pith. sign in

arxiv: 2606.10111 · v1 · pith:UUG7QHR3new · submitted 2026-06-08 · 💻 cs.LG · cs.SY· eess.SY· stat.ML

Nonlinear Estimator: Dual Bayesian Affine Estimators for Parameter Learning

Sasan Vakili , Dani\"el Woonings , Pradyumna Paruchuri , Peyman Mohajerin Esfahani This is my paper

Pith reviewed 2026-06-27 17:32 UTC · model grok-4.3

classification 💻 cs.LG cs.SYeess.SYstat.ML
keywords nonlinear estimationWiener-type modelsaffine MMSE estimatorsdual Bayesian estimatorsfixed-point iterationdynamic basis statisticsparameter learningstate-space models
0
0 comments X

The pith

Coupling two affine MMSE estimators in a fixed-point loop yields a nonlinear parameter estimator with the lowest error for Wiener-type state-space models.

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

The paper develops a nonlinear parameter estimator for Wiener-type state-space models as a fixed-point architecture that couples an affine MMSE estimator for unknown parameters with an affine MMSE estimator for latent variables. The architecture uses Dynamic Basis Statistics to summarize nonlinear basis-function evaluations while preserving the affine structure. Two variants are presented, and Monte Carlo experiments show that the dual state-parameter version, which derives Dynamic Basis Statistics from state-estimate statistics via a Gaussian operator, attains the lowest parameter mean-squared error. This outperforms the dual basis-parameter version, the purely affine estimator, and sequential Monte Carlo versions of Particle Gibbs and Expectation-Maximization. A sympathetic reader would care because such estimators enable accurate learning of system parameters from noisy observations without relying on computationally intensive sampling methods.

Core claim

The dual state-parameter estimator achieves the lowest parameter mean-squared error by first computing affine state estimates and their covariances, mapping these through a Gaussian Dynamic Basis Statistics operator, and then alternating with the affine parameter estimator in a fixed-point iteration that uses plug-in statistics from the previous iteration; this performance is superior to that of the dual basis-parameter estimator, the purely affine parameter estimator, and sequential Monte Carlo variants of Particle Gibbs and Expectation-Maximization schemes.

What carries the argument

The fixed-point architecture coupling an affine minimum mean-squared error estimator for parameters with one for latent variables via Dynamic Basis Statistics estimates that summarize nonlinear basis-function evaluations.

If this is right

  • The dual state-parameter estimator attains lower parameter mean-squared error than the dual basis-parameter estimator.
  • The dual state-parameter estimator attains lower parameter mean-squared error than the purely affine parameter estimator.
  • The dual state-parameter estimator attains lower parameter mean-squared error than sequential Monte Carlo variants of Particle Gibbs and Expectation-Maximization.
  • Both the dual basis-parameter and dual state-parameter estimators admit fixed-point characterizations that alternate between the two components.

Where Pith is reading between the lines

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

  • The fixed-point method could be applied to parameter estimation in other nonlinear state-space model classes.
  • Improved parameter accuracy may enhance performance in related tasks such as state filtering or prediction.
  • Convergence properties of the iterations could be analyzed to determine conditions for guaranteed improvement.
  • The approach offers a deterministic alternative that may scale better than particle methods for large data sets.

Load-bearing premise

The fixed-point iterations that alternate between the two affine estimators using plug-in statistics from the previous iteration converge to a useful solution that improves upon the purely affine estimator.

What would settle it

Monte Carlo experiments on Wiener-type state-space models in which the dual state-parameter estimator does not achieve the lowest parameter mean-squared error compared to the dual basis-parameter estimator, the purely affine estimator, and the particle-based methods.

Figures

Figures reproduced from arXiv: 2606.10111 by Dani\"el Woonings, Peyman Mohajerin Esfahani, Pradyumna Paruchuri, Sasan Vakili.

Figure 1
Figure 1. Figure 1: Optimal MMSE in (5) vs. affine MMSE in (10a) for estimating θ given y In this work, we propose an analytical approach to construct a nonlinear estimator that improves upon the affine class. We revisit Example 1 throughout the paper to illustrate the behaviour of the proposed estimation algorithms. Contributions. Building on this direction, we propose a nonlinear parameter estimator that retains the functio… view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of the proposed nonlinear parameter estimator induces an interdependence between the two blocks: the parameter estimator requires DBS statistics, and the DBS estimator, in turn, depends on the parameter estimates. We resolve this interdependence via a fixed-point characterization that iteratively refines both sets of statistics. Leveraging the closed￾form structure of the affine MMSE estimator… view at source ↗
Figure 3
Figure 3. Figure 3: Architectures of the Dynamic Basis Statistics estimator Combining either of the proposed DBS estimation methods with the affine parameter estimator in￾duces an interdependence between the parameters and the DBS statistics, resolved by the algebraic equations defining DB-P and DS-P estimators. These estimators naturally lead to fixed-point charac￾terizations, implemented by Algorithm 1, that alternate betwe… view at source ↗
Figure 4
Figure 4. Figure 4: DB-P, optimal MMSE (5), and affine (10a) estimators. In the above example, the basis function was linear, so estimating basis-function evaluations is equivalent to estimating the latent state. However, for nonlinear basis functions, this equivalence no longer holds, and the dual basis-parameter estimator can become both statistically and computationally inefficient. These limitations motivate an alternativ… view at source ↗
Figure 5
Figure 5. Figure 5: DS-P, DB-P, optimal MMSE (5), and affine MMSE (10a) estimators 3.3. Fixed-point algorithm The dual basis-parameter (DB-P) and dual state-parameter (DS-P) estimators can be equivalently characterized by a fixