pith. sign in

arxiv: 2605.26881 · v1 · pith:2POEDMZBnew · submitted 2026-05-26 · 🧮 math.ST · math.DS· stat.ME· stat.TH

Robust ensemble Kalman filtering under observation noise misspecification via diffusion score matching

Hans Reimann , Sebastian Reich This is my paper

Pith reviewed 2026-07-01 16:12 UTC · model grok-4.3

classification 🧮 math.ST math.DSstat.MEstat.TH
keywords robust Kalman filteringdiffusion score matchingobservation noise misspecificationensemble Kalman filtergeneralized Bayesian inferencedata assimilationLorenz system
0
0 comments X

The pith

Diffusion score matching in the analysis step yields a robust Kalman filter that preserves uncertainty quantification under noise misspecification.

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

The paper introduces a Kalman filter that replaces the standard Bayesian update in the analysis step with diffusion score matching to handle mismatches between assumed and actual observation noise, especially outliers from differing tail decay. This matters because conventional fixes such as detect-and-delete schemes or covariance inflation can destabilize the filter or degrade uncertainty estimates when forecast uncertainty is already high. Theoretical results establish conjugacy, closed-form parameter updates, robustness, covariance stability and high-dimensional consistency for the linear Gaussian case. Ensemble approximations via stochastic and deterministic coupling, together with localization, produce practical EnKF, ESRF and LETKF variants. Simulations on target tracking, Lorenz 63 and 40-dimensional Lorenz 96 systems illustrate the method's performance under nonlinear dynamics and non-Gaussian noise.

Core claim

The diffusion score matching Kalman filter adjusts information processing in the analysis step by employing diffusion score matching for inference to obtain robustness while maintaining well-quantified uncertainties. In linear Gaussian state space systems it exhibits conjugacy and closed-form parameter updates, robustness, covariance stability, and high-dimensional consistency. Ensemble approximations are derived via stochastic and deterministic coupling as well as implementing localization to obtain EnKF, ESRF and LETKF varieties. The method is evaluated in simulation studies on target-tracking, the chaotic Lorenz 63 system and the Lorenz 96 system in 40 dimensions.

What carries the argument

Diffusion score matching applied to inference in the analysis step of the Kalman filter, enabling generalized Bayesian updates that are robust to observation noise misspecification.

If this is right

  • The filter remains conjugate with closed-form updates in linear Gaussian state space systems.
  • Covariance stability and high-dimensional consistency hold under the stated conditions.
  • Stochastic and deterministic ensemble couplings with localization yield practical EnKF, ESRF and LETKF implementations.
  • The approach improves assimilation when dynamics are nonlinear and observation noise may be non-Gaussian.

Where Pith is reading between the lines

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

  • The identified robustness-stability trade-off could guide design of other generalized Bayesian filters beyond the Kalman setting.
  • The method may be combined with existing localization or inflation techniques to handle even higher-dimensional or more severely misspecified cases.
  • Testing on systems with explicitly heavier-tailed observation noise would directly probe the claimed robustness gains.

Load-bearing premise

That diffusion score matching can be used in the analysis step to navigate the robustness-stability trade-off without destabilizing the filter in challenging non-linear dynamics with non-Gaussian noise.

What would settle it

In the 40-dimensional Lorenz 96 simulation, an instance where the diffusion score matching Kalman filter produces ensemble covariances that fail to cover the true state trajectory or where the filter diverges when observation outliers are present.

Figures

Figures reproduced from arXiv: 2605.26881 by Hans Reimann, Sebastian Reich.

Figure 1
Figure 1. Figure 1: provides a visual intuition of the different choices of weight kernel via a comparison of the shapes of the different loss functions resulting from the regular, KL discrepancy based posterior (the log-likelihood, left), the weighted log-likelihood utilized in the WoLF KF in [4] (middle) and the the DSM based loss (right) for assuming a standard Gaussian likelihood. We observe the described behaviour of the… view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of the curves the Shannon information criterion (blue) and [PITH_FULL_IMAGE:figures/full_fig_p021_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Centred trajectories for the different methods and their [PITH_FULL_IMAGE:figures/full_fig_p023_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Centred trajectories for the different methods and their [PITH_FULL_IMAGE:figures/full_fig_p024_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Example position trajectories for the different methods and contaminated observations. [PITH_FULL_IMAGE:figures/full_fig_p025_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Averaged RMSE over MMC = 2500 repetitions of the Kalman filter varieties for different frequencies and degrees. time windows with Tend = 50 with discretization step size ∆t = 0.001 and observations generated every tout = 0.05, so every 50 steps. The fairly long time windows and resolution is chosen to investigate stability. We adjust the model as given in [18]. The state is updated according to xn = xn−1 +… view at source ↗
Figure 7
Figure 7. Figure 7: Averaged IC over MMC = 2500 repetitions of the Kalman filter varieties for different frequencies and degrees. of Mens = 10. For the DSM and WoLF EnKF variants we use the average particle variants. The initial ensemble is produced by sampling x a,(i) 0 ∼iid N (x0, 0.1 · 13×3) for i = 1, 2, . . . , Mens. An example trajectory with contaminated observations for frequency ϵ = 0.25 and degree λ = 252 is present… view at source ↗
Figure 8
Figure 8. Figure 8: Example trajectories for the different stochastic EnKF variants and contaminated observations. [PITH_FULL_IMAGE:figures/full_fig_p026_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Centred trajectories of the unobserved x2-component for the different methods and their Gaussian approxima￾tion 95%-CIs in the contaminated model. The experiment was repeated MMC = 1000 times for different combinations of frequency and degree respectively via ϵ ∈ {0, 0.025, 0.05, ..., 0.25} and √ λ ∈ {2.5, 5, 7.5, ..., 27.5} including the well-specified case for ϵ = 0. The results are presented for the RMS… view at source ↗
Figure 10
Figure 10. Figure 10: Averaged RMSE over MMC = 1000 repetitions of the stochastic variants of the regular EnKF (left), DSM EnKF (middle) and WoLF EnKF (right) for different frequencies and degrees. 0 10 20 30 0.0 0.1 0.2 Eps Lam 0 10 20 30 0.0 0.1 0.2 Eps Lam 0 10 20 30 0.0 0.1 0.2 Eps Lam IC 4 6 8 [PITH_FULL_IMAGE:figures/full_fig_p028_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Averaged IC over MMC = 1000 repetitions of the stochastic variants of the regular EnKF (left), DSM EnKF (middle) and WoLF EnKF (right) for different frequencies and degrees. 1 3 5 10 30 100 300 Ensemble Size RMSE 1 3 10 10 30 100 300 Ensemble Size Reg. EnKF DSM EnKF WoLF EnKF [PITH_FULL_IMAGE:figures/full_fig_p028_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Averaged RMSE over MMC = 1000 repetitions of stochastic EnKF variants for different ensemble sizes in the well-specified (left) and contaminated model (right). The the dotted line indicates the Monte Carlo rate. This supports a heuristic redistributing information rather than just discarding information as with the DSM EnKF and help explain the results of the numerical experiments. 28 [PITH_FULL_IMAGE:fi… view at source ↗
Figure 13
Figure 13. Figure 13: Averaged IC over MMC = 1000 repetitions of stochastic EnKF variants for different ensemble sizes in the well-specified (left) and contaminated model (right). Additional results on the Lorenz-63 system with well-specified observation are provided in apx. I.2 via an example trajectory in fig. 23 and tab. 5 and the corresponding estimated Gaussian approximation 95%-CIs for the unobserved x2-component fig. 9.… view at source ↗
Figure 14
Figure 14. Figure 14: Example trajectories of the x1-component for the different LETKF variants and contaminated observations. reg. KF DSM KF WoLF KF RMSE 9.309 0.328 0.359 q-IC 8.523 0.507 0.494 [PITH_FULL_IMAGE:figures/full_fig_p030_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Centred trajectories of the x1-component for the different LETKF variants and their Gaussian approximation 95%-CIs in the contaminated model. 5 10 15 20 0.00 0.05 0.10 0.15 0.20 Eps Lam 5 10 15 20 0.00 0.05 0.10 0.15 0.20 Eps Lam 5 10 15 20 0.00 0.05 0.10 0.15 0.20 Eps Lam RMSE 1 2 3 4 5 [PITH_FULL_IMAGE:figures/full_fig_p031_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Averaged RMSE over MMC = 100 repetitions of the regular LETKF (left), DSM LETKF (middle) and WoLF LETKF (right) for different frequencies and degrees. other words, for Mens ≥ 16 and given the setup, the DSM LETKF can improve uncertainty quantification by adjusting implicit mis-specification in forecast-observation mis-match resulting from the still fairly small ensemble size yet with the forecast ensemble… view at source ↗
Figure 17
Figure 17. Figure 17: Averaged IC over MMC = 100 repetitions of the regular LETKF (left), DSM LETKF (middle) and WoLF LETKF (right) for different frequencies and degrees. 0.26 0.27 0.28 0.29 10.0 12.5 15.0 17.5 20.0 Ensemble Size RMSE 0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0 Ensemble Size 0.30 0.35 0.40 0.45 10 15 20 Reg. LETKF DSM LETKF WoLF LETKF [PITH_FULL_IMAGE:figures/full_fig_p032_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Averaged RMSE over MMC = 100 repetitions of LETKF variants for different ensemble sizes in the well-specified (left) and contaminated model (right). Uncertainty quantification evaluated via the q-IC deteriorates for increasing frequency of contamination, yet strongly improving for the DSM LETKF with increasing ensemble size. As will be point of discussion in sec. 6 and observed also for the Lorenz-63 simu… view at source ↗
Figure 19
Figure 19. Figure 19: Averaged IC over MMC = 100 repetitions of LETKF variants for different ensemble sizes in the well￾specified (left) and contaminated model (right). basic concepts in robust statistics are based on this model, including the works on robust Bayesian inference providing foundation for the work at hand. What it describes is essentially, that the assumed model, here the observation marginal resulting from the m… view at source ↗
Figure 20
Figure 20. Figure 20: exp imq I.2 Complementing simulation results −1 0 1 2 0 25 50 75 100 Time Value Reg. KF DSM KF WoLF KF [PITH_FULL_IMAGE:figures/full_fig_p053_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Centred trajectories for the different methods in the well-specified model. [PITH_FULL_IMAGE:figures/full_fig_p053_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Centred trajectories for the different methods in the contaminated model. [PITH_FULL_IMAGE:figures/full_fig_p053_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Example trajectories for the different stochastic EnKF variants in the well-specified model. [PITH_FULL_IMAGE:figures/full_fig_p054_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: Centred trajectories of the unobserved x2-component for the different methods and their Gaussian approxi￾mation 95%-CIs in the well-specified model. −5 0 5 10 0.0 2.5 5.0 7.5 10.0 12.5 Time Value True Obs. Reg. LETKF DSM LETKF WoLF LETKF [PITH_FULL_IMAGE:figures/full_fig_p055_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Example trajectories of the x1-component for the different LETKF variants in the well-specified model. 55 [PITH_FULL_IMAGE:figures/full_fig_p055_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: Centred trajectories of the x1-component for the different LETKF variants and their Gaussian approximation 95%-CIs in the well-specified model. 56 [PITH_FULL_IMAGE:figures/full_fig_p056_26.png] view at source ↗
read the original abstract

