Pith. sign in

REVIEW 4 major objections 5 minor 46 references

DKFNet: Differentiable Kalman Filter for Field Inversion and Machine Learning

T0 review · 4 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read A Kalman filter whose transition operator is optimized by adjoint gradients, then approximated by a neural network, recovers missing dynamics and cuts reconstruction error by at least 90% on two test systems.

desk verdict Solid adjoint-derivation work, but the headline 90% improvement is a training-set fit: same observations are optimized and scored, with no held-out evaluation, no baselines, and no numeric error tables. read the letter →

arxiv 2509.07474 v1 pith:7VXRUSAU submitted 2025-09-09 math.OC

Yuan Wu , Sicheng He This is my paper · ORCID
classification math.OC MSC 93E1165K1068T07
keywords differentiableKalmanfilterfieldinversionadjointsensitivitymodel-formerrorclosuremodelAllen-Cahnequationstateestimationuncertaintyquantification
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 claims that the classical Kalman filter's fixed dynamics model can be made adaptive by treating the state-transition operator as an optimization variable. A field-inversion stage minimizes the mismatch between predicted and observed measurements using analytic adjoint gradients, and a second stage trains a neural network to reproduce the optimized operators as a state-dependent closure model. On a rocket trajectory problem and an Allen-Cahn reaction-diffusion boundary value problem, the two-stage DKF lowers reconstruction error by at least 90% across tested noise levels and keeps the filter's covariance-based uncertainty estimates close to the true errors. The contribution matters because it sketches a way to combine data assimilation with machine learning while retaining the Kalman filter's structure and uncertainty propagation.

What carries the argument

The differentiable Kalman filter (DKF): the standard Kalman predict-update loop is reinterpreted as a root-finding system whose residuals r_k stack state and covariance equations, with the transition matrices F_k as learnable parameters. The argument is carried by analytic adjoint Jacobian blocks (Eq. 17)—expressions for A_{k,k-1} and A_{k,k} with respect to state x and covariance P—which provide gradients of the observation-mismatch loss without unrolling the recursion. A neural closure DNN(x,d;theta) trained by Frobenius loss on the optimized operators then serves as the state-dependent transition law.

What would settle it

Use one trajectory: run the DKF field inversion on the first half of the time series and evaluate reconstruction on the held-out second half, at sigma=0.005, 0.025, and 0.125. If the at-least-90% error reduction observed in the paper does not persist on the held-out segment, the optimized operator is fitting the evaluation sequence rather than discovering the dynamics. The same test applies to the Allen-Cahn case with independent noise draws; additionally, compare per-step parameter count (four F_k entries per scalar rocket observation) with attained recovery of the true F_k.

Watch

Extended reading notes

Core claim

The central claim is that model-form error—the gap between the approximate transition operator carried by the filter and the true dynamics—can be removed by solving a least-squares field inversion problem at filter level. The filter's predict-update recursion is treated as a differentiable residual system, the transition matrices F_k are the design variables, and adjoint equations derived from the residual Jacobian blocks give the gradient of the observation mismatch. After the optimized F_k sequence is found, a DNN trained to map state and design variables to those operators serves as the new transition law. In the paper's two examples the optimized filter produces nearly true trajectories

Load-bearing premise

The evaluation assumes that optimizing the transition operator on the very observation sequence used for the reported errors is a valid proxy for reconstruction on unseen data; with more fitted parameters per time step than independent observations, the filter can fit noise without learning real dynamics.

Editorial extensions

If this is right

  • If DKF works as claimed, any Kalman-filter application with an imperfect model can replace its fixed transition operator with a data-corrected one, improving tracking without giving up recursive filtering.
  • The learned closure model, trained on optimized operators, carries the correction into new states, so the benefit is not confined to the exact trajectory used for inversion.
  • The analytic adjoint gradient path avoids storing a full unrolled computation graph, which would make the approach more scalable to long horizons.
  • Since the covariance recursion is retained, the corrected filter still outputs principled uncertainty around its state estimate rather than a point forecast.
  • The two-stage workflow gives an interpretable sequence of discovered transition operators that can be inspected as data-driven corrections of the physics.

Reading between the lines

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

  • The paper does not test out-of-sample generalization; we infer that splitting the observation window into fit and hold-out segments would reveal how much of the reported margin is memorization of the fitted trajectory.
  • The analytic Jacobian blocks of Eq. (17) are generic to Kalman recursion and, we infer, could accelerate training of other learned filtering components such as measurement models or noise covariances.
  • The '90% better than classical KF' comparison uses a deliberately un-optimized baseline; we infer the margin would be smaller, and more informative, against a well-tuned extended or unscented filter.
  • In the rocket case the four free entries of F_k per time step exceed the single scalar observation per step; we infer that identifiability is the limiting factor and that richer observations or regularization would be needed for reliable dynamics discovery.
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 introduces the Differentiable Kalman Filter (DKF), a two-level optimization framework in which the state transition matrix F_k of a Kalman filter is treated as a design variable and optimized via analytically derived adjoint gradients against observation mismatch. After inversion, a neural-network closure model is trained to reproduce the optimized operators. The method is demonstrated on a rocket dynamics model and an Allen–Cahn reaction–diffusion problem, and the authors claim that DKF reduces state reconstruction error by at least 90% relative to the classical Kalman filter while maintaining robust uncertainty quantification. The adjoint Jacobian derivation is presented in Appendix B and verified against automatic differentiation and finite differences in Appendix C.

Significance. If the central numerical claim were supported, the paper would make a useful contribution: an analytic adjoint formulation for differentiating through the Kalman filter is valuable, and the two-level field-inversion-plus-closure workflow is an attractive way to combine data assimilation with machine learning. The Jacobian verification in Appendix C is a concrete strength and gives confidence that the gradient machinery is correct. However, the reported 90% improvement is the paper's headline result, and the current evaluation does not substantiate it as predictive performance because the same observation sequence is used for fitting and for reporting errors. The contribution is therefore promising but not yet demonstrated.

major comments (4)
  1. [Section 4, Eqs. (24), (31), Figs. 1, 4, 5] The evaluation is circular for the paper's central claim. In the rocket case, Eq. (24) minimizes the squared observation error over F_k on the same sequence z_k used to report reconstruction errors in Fig. 1. In the Allen–Cahn case, Eq. (31) minimizes the whitened innovation on the trajectory whose reconstructions are shown in Figs. 4 and 5. With per-time-step free parameters, the optimizer can fit the training observations without learning a generalizable model. The reported at-least-90% error reduction is therefore a training-fit result unless the authors evaluate on held-out observation sequences or on independent trajectories. A train/test split should be added, or the claims should be explicitly reframed as fitting/reconstruction accuracy.
  2. [Section 4.1, Eq. (24), Table 2, Fig. 1] The rocket inversion is underdetermined: F_k has 4 free entries per time step, while each observation is a scalar (H_k = [1 0], Eq. (23)), giving 4 n_t unknowns against n_t measurements. Minimizing Eq. (24) can therefore match the observed altitude without recovering the true dynamics. Table 2 shows the optimized F_k close to the true F, but this is not a consequence of identifiability; it likely reflects the initialization as 'small perturbations around a physically reasonable baseline' (Sec. 4.1.1). The authors should provide an identifiability or regularization analysis, report reconstruction error for the unobserved velocity component, and validate on unseen data before claiming that the dynamics have been discovered.
  3. [Section 4.2.5, Fig. 5] The uncertainty-quantification claim is not quantitatively supported. The DKF posterior variance is produced by a filter whose model was optimized on the same trajectory; matching the supposedly true [P_k]_{ii} on that trajectory is a consistency check, not a calibration test. No coverage probabilities, reliability metrics, or ensemble-spread diagnostics are reported, and each noise level appears to use a single realization without error bars. The phrase 'robust uncertainty quantification' in the abstract requires statistical calibration evidence on independent data.
  4. [Section 4.2.2-4.2.4] The Allen–Cahn experiments use a single trajectory for both inversion and evaluation. The tabulated diffusivity values are optimized against the same observations used to produce the reconstructions in Fig. 4 and the variance comparison in Fig. 5; the DNN is then trained on those same inversion results. Consequently, the DNN reconstructions demonstrate interpolation on the training trajectory, not generalization. Please add a train/test split across initial conditions or noise realizations, and report the corresponding test errors.
minor comments (5)
  1. [Section 1, Table 1] The table marks DEKF with '×' for Learning/Adaptation and End-to-end Gradient, but the text states that the differentiable EKF 'allows the entire state estimation pipeline to be trained via gradient-based optimization.' This inconsistency should be resolved.
  2. [Eq. (31), Section 4.2.2] The notation n_t is used both for the number of time steps and as the upper index for the tabulated diffusivity values d~(v_{n_t}). Since these are different quantities, please use separate symbols (e.g., n_v for the number of table points).
  3. [Figs. 4 and 5] The captions do not specify exactly how the 'DNN predictions' are obtained: is the neural-network surrogate used inside the Kalman filter as the transition operator, or is it used to reconstruct the state directly? Please clarify in the captions and text.
  4. [Section 4.1] The experimental setup is underspecified: the measurement noise covariance R, process noise covariance Q, thrust force, mass, burn time, and the L-BFGS convergence criteria are not reported. These details are needed to reproduce the results and to interpret the noise-level comparison.
  5. [Section 5, abstract] The conclusion states that the Allen–Cahn case achieves an RMSE of d(v) below 1e-2 and an improvement of 'at least two orders of magnitude' compared to the classical Kalman filter, while the abstract reports a '90%' reduction in state reconstruction error. The metrics should be aligned and clearly defined.

Circularity Check

2 steps flagged · score 6.0 of 10

The headline 90% improvement is evaluated on the same observation sequence used to fit the transition operator; the reported 'prediction' reduces to in-sample optimization, not independent predictive skill.

  1. fitted input called prediction [Section 4.1, Eq. (24), Table 2, Fig. 1]
    "Given a sequence of noisy observation {z_k}, the objective is to determine the transition matrices F_k that minimizes the discrepancy between the predicted (filtered) states and the observations over the trajectory. The estimation problem is formulated as the following optimization: min sum || H_k x_k - z_k ||^2 w.r.t. F_1, ..., F_nt. ... The results in Fig. 1 illustrate the filtering performance of the optimized system under various observation noise levels."

    The same observation sequence z_k appears both in the Eq. (24) objective and in the Fig. 1 evaluation. F_k has four free entries per time step while each z_k is a scalar, so the inversion is underdetermined; minimizing Eq. (24) can match the training observations without recovering true dynamics. The reported state-reconstruction advantage of DKF over the unoptimized KF on this same trajectory is therefore an in-sample fitting result, and the abstract's 'reduces state reconstruction error by at least 90%' is presented as predictive skill without a held-out test.

  2. fitted input called prediction [Section 4.2.2, Eq. (31), Sections 4.2.4/4.2.5, Figs. 4-5]
    "These table values {d(v_j)} are optimized to minimize the mean squared innovation between predicted and observed data. Formally, the loss is defined as min 1/nt sum || S_t^{-1/2} (z_t - H x_t) ||^2 w.r.t. d(v_1), ..., d(v_nt) ... The optimized diffusivity table then defines the operator sequence F_k, which provides a data-driven approximation of the true dynamics and is used to reconstruct the state trajectory."

    The diffusivity table is fit by minimizing the whitened innovation on the same trajectory that Figs. 4 and 5 use to report the DKF and DNN reconstruction and variance errors. The neural-network closure is then trained on these field-inverted operators and evaluated on the same trajectory, so its claimed 'strong generalization' is never tested on unseen data. The 90%/order-of-magnitude improvement over the classical filter is thus a measure of fit quality on the training sequence, not an out-of-sample prediction; by construction it cannot support the abstract's predictive claim.

