    2019

  11. [19]

    Convergence and stability analysis of neural stochastic differential equations with poisson jumps.Chaos, Solitons & Fractals, 2026

    Qi Rong, Liang Yan, and Xiaojun Duan. Convergence and stability analysis of neural stochastic differential equations with poisson jumps.Chaos, Solitons & Fractals, 2026

    2026

  12. [20]

    Variational inference with normalizing flows

    Danilo Rezende and Shakir Mohamed. Variational inference with normalizing flows. InPro- ceedings of the 32nd International Conference on Machine Learning, volume 37 ofProceedings of Machine Learning Research, 2015

    2015

  13. [21]

    Karniadakis

    Maziar Raissi, Paris Perdikaris, and George E. Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations.Journal of Computational Physics, 378:686–707, 2019

    2019

  14. [22]

    Williams, Ioannis G

    Matthew O. Williams, Ioannis G. Kevrekidis, and Clarence W. Rowley. A data-driven ap- proximation of the koopman operator: Extending dynamic mode decomposition.Journal of Nonlinear Science, 25(6):1307–1346, 2015

    2015

  15. [23]

    A differential geometric approach to nonlinear filtering: The projection filter.IEEE Transactions on Automatic Control, 43(2):247– 252, 1998

    Damiano Brigo, Bernard Hanzon, and François Le Gland. A differential geometric approach to nonlinear filtering: The projection filter.IEEE Transactions on Automatic Control, 43(2):247– 252, 1998

    1998

  16. [24]

    Approximate nonlinear filtering by projection on exponential manifolds of densities.Bernoulli, 5(3):495–534, 1999

    Damiano Brigo, Bernard Hanzon, and François Le Gland. Approximate nonlinear filtering by projection on exponential manifolds of densities.Bernoulli, 5(3):495–534, 1999

    1999

  17. [25]

    Bernard Hanzon and Remi L. M. Hut. New results on the projection filter. InProceedings of the European Control Conference, pages 623–628, Grenoble, France, 1991

    1991

  18. [26]

    Approximate solution of the Fokker–Planck–Kolmogorov equation.Probabilistic Engineering Mechanics, 17(4):369–384, 2002

    Mario Di Paola and Alba Sofi. Approximate solution of the Fokker–Planck–Kolmogorov equation.Probabilistic Engineering Mechanics, 17(4):369–384, 2002

    2002

  19. [27]

    Gingold and Joseph J

    Robert A. Gingold and Joseph J. Monaghan. Smoothed particle hydrodynamics: Theory and application to non-spherical stars.Monthly Notices of the Royal Astronomical Society, 181(3):375–389, 1977

    1977

  20. [28]

    Leon B. Lucy. A numerical approach to the testing of the fission hypothesis.Astronomical Journal, 82:1013–1024, 1977

    1977

  21. [29]

    H. F. Trotter. On the product of semi-groups of operators.Proceedings of the American Mathematical Society, 10(4):545–551, 1959

    1959

  22. [30]

    On the construction and comparison of difference schemes.SIAM Journal on Numerical Analysis, 5(3):506–517, 1968

    Gilbert Strang. On the construction and comparison of difference schemes.SIAM Journal on Numerical Analysis, 5(3):506–517, 1968

    1968

  23. [31]

    Rousseeuw and Christophe Croux

    Peter J. Rousseeuw and Christophe Croux. Alternatives to the median absolute deviation. Journal of the American Statistical Association, 88(424):1273–1283, 1993

    1993

  24. [32]

    Christopher M. Bishop. Mixture density networks. Technical Report NCRG/94/004, Aston University, 1994

    1994

  25. [33]

    Neural sdes as a unified approach to continuous-domain sequence modeling.arXiv preprint arXiv:2501.18871, 2025

    Macheng Shen and Chen Cheng. Neural sdes as a unified approach to continuous-domain sequence modeling.arXiv preprint arXiv:2501.18871, 2025

    2025 arXiv

  26. [34]

    Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    2017 arXiv

  27. [35]

    American Mathematical Society and Oxford University Press, 2000

    Shun-ichi Amari and Hiroshi Nagaoka.Methods of Information Geometry, volume 191 of Translations of Mathematical Monographs. American Mathematical Society and Oxford University Press, 2000. 14

    2000

  28. [36]

    Singularity structures and impacts on parameter estimation in finite mixtures of distributions.SIAM Journal on Mathematics of Data Science, 1(4):730–758, 2019

    Nhat Ho and XuanLong Nguyen. Singularity structures and impacts on parameter estimation in finite mixtures of distributions.SIAM Journal on Mathematics of Data Science, 1(4):730–758, 2019

    2019

  29. [37]

    Dempster, Nan M

    Arthur P. Dempster, Nan M. Laird, and Donald B. Rubin. Maximum likelihood from incomplete data via the EM algorithm.Journal of the Royal Statistical Society: Series B, 39(1):1–38, 1977

    1977

  30. [38]

    From molecular dynamics to brownian dynamics.Proceedings

    Radek Erban. From molecular dynamics to brownian dynamics.Proceedings. Mathematical, Physical, and Engineering Sciences/The Royal Society, 470(2167):20140036, 2014. 15 A Symbols and notation The paper alternates between the microscopic SDE picture, given in terms of particles ...

    2014

  31. [39]

    two Gaussian components coincide,

  32. [40]

    Consequently,G K becomes a stratified space rather than a smooth manifold [36]

    or a mixture weight vanishes. Consequently,G K becomes a stratified space rather than a smooth manifold [36]. B.2 Projection Operator We will defineΠin terms a divergenceDonM(R d): ΠK(p) := arg min q∈GK D(p∥q).(46) ChoosingD= KLyields the population maximum-likelihood projecti...

  33. [41]

    Permutation symmetry.If λ⋆ is an optimal set of mixture parameters, then any permutation of its Gaussian components yields the same density and is therefore also optimal

  34. [42]

    Proof sketch.Permuting Gaussian components leaves the mixture density unchanged

    Redundant representations.If p∈ GK⋆ for some K ⋆ < K, then infinitely many distinct parameter vectors inG K represent the same density and attain the same projection error. Proof sketch.Permuting Gaussian components leaves the mixture density unchanged. Therefore, if pλ⋆ = KX ...

  35. [43]

    use small mixture counts,

  36. [44]

    regularize covariances,

  37. [45]

    prune low-weight components,

  38. [46]

    18 C Lifted Dynamics This appendix provides theoretical justification for the approximations underlying Neural Kolmogorov Equations

    enforce spatial smoothness. 18 C Lifted Dynamics This appendix provides theoretical justification for the approximations underlying Neural Kolmogorov Equations. We show that:

  39. [47]

    Localized Gaussian packets remain approximately Gaussian under short-time advection– diffusion evolution

  40. [48]

    Sufficiently small jumps behave as an effective diffusion process

  41. [49]

    Throughout this section, we assume f∈C 3 b (Rd,R d), a:=gg ⊤ ∈C 2 b (Rd,R d×d), and that all jump measures possess finite third moments

    State-dependent jump distributions may be frozen at the packet center with controlled error. Throughout this section, we assume f∈C 3 b (Rd,R d), a:=gg ⊤ ∈C 2 b (Rd,R d×d), and that all jump measures possess finite third moments. We denote by L=A+D+J the Kolmogorov generator, ...

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.