We address the problem of observation noise misspecification in Bayesian filtering of dynamical systems via recent advances in generalised Bayesian inference. Mis-match in tail decay between the true data generating process and an assumed observation model, often showing via frequent outliers, can strongly impact Bayesian updates and analysis in Kalman filtering. Existing approaches often employ detect-and-delete-schemes or covariance inflation to avoid assimilation of influential instances of mis-specification. In challenging settings where the analysis updates are barely sufficient to counteract the induced forecast uncertainty, these strategies may destabilize or struggle to provide reliable uncertainty quantification. We consider a novel Kalman filter adjusting information processing in the analysis step by employing diffusion score matching for inference to obtain robustness while maintaining well-quantified uncertainties. We provide theoretical properties of the diffusion score matching Kalman filter in linear Gaussian state space systems covering conjugacy and closed form parameter update in the analysis step, robustness, covariance stability, and tuning as well as high-dimensional consistency. We derive ensemble approximations via stochastic and deterministic coupling as well as implementing localization to obtain EnKF, ESRF and LETKF varieties. We evaluate the methods in appropriate simulation studies on target-tracking, the chaotic Lorenz 63 system and the Lorenz 96 system in 40 dimensions. Our insights highlight a critical trade-off between robustness and stability in Bayesian filtering. Methods employing generalized Bayesian inference can navigate this balance and improve data assimilation in challenging environments combining non-linear dynamics and potentially non-Gaussian observation noise.

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 paper introduces a diffusion score matching Kalman filter to achieve robustness against observation noise misspecification in ensemble Kalman filtering. It establishes theoretical properties (conjugacy, closed-form updates, robustness, covariance stability, and high-dimensional consistency) for linear Gaussian state-space systems, derives ensemble approximations (EnKF, ESRF, LETKF) with localization, and evaluates performance via simulations on target tracking and the chaotic Lorenz-63/96 systems.