full rationale

The paper is not circular in its derivation of the adjoint sensitivities: the analytic Jacobian blocks in Section 3.1.2 and Appendix B are checked against finite differences and automatic differentiation (Appendix C), and the two-level DKF optimization is a legitimate way to tune F_k or d(v). Nor is there a load-bearing self-citation chain: the cited prior work (FIML [40], adjoint control theory [42]) is used as background, and no uniqueness theorem is imported from the authors' own work. However, the central numerical claim—that the DKF 'consistently reduces state reconstruction error by at least 90%'—is supported only by Figs. 1, 4, and 5, all of which evaluate on the same observation trajectory used to optimize the transition operator/diffusivity in Eqs. (24) and (31). Because the quality metric is computed on the fitting sequence, the margin is a training-set improvement (and, in the rocket case, an underdetermined fit of four F_k entries per scalar observation), not an independent predictive result. This is a fitted-input-called-prediction circularity, warranting a 6. The adjoint verification and operator-recovery plots are informative, but they do not cure the in-sample evaluation of the headline improvement.

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

The method is primarily a fitting procedure: it adds per-time-step matrices F_k or tabulated diffusivities d(v) as optimization variables, then reports errors on the same observations used to fit them. All other structure, covariances, observation operator, and discretization, is assumed known. No new physical particles, forces, or conserved quantities are introduced.

free parameters (3)
  • per-time-step transition matrix F_k (rocket) = Table 2: near [[1,0.1],[0,1]] for sigma=0.005, 0.025, 0.125
    These are the design variables in Eq. (24), optimized against the observation sequence; the reported reconstruction improvements are generated by these fitted matrices.
  • tabulated diffusivity values d~(v_j) (Allen-Cahn) = not tabulated numerically; shown as dots in Fig. 2
    Optimization variables in Eq. (31) representing d(v) on a state grid; the learned d(v) drives the reported accuracy and the neural-network surrogate is trained on it.
  • DNN weights theta for closure model = not reported
    Trained to minimize the Frobenius loss in Eq. (19) against the field-inversion operators; the DNN's state and variance reconstructions are reported as results.
assumptions (6)
  • domain assumption The true and model dynamics are linear in state with Gaussian additive noise, as in Eqs. (1)-(2), and the Kalman filter with fixed covariances is the correct estimator.
    Section 2; required for the residual equations in Eq. (3) and the adjoint derivation to apply.
  • domain assumption The covariance matrices Q_k and R_k are known exactly and are held fixed ('frozen') during field inversion.
    Section 3.1.1 explicitly freezes uncertainty; if Q/R are misspecified, the optimized F_k can absorb the mismatch, breaking the interpretation of the learned dynamics and the uncertainty quantification.
  • domain assumption Model-form error is representable entirely by modifying the transition operator F_k or the diffusivity function d(v), not by changing the observation model or noise structure.
    Eq. (6) and Eq. (31) optimize only F or d; other sources of discrepancy are excluded by construction.
  • standard math The residual system r(x,d)=0 in Eq. (4) has a locally unique solution and the analytic Jacobian blocks in Eq. (17) are correct.
    Appendix B derives the blocks and Appendix C verifies them numerically against finite differences and automatic differentiation; needed for the adjoint gradient in Eq. (8).
  • domain assumption The explicit Euler discretization of the Allen-Cahn equation in Eq. (30) with N=16 and dt=0.01 faithfully represents the continuous problem for the tested regime.
    Section 4.2.2; no stability or convergence study is reported.
  • domain assumption The neural network DNN(x,d;theta) can approximate the fitted operator sequence well enough to generalize to new states.
    Section 3.2 relies on the trained surrogate, but no architecture, training details, or generalization guarantee is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DKFNet: Differentiable Kalman Filter for Field Inversion and Machine Learning." pith.science (2026). https://pith.science/paper/7VXRUSAU

@misc{pith2026250907474,
  author       = {Pith},
  title        = {Pith review of: DKFNet: Differentiable Kalman Filter for Field Inversion and Machine Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7VXRUSAU}},
  note         = {Machine review of arXiv:2509.07474}
}
read the original abstract

The Kalman filter is a fundamental tool for state estimation in dynamical systems. While originally developed for linear Gaussian settings, it has been extended to nonlinear problems through approaches such as the extended and unscented Kalman filters. Despite its broad use, a persistent limitation is that the underlying approximate model is fixed, which can lead to significant deviations from the true system dynamics. To address this limitation, we introduce the differentiable Kalman filter (DKF), an adjoint-based two-level optimization framework designed to reduce the mismatch between approximate and true dynamics. Within this framework, a field inversion step first uncovers the discrepancy, after which a closure model is trained to capture the discovered dynamics, allowing the filter to adapt flexibly and scale efficiently. We illustrate the capabilities of the DKF using two representative examples: a rocket dynamics model and the Allen-Cahn boundary value problem. In both cases, and across a range of noise levels, the DKF consistently reduces state reconstruction error by at least 90% compared to the classical Kalman filter, while also maintaining robust uncertainty quantification. These results demonstrate that the DKF not only improves estimation accuracy by large margins but also enhances interpretability and scalability, offering a principled pathway for combining data assimilation with modern machine learning.

Figures

Figures reproduced from arXiv: 2509.07474 by the authors.

Figure 1
Figure 1. Optimization results for three levels of observation noise: [PITH_FULL_IMAGE:figures/full_fig_p015_1.png] view at source ↗
Figure 2
Figure 2. Comparison of d(v) estimated by different methods under different noise levels σ = 0.0025, 0.005, and 0.01. Dots indicate field inversion results, solid lines indicate DNN fits, the black line shows the ground truth d(v) = 0.1 tanh(v), and the red line denotes the constant initial guess d(v) ≡ 1.0. via DKF optimization successfully reconstructs the nonlinear diffusivity d(v) across all noise levels. Even under relat… view at source ↗
Figure 3
Figure 3. Time evolution of the relative Frobenius norm error [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Spatiotemporal reconstruction under different observation noise levels ( [PITH_FULL_IMAGE:figures/full_fig_p023_4.png]
Figure 5
Figure 5. Figure 5: Evolution of the diagonal elements of the covariance matrix [PITH_FULL_IMAGE:figures/full_fig_p024_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 40 canonical work pages

  1. [1]

    Rezaei, R

    S. Rezaei, R. Sengupta, Kalman filter based integration of DGPS and ve- hicle sensors for localization, in: IEEE International Conference Mecha- tronics and Automation, 2005, ICMA-05, IEEE, 2005, p. 455–460. doi:10.1109/icma.2005.1626590. 26

    work page arXiv 2005

  2. [2]

    J. Liu, G. Guo, Vehicle localization during GPS outages with extended Kalman filter and deep learning, IEEE Transactions on Instrumentation and Measurement 70 (2021) 1–10.doi:10.1109/tim.2021.3097401

    work page arXiv 2021

  3. [3]

    Y. Lu, H. Ma, E. Smart, H. Yu, Real-time performance-focused local- ization techniques for autonomous vehicle: A review, IEEE Transac- tions on Intelligent Transportation Systems 23 (7) (2022) 6082–6100. doi:10.1109/tits.2021.3077800

    work page arXiv 2022

  4. [4]

    Chalvatzaras, I

    A. Chalvatzaras, I. Pratikakis, A. A. Amanatiadis, A survey on map- based localization techniques for autonomous vehicles, IEEE Transac- tions on Intelligent Vehicles 8 (2) (2023) 1574–1596.doi:10.1109/tiv. 2022.3192102

    work page arXiv 2023

  5. [5]

    B. O. Teixeira, M. A. Santillo, R. S. Erwin, D. S. Bernstein, Space- craft tracking using sampled-data Kalman filters, IEEE Control Systems Magazine 28 (4) (2008) 78–94.doi:10.1109/mcs.2008.923231

    work page arXiv 2008

  6. [6]

    Zahaby, P

    M. Zahaby, P. Gaonjur, S. Farajian, Location tracking in GPS using Kalman filter through SMS, in: IEEE EUROCON 2009, IEEE, 2009, p. 1707–1711.doi:10.1109/eurcon.2009.5167873

    work page arXiv 2009

  7. [7]