Significance. If the central claims hold, the work provides a new mechanism for balancing robustness and uncertainty quantification in data assimilation under misspecification, extending generalized Bayesian ideas to ensemble filters with explicit linear-Gaussian theory and practical ensemble implementations. This could be relevant for applications involving outliers or non-Gaussian noise in high-dimensional dynamical systems.

major comments (2)
  1. [Abstract, §3] Abstract and §3 (theoretical properties): All stated guarantees (conjugacy, closed-form parameter updates, robustness, covariance stability, high-dimensional consistency) are derived exclusively for linear Gaussian state-space systems. The abstract and introduction claim the method 'navigates this balance' in 'challenging environments combining non-linear dynamics and potentially non-Gaussian observation noise,' but no extension of the proofs or stability analysis to the non-linear regime is indicated; the non-linear claims therefore rest entirely on the ensemble simulation results in §5.
  2. [§5] §5 (simulation studies): The Lorenz-63 and Lorenz-96 experiments (non-linear dynamics) report improved robustness relative to standard EnKF/ESRF/LETKF, but the manuscript provides no diagnostic (e.g., posterior coverage, filter divergence rates, or bias-variance decomposition) that would confirm the advertised 'well-quantified uncertainties' are preserved when the linear-Gaussian conjugacy no longer holds. This is load-bearing for the central trade-off claim.
minor comments (2)
  1. [§2, §3] Notation for the diffusion score-matching objective and its relation to the analysis-step likelihood should be introduced earlier and used consistently across the linear theory and ensemble sections.
  2. [§3] The high-dimensional consistency result would benefit from an explicit statement of the scaling regime (e.g., dimension vs. ensemble size) under which the result holds.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments on our manuscript. We address each major comment below and outline the corresponding revisions.

read point-by-point responses

  1. Referee: [Abstract, §3] Abstract and §3 (theoretical properties): All stated guarantees (conjugacy, closed-form parameter updates, robustness, covariance stability, high-dimensional consistency) are derived exclusively for linear Gaussian state-space systems. The abstract and introduction claim the method 'navigates this balance' in 'challenging environments combining non-linear dynamics and potentially non-Gaussian observation noise,' but no extension of the proofs or stability analysis to the non-linear regime is indicated; the non-linear claims therefore rest entirely on the ensemble simulation results in §5.

    Authors: We agree that the theoretical results (conjugacy, closed-form updates, robustness, covariance stability, and high-dimensional consistency) are derived exclusively under the linear-Gaussian assumption in §3. The abstract and introduction statements regarding non-linear dynamics and non-Gaussian noise are supported only by the empirical results in §5. We will revise the abstract and the relevant paragraphs in the introduction to explicitly separate the linear-Gaussian theory from the simulation-based evidence for non-linear systems. revision: yes

  2. Referee: [§5] §5 (simulation studies): The Lorenz-63 and Lorenz-96 experiments (non-linear dynamics) report improved robustness relative to standard EnKF/ESRF/LETKF, but the manuscript provides no diagnostic (e.g., posterior coverage, filter divergence rates, or bias-variance decomposition) that would confirm the advertised 'well-quantified uncertainties' are preserved when the linear-Gaussian conjugacy no longer holds. This is load-bearing for the central trade-off claim.

    Authors: The referee is correct that §5 does not include explicit diagnostics such as posterior coverage, divergence rates, or bias-variance decompositions for the non-linear cases. The reported improvements are based on RMSE and stability metrics. We will add posterior coverage checks (via rank histograms) and divergence-rate statistics to the revised §5 to provide direct support for the uncertainty-quantification claims under non-linear dynamics. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The paper introduces diffusion score matching into the Kalman analysis step as a novel adjustment for robustness under noise misspecification. Theoretical properties (conjugacy, closed-form updates, robustness, covariance stability, high-dimensional consistency) are stated to be derived specifically for linear-Gaussian state-space systems, while non-linear cases (Lorenz-63/96) are evaluated via ensemble simulations. No equations or steps are shown that reduce a claimed result to a fitted parameter or self-citation by construction; the central claims rest on explicit derivations and external simulation benchmarks rather than tautological re-labeling of inputs. This is the common honest case of a self-contained methodological contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Insufficient information from abstract to identify specific free parameters, axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5793 in / 880 out tokens · 30322 ms · 2026-07-01T16:12:03.801605+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

59 extracted references · 59 canonical work pages · 1 internal anchor

  1. [1]

    A Kalman filter in the presence of outliers.Communi- cations in Statistics-Theory and Methods, 20(5-6):1803–1820, 1991

    Nihal Yatawara, Bovas Abraham, and John F MacGregor. A Kalman filter in the presence of outliers.Communi- cations in Statistics-Theory and Methods, 20(5-6):1803–1820, 1991

    work page 1991

  2. [2]

    Robust Kalman filtering for uncertain systems.Systems & Control Letters, 22(2):123–129, 1994

    Lihua Xie and Yeng Chai Soh. Robust Kalman filtering for uncertain systems.Systems & Control Letters, 22(2):123–129, 1994

    work page 1994

  3. [3]

    Generalised Bayesian filtering via sequential Monte Carlo.Advances in Neural Information Processing Systems, 33:418–429, 2020

    Ayman Boustati, Omer Deniz Akyildiz, Theodoros Damoulas, and Adam Johansen. Generalised Bayesian filtering via sequential Monte Carlo.Advances in Neural Information Processing Systems, 33:418–429, 2020

    work page 2020

  4. [4]

    Outlier-robust Kalman filtering through generalised Bayes.arXiv preprint arXiv:2405.05646, 2024

    Gerardo Duran-Martin, Matias Altamirano, Alexander Y Shestopaloff, Leandro Sánchez-Betancourt, Jeremias Knoblauch, Matt Jones, François-Xavier Briol, and Kevin Murphy. Outlier-robust Kalman filtering through generalised Bayes.arXiv preprint arXiv:2405.05646, 2024

    work page arXiv 2024

  5. [5]

    Nudging state-space models for Bayesian filtering under misspecified dynamics.Statistics and Computing, 35(4):112, 2025

    Fabián González, O Deniz Akyildiz, Dan Crisan, and Joaquín Míguez. Nudging state-space models for Bayesian filtering under misspecified dynamics.Statistics and Computing, 35(4):112, 2025

    work page 2025

  6. [6]

    Generalized variational infer- ence: Three arguments for deriving new posteriors.arXiv preprint arXiv:1904.02063, 2019

    Jeremias Knoblauch, Jack Jewson, and Theodoros Damoulas. Generalized variational inference: Three arguments for deriving new posteriors.arXiv preprint arXiv:1904.02063, 2019

    work page arXiv 1904

  7. [7]

    An optimization-centric view on Bayes’ rule: Reviewing and generalizing variational inference.Journal of Machine Learning Research, 23(132):1–109, 2022

    Jeremias Knoblauch, Jack Jewson, and Theodoros Damoulas. An optimization-centric view on Bayes’ rule: Reviewing and generalizing variational inference.Journal of Machine Learning Research, 23(132):1–109, 2022

    work page 2022

  8. [8]

    Some PAC-Bayesian theorems

    David A McAllester. Some PAC-Bayesian theorems. InProceedings of the eleventh annual Conference on Computational Learning Theory, pages 230–234, 1998

    work page 1998

  9. [9]

    Robust Bayes estimation using the density power divergence.Annals of the Institute of Statistical Mathematics, 68:413–437, 2016

    Abhik Ghosh and Ayanendranath Basu. Robust Bayes estimation using the density power divergence.Annals of the Institute of Statistical Mathematics, 68:413–437, 2016

    work page 2016

  10. [10]

    A general framework for updating belief distributions.Journal of the Royal Statistical Society Series B: Statistical Methodology, 78(5):1103–1130, 2016

    Pier Giovanni Bissiri, Chris C Holmes, and Stephen G Walker. A general framework for updating belief distributions.Journal of the Royal Statistical Society Series B: Statistical Methodology, 78(5):1103–1130, 2016

    work page 2016

  11. [11]

    Principles of Bayesian inference using general divergence criteria

    Jack Jewson, Jim Q Smith, and Chris Holmes. Principles of Bayesian inference using general divergence criteria. Entropy, 20(6):442, 2018

    work page 2018

  12. [12]

    Towards robust inference for Bayesian filtering of linear Gaussian dynamical systems subject to additive change

    Hans Reimann. Towards robust inference for Bayesian filtering of linear Gaussian dynamical systems subject to additive change. Master’s thesis, Universität Potsdam, 2024

    work page 2024

  13. [13]

    Minimum Stein discrepancy estimators.Advances in Neural Information Processing Systems, 32, 2019

    Alessandro Barp, François-Xavier Briol, Andrew Duncan, Mark Girolami, and Lester Mackey. Minimum Stein discrepancy estimators.Advances in Neural Information Processing Systems, 32, 2019

    work page 2019

  14. [14]

    Robust and scalable Bayesian online changepoint detection.arXiv preprint arXiv:2302.04759, 2023

    Matias Altamirano, François-Xavier Briol, and Jeremias Knoblauch. Robust and scalable Bayesian online changepoint detection.arXiv preprint arXiv:2302.04759, 2023

    work page arXiv 2023

  15. [15]

    Robust and conjugate Gaussian process regression.arXiv preprint arXiv:2311.00463, 2023

    Matias Altamirano, François-Xavier Briol, and Jeremias Knoblauch. Robust and conjugate Gaussian process regression.arXiv preprint arXiv:2311.00463, 2023

    work page arXiv 2023

  16. [16]