    Y. Yang, X. Yue, A. G. Dempster, GPS-based onboard real-time orbit determination for leo satellites using consider Kalman filter, IEEE Transactions on Aerospace and Electronic Systems 52 (2) (2016) 769–777.doi:10.1109/taes.2015.140758

    work page arXiv 2016

  8. [8]

    W. Pei, X. Lu, Moving object tracking in satellite videos by kernel- ized correlation filter based on color-name features and Kalman pre- diction, Wireless Communications and Mobile Computing 2022 (2022) 1–16.doi:10.1155/2022/9735887

    work page doi:10.1155/2022/9735887 2022

Show all 46 references
  1. [9]

    Huang, N.-Y

    S.-C. Huang, N.-Y. Wang, T.-Y. Li, Y.-C. Lee, L.-F. Chang, T.-H. Pan, Financial forecasting by modified Kalman filters and kernel ma- chines, Journal of Statistics and Management Systems 16 (2–03) (2013) 163–176.doi:10.1080/09720510.2013.777575

    2013

  2. [10]

    X. Bao, Q. Tao, H. Fu, Dynamic financial distress prediction based on Kalman filtering, Journal of Applied Statistics 42 (2) (2014) 292–308. doi:10.1080/02664763.2014.947359. 27

    2014

  3. [11]

    Khashei, B

    M. Khashei, B. Mahdavi Sharif, A Kalman filter-based hybridization model of statistical and intelligent approaches for exchange rate fore- casting, Journal of Modelling in Management 16 (2) (2020) 579–601. doi:10.1108/jm2-12-2019-0277

    2020 doi

  4. [12]

    G. Rodriguez, Kalman filtering, smoothing, and recursive robot arm for- ward and inverse dynamics, IEEE Journal on Robotics and Automation 3 (6) (1987) 624–639.doi:10.1109/jra.1987.1087147

    1987

  5. [13]

    Gautier, P

    M. Gautier, P. Poignet, Extended Kalman filtering and weighted least squares dynamic identification of robot, Control Engineering Practice 9 (12) (2001) 1361–1372.doi:10.1016/s0967-0661(01)00105-8

    2001 doi

  6. [14]

    G. Du, P. Zhang, A markerless human–robot interface using particle filter and Kalman filter for dual robots, IEEE Transactions on Industrial Electronics 62 (4) (2015) 2257–2264.doi:10.1109/tie.2014.2362095

    2015

  7. [15]

    Martin, D. I. H. Putri, Riyanto, C. Machbub, Gait controllers on hu- manoid robot using Kalman filter and PD controller, in: 2018 15th International Conference on Control, Automation, Robotics and Vi- sion (ICARCV), IEEE, 2018, p. 36–41.doi:10.1109/icarcv.2018. 8581061

    2018 doi

  8. [16]

    K.Lee, K.T.Carlberg, Modelreductionofdynamicalsystemsonnonlin- earmanifolds using deepconvolutionalautoencoders, Journalof Compu- tational Physics 404 (2020) 108973.doi:10.1016/j.jcp.2019.108973

    2020

  9. [17]

    Humpherys, P

    J. Humpherys, P. Redd, J. West, A fresh look at the Kalman filter, SIAM Review 54 (4) (2012) 801–823.doi:10.1137/100799666

    2012 doi

  10. [18]

    J. Kao, D. Flicker, R. Henninger, S. Frey, M. Ghil, K. Ide, Data assim- ilation with an extended Kalman filter for impact-produced shock-wave dynamics, Journal of Computational Physics 196 (2) (2004) 705–723. doi:10.1016/j.jcp.2003.11.028

    2004 doi

  11. [19]

    Branicki, B

    M. Branicki, B. Gershgorin, A. Majda, Filtering skill for turbulent signals for a suite of nonlinear and linear extended Kalman filters, Journal of Computational Physics 231 (4) (2012) 1462–1498.doi: 10.1016/j.jcp.2011.10.029. 28

    2012 doi

  12. [20]

    György, A

    K. György, A. Kelemen, L. Dávid, Unscented Kalman filters and particle filter methods for nonlinear state estimation, Procedia Technology 12 (2014) 65–74.doi:10.1016/j.protcy.2013.12.457