    An adaptive ensemble filter for heavy-tailed distributions: tuning-free inflation and localization.arXiv preprint arXiv:2310.08741, 2023

    Mathieu Le Provost, Ricardo Baptista, Jeff D Eldredge, and Youssef Marzouk. An adaptive ensemble filter for heavy-tailed distributions: tuning-free inflation and localization.arXiv preprint arXiv:2310.08741, 2023

    work page arXiv 2023

  17. [17]

    A generalized t-distribution-based kernel adaptive filtering algorithm.IEEE Transactions on Circuits and Systems II: Express Briefs, 2024

    Huchuan Tang, Hongyu Han, Sheng Zhang, and Wenting Feng. A generalized t-distribution-based kernel adaptive filtering algorithm.IEEE Transactions on Circuits and Systems II: Express Briefs, 2024

    work page 2024

  18. [18]

    Cambridge University Press, 2015

    Sebastian Reich and Colin Cotter.Probabilistic forecasting and Bayesian data assimilation. Cambridge University Press, 2015

    work page 2015

  19. [19]

    Optimal information processing and Bayes’s theorem.The American Statistician, 42(4):278–280, 1988

    Arnold Zellner. Optimal information processing and Bayes’s theorem.The American Statistician, 42(4):278–280, 1988

    work page 1988

  20. [20]

    Courier Corporation, 1997

    Solomon Kullback.Information theory and statistics. Courier Corporation, 1997

    work page 1997

  21. [21]

    John Wiley & Sons, 2004

    Peter J Huber.Robust statistics, volume 523. John Wiley & Sons, 2004

    work page 2004

  22. [22]

    On the generalized distance in statistics.Sankhy ¯a: The Indian Journal of Statistics, Series A (2008-), 80:S1–S7, 2018

    Prasanta Chandra Mahalanobis. On the generalized distance in statistics.Sankhy ¯a: The Indian Journal of Statistics, Series A (2008-), 80:S1–S7, 2018. 35 Robust EnKF

    work page 2008

  23. [23]

    Robust generalised Bayesian inference for intractable likelihoods.Journal of the Royal Statistical Society Series B: Statistical Methodology, 84(3):997–1022, 2022

    Takuo Matsubara, Jeremias Knoblauch, François-Xavier Briol, and Chris J Oates. Robust generalised Bayesian inference for intractable likelihoods.Journal of the Royal Statistical Society Series B: Statistical Methodology, 84(3):997–1022, 2022

    work page 2022

  24. [24]

    Stability of the Kalman filter with stochastic time-varying parameters

    Victor Solo. Stability of the Kalman filter with stochastic time-varying parameters. InProceedings of 35th IEEE Conference on Decision and Control, volume 1, pages 57–61. IEEE, 1996

    work page 1996

  25. [25]

    Bounds on the Jensen gap, and implications for mean- concentrated distributions.arXiv preprint arXiv:1712.05267, 2017

    Xiang Gao, Meera Sitharam, and Adrian E Roitberg. Bounds on the Jensen gap, and implications for mean- concentrated distributions.arXiv preprint arXiv:1712.05267, 2017

    work page arXiv 2017

  26. [26]

    Cambridge University Press, 2023

    Simo Särkkä and Lennart Svensson.Bayesian filtering and smoothing, volume 17. Cambridge University Press, 2023

    work page 2023

  27. [27]

    Non-Gaussian state-space modeling of nonstationary time series.Journal of the American Statistical Association, 82(400):1032–1041, 1987

    Genshiro Kitagawa. Non-Gaussian state-space modeling of nonstationary time series.Journal of the American Statistical Association, 82(400):1032–1041, 1987

    work page 1987

  28. [28]

    Efficient data assimilation for spatiotemporal chaos: A local ensemble transform Kalman filter.Physica D: Nonlinear Phenomena, 230(1-2):112–126, 2007

    Brian R Hunt, Eric J Kostelich, and Istvan Szunyogh. Efficient data assimilation for spatiotemporal chaos: A local ensemble transform Kalman filter.Physica D: Nonlinear Phenomena, 230(1-2):112–126, 2007

    work page 2007

  29. [29]

    A nonparametric ensemble transform method for Bayesian inference.SIAM Journal on Scientific Computing, 35(4):A2013–A2024, 2013

    Sebastian Reich. A nonparametric ensemble transform method for Bayesian inference.SIAM Journal on Scientific Computing, 35(4):A2013–A2024, 2013

    work page 2013

  30. [30]