    2014 doi

  13. [21]

    W. Kang, S. King, L. Xu, A Sparse-grid UKF For The State Estima- tion of PDEs, Society for Industrial and Applied Mathematics, 2017, p. 101–106.doi:10.1137/1.9781611975024.14

    2017 doi

  14. [22]

    U. Z. Ijaz, A. K. Khambampati, J. S. Lee, S. Kim, K. Y. Kim, Non- stationary phase boundary estimation in electrical impedance tomogra- phy using unscented Kalman filter, Journal of Computational Physics 227 (15) (2008) 7089–7112.doi:10.1016/j.jcp.2007.12.025

    2008 doi

  15. [23]

    Evensen, The ensemble Kalman filter: theoretical formulation and practical implementation, Ocean Dynamics 53 (4) (2003) 343–367.doi: 10.1007/s10236-003-0036-9

    G. Evensen, The ensemble Kalman filter: theoretical formulation and practical implementation, Ocean Dynamics 53 (4) (2003) 343–367.doi: 10.1007/s10236-003-0036-9

    2003 doi

  16. [24]

    Grooms, Y

    I. Grooms, Y. Lee, A. J. Majda, Ensemble Kalman filters for dynamical systems with unresolved turbulence, Journal of Computational Physics 273 (2014) 435–452.doi:10.1016/j.jcp.2014.05.037

    2014 doi

  17. [25]

    Harlim, A

    J. Harlim, A. Mahdi, A. J. Majda, An ensemble Kalman filter for sta- tistical estimation of physics constrained nonlinear regression models, Journal of Computational Physics 257 (2014) 782–812.doi:10.1016/ j.jcp.2013.10.025

    2014

  18. [26]

    W. Xie, Z. Wang, J. Kim, X. Sun, Y. Li, A novel ensemble Kalman filter based data assimilation method with an adaptive strategy for dendritic crystal growth, Journal of Computational Physics 524 (2025) 113711. doi:10.1016/j.jcp.2024.113711

    2025

  19. [27]

    Sebacher, R

    B. Sebacher, R. Hanea, A. Heemink, A probabilistic parametrization for geological uncertainty estimation using the ensemble Kalman filter (EnKF), Computational Geosciences 17 (5) (2013) 813–832.doi:10. 1007/s10596-013-9357-z

    2013

  20. [28]

    Del Moral, A

    P. Del Moral, A. Doucet, S. S. Singh, Uniform stability of a particle approximation of the optimal filter derivative, SIAM Journal on Control and Optimization 53 (3) (2015) 1278–1304.doi:10.1137/140993703. 29

    2015 doi

  21. [29]

    Kloss, G

    A. Kloss, G. Martius, J. Bohg, How to train your differentiable filter, Autonomous Robots 45 (4) (2021) 561–578.doi:10.1007/ s10514-021-09990-9

    2021

  22. [31]

    S. Shen, J. Chen, G. Yu, Z. Zhai, P. Han, KalmanFormer: using trans- former to model the Kalman gain in Kalman filters, Frontiers in Neuro- robotics 18 (Jan. 2025).doi:10.3389/fnbot.2024.1460255

    2025

  23. [32]

    N. A. Piga, U. Pattacini, L. Natale, A differentiable extended Kalman filter for object tracking under sliding regime, Frontiers in Robotics and AI 8 (Aug. 2021).doi:10.3389/frobt.2021.686447

    2021

  24. [33]

    Revach, N

    G. Revach, N. Shlezinger, X. Ni, A. L. Escoriza, R. J. G. van Sloun, Y. C. Eldar, Kalmannet: Neural network aided Kalman filtering for partially known dynamics, IEEE Transactions on Signal Processing 70 (2022) 1532–1547.doi:10.1109/tsp.2022.3158588

    2022

  25. [34]

    X. Liu, S. Ikemoto, Y. Yoshimitsu, H. B. Amor, Learning soft robot dynamics using differentiable Kalman filters and spatio-temporal em- beddings, in: 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, 2023, p. 2550–2557.doi:10.1109/ iros5555...

    2023

  26. [35]

    Y. Chen, D. Sanz-Alonso, R. Willett, Autodifferentiable ensemble Kalman filters, SIAM Journal on Mathematics of Data Science 4 (2) (2022) 801–833.doi:10.1137/21m1434477

    2022 doi

  27. [36]

    Corenflos, J

    A. Corenflos, J. Thornton, G. Deligiannidis, A. Doucet, Differentiable particle filtering via entropy-regularized optimal transport (2021).doi: 10.48550/ARXIV.2102.07850

    2021 doi arXiv

  28. [37]

    B. Cox, S. Pérez-Vieites, N. Zilberstein, M. Sevilla, S. Segarra, V. Elvira, End-to-end learning of Gaussian mixture proposals using differentiable particle filters and neural networks, in: ICASSP 2024 - 2024 IEEE 30 International Conference on Acoustics, Speech and Signal Pro...

    2024

  29. [38]

    Y. Yin, C. Kou, S. Jia, L. Lu, X. Yuan, Y. Luo, PCDMD: Physics- constrained dynamic mode decomposition for accurate and robust fore- casting of dynamical systems with imperfect data and physics, Com- puter Physics Communications 304 (2024) 109303.doi:10.1016/j. cpc.2024.109303

    2024

  30. [39]

    Jiang, N

    L. Jiang, N. Liu, Correcting noisy dynamic mode decomposition with Kalman filters, Journal of Computational Physics 461 (2022) 111175. doi:10.1016/j.jcp.2022.111175

    2022

  31. [40]

    E. J. Parish, K. Duraisamy, A paradigm for data-driven predictive mod- eling using field inversion and machine learning, Journal of Computa- tional Physics 305 (2016) 758–774.doi:10.1016/j.jcp.2015.11.012

    2016 doi

  32. [41]

    R. E. Kalman, A new approach to linear filtering and prediction prob- lems, Journal of Basic Engineering 82 (1) (1960) 35–45.doi:10.1115/ 1.3662552

    1960

  33. [42]

    Jameson, Aerodynamic design via control theory, Journal of Scientific Computing 3 (3) (1988) 233–260.doi:10.1007/bf01061285

    A. Jameson, Aerodynamic design via control theory, Journal of Scientific Computing 3 (3) (1988) 233–260.doi:10.1007/bf01061285

    1988 doi

  34. [43]

    Bradbury, R

    J. Bradbury, R. Frostig, P. Hawkins, M. J. Johnson, C. Leary, D. Maclaurin, G. Necula, A. Paszke, J. VanderPlas, S. Wanderman- Milne, Q. Zhang, JAX: composable transformations of Python+NumPy programs (2018)

    2018

  35. [44]

    Zhang, Q

    J. Zhang, Q. Du, Numerical studies of discrete approximations to the Allen–Cahn equation in the sharp interface limit, SIAM Journal on Sci- entific Computing 31 (4) (2009) 3042–3063.doi:10.1137/080738398

    2009 doi

  36. [45]

    X. Qi, Y. Zhang, C. Xu, An efficient approximation to the stochastic Allen-Cahn equation with random diffusion coefficient field and multi- plicative noise, Advances in Computational Mathematics 49 (5) (Sep. 2023).doi:10.1007/s10444-023-10072-w

    2023 doi

  37. [46]

    Mohammadi, D

    V. Mohammadi, D. Mirzaei, M. Dehghan, Numerical simulation and error estimation of the time-dependent Allen–Cahn equation on surfaces 31 withradialbasisfunctions, JournalofScientificComputing79(1)(2018) 493–516.doi:10.1007/s10915-018-0859-7

    2018 doi

  38. [47]

    TH” represents the formula results that we have derived, and “AD

    P. Benner, M. Stoll, Optimal control for Allen-Cahn equations enhanced by model predictive control, IFAC Proceedings Volumes 46 (26) (2013) 139–143.doi:10.3182/20130925-3-fr-4043.00062. Appendix A. Kalman filter derivation In this section, we present the detailed derivation of...

    2013

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.