    World Scientific, 2022

    Sabir Umarov and Constantino Tsallis.Mathematical Foundations of Nonextensive Statistical Mechanics. World Scientific, 2022

    work page 2022

  31. [31]

    Edward N. Lorenz. Deterministic nonperiodic flow.Journal of Atmospheric Sciences, 20(2):130–141, 1963

    work page 1963

  32. [32]

    Control simulation experiments of extreme events with the Lorenz-96 model.Nonlinear Processes in Geophysics Discussions, 2022:1–18, 2022

    Qiwen Sun, Takemasa Miyoshi, and Serge Richard. Control simulation experiments of extreme events with the Lorenz-96 model.Nonlinear Processes in Geophysics Discussions, 2022:1–18, 2022

    work page 2022

  33. [33]

    Predictability: A problem partly solved

    Edward N Lorenz. Predictability: A problem partly solved. InProceedings of the Seminar on Predictability, volume 1, pages 1–18. Reading, 1996

    work page 1996

  34. [34]

    On Huber’s contaminated model.Journal of Complexity, 77:101745, 2023

    Weiyan Mu and Shifeng Xiong. On Huber’s contaminated model.Journal of Complexity, 77:101745, 2023

    work page 2023

  35. [35]

    Kota Takeda and Takashi Sakajo. Uniform error bounds of the ensemble transform Kalman filter for chaotic dynamics with multiplicative covariance inflation.SIAM/ASA Journal on Uncertainty Quantification, 12(4):1315– 1335, 2024

    work page 2024

  36. [36]

    Quantifying the minimum ensemble size for asymptotic accuracy of the ensemble Kalman filter using the degrees of instability.EGUsphere, 2025:1–18, 2025

    Kota Takeda and Takemasa Miyoshi. Quantifying the minimum ensemble size for asymptotic accuracy of the ensemble Kalman filter using the degrees of instability.EGUsphere, 2025:1–18, 2025

    work page 2025

  37. [37]

    Posterior consistency for Gaussian process approximations of Bayesian posterior distributions.Mathematics of Computation, 87(310):721–753, 2018

    Andrew Stuart and Aretha Teckentrup. Posterior consistency for Gaussian process approximations of Bayesian posterior distributions.Mathematics of Computation, 87(310):721–753, 2018

    work page 2018

  38. [38]

    A robust generalized t distribution-based Kalman filter

    Mingming Bai, Chengjiao Sun, and Yonggang Zhang. A robust generalized t distribution-based Kalman filter. IEEE Transactions on Aerospace and Electronic Systems, 58(5):4771–4781, 2022

    work page 2022

  39. [39]

    JHU Press, 2013

    Gene H Golub and Charles F Van Loan.Matrix computations. JHU Press, 2013

    work page 2013

  40. [40]

    Generalizing Bayesian inference: Updating a 250 years old theorem for the 21st century, August 2021

    Lorenzo Pacchiardi. Generalizing Bayesian inference: Updating a 250 years old theorem for the 21st century, August 2021. Accessed: 2024-05-22

    work page 2021

  41. [41]

    Data assimilation: mathematics for merging models and data.Snapshots of modern mathematics from Oberwolfach, 2018(11), 2018

    Matthias Morzfeld and Sebastian Reich. Data assimilation: mathematics for merging models and data.Snapshots of modern mathematics from Oberwolfach, 2018(11), 2018

    work page 2018

  42. [42]

    Generalized Bayesian likelihood-free inference using scoring rules estimators.arXiv preprint arXiv:2104.03889, 2021

    Lorenzo Pacchiardi and Ritabrata Dutta. Generalized Bayesian likelihood-free inference using scoring rules estimators.arXiv preprint arXiv:2104.03889, 2021

    work page arXiv 2021

  43. [43]

    User-friendly introduction to PAC-Bayes bounds.Foundations and Trends® in Machine Learning, 17(2):174–303, 2024

    Pierre Alquier et al. User-friendly introduction to PAC-Bayes bounds.Foundations and Trends® in Machine Learning, 17(2):174–303, 2024

    work page 2024

  44. [44]

    Spatio-temporal Bayesian on-line changepoint detection with model selection

    Jeremias Knoblauch and Theodoros Damoulas. Spatio-temporal Bayesian on-line changepoint detection with model selection. InInternational Conference on Machine Learning, pages 2718–2727. PMLR, 2018

    work page 2018

  45. [45]

    Springer, 2020

    Nicolas Chopin, Omiros Papaspiliopoulos, et al.An introduction to sequential Monte Carlo, volume 4. Springer, 2020

    work page 2020

  46. [46]

    An adaptive sequential Monte Carlo method for approximate Bayesian computation.Statistics and Computing, 22(5):1009–1020, 2012

    Pierre Del Moral, Arnaud Doucet, and Ajay Jasra. An adaptive sequential Monte Carlo method for approximate Bayesian computation.Statistics and Computing, 22(5):1009–1020, 2012

    work page 2012

  47. [47]

    On the choice of MCMC kernels for approximate Bayesian computation with SMC samplers

    Anthony Lee. On the choice of MCMC kernels for approximate Bayesian computation with SMC samplers. In Proceedings of the 2012 Winter Simulation Conference (WSC), pages 1–12. IEEE, 2012

    work page 2012

  48. [48]

    Everitt, Adam M

    Xavier Didelot, Richard G. Everitt, Adam M. Johansen, and Daniel J. Lawson. Likelihood-free estimation of model evidence.Bayesian Analysis, 6(1):49–76, 2011. 36 Robust EnKF

    work page 2011

  49. [49]

    Improving approximate Bayesian computation via quasi-Monte Carlo

    Alexander Buchholz and Nicolas Chopin. Improving approximate Bayesian computation via quasi-Monte Carlo. Journal of Computational and Graphical Statistics, 28(1):205–219, 2019

    work page 2019

  50. [50]

    Interpreting diffusion score matching using normalizing flow.arXiv preprint arXiv:2107.10072, 2021

    Wenbo Gong and Yingzhen Li. Interpreting diffusion score matching using normalizing flow.arXiv preprint arXiv:2107.10072, 2021

    work page arXiv 2021

  51. [51]

    Estimation of non-normalized statistical models by score matching.Journal of Machine Learning Research, 6(4), 2005

    Aapo Hyvärinen and Peter Dayan. Estimation of non-normalized statistical models by score matching.Journal of Machine Learning Research, 6(4), 2005

    work page 2005

  52. [52]

    Measuring sample quality with diffusions.The Annals of Applied Probability, 29(5):2884–2928, 2019

    Jackson Gorham, Andrew B Duncan, Sebastian J V ollmer, and Lester Mackey. Measuring sample quality with diffusions.The Annals of Applied Probability, 29(5):2884–2928, 2019

    work page 2019

  53. [53]

    Estimating density models with truncation boundaries using score matching.The Journal of Machine Learning Research, 23(1):8448–8485, 2022

    Song Liu, Takafumi Kanamori, and Daniel J Williams. Estimating density models with truncation boundaries using score matching.The Journal of Machine Learning Research, 23(1):8448–8485, 2022

    work page 2022

  54. [54]

    Towards healing the blindness of score matching.arXiv preprint arXiv:2209.07396, 2022

    Mingtian Zhang, Oscar Key, Peter Hayes, David Barber, Brooks Paige, and François-Xavier Briol. Towards healing the blindness of score matching.arXiv preprint arXiv:2209.07396, 2022

    work page arXiv 2022

  55. [55]

    Interpretation and Generalization of Score Matching

    Siwei Lyu. Interpretation and generalization of score matching.arXiv preprint arXiv:1205.2629, 2012

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  56. [56]

    Stein’s method meets computational statistics: A review of some recent developments.Statistical Science, 38(1):120–139, 2023

    Andreas Anastasiou, Alessandro Barp, François-Xavier Briol, Bruno Ebner, Robert E Gaunt, Fatemeh Ghader- inezhad, Jackson Gorham, Arthur Gretton, Christophe Ley, Qiang Liu, et al. Stein’s method meets computational statistics: A review of some recent developments.Statistical Science, 38(1):120–139, 2023

    work page 2023

  57. [57]

    A kernelized Stein discrepancy for goodness-of-fit tests

    Qiang Liu, Jason Lee, and Michael Jordan. A kernelized Stein discrepancy for goodness-of-fit tests. InInterna- tional Conference on Machine Learning, pages 276–284. PMLR, 2016

    work page 2016

  58. [58]

    Springer, 2004

    Pierre Del Moral.Feynman-Kac formulae: genealogical and interacting particle systems with applications. Springer, 2004

    work page 2004

  59. [59]

    On the convergence of two sequential Monte Carlo methods for maximum a posteriori sequence estimation and stochastic global optimization.Statistics and Computing, 23:91–107, 2013

    Joaquín Míguez, Dan Crisan, and Petar M Djuri´c. On the convergence of two sequential Monte Carlo methods for maximum a posteriori sequence estimation and stochastic global optimization.Statistics and Computing, 23:91–107, 2013. 37 Robust EnKF Appendix A Constructing the Kalman filter from conjugacy We provide the construction of the regular Kalman filter...

    work page 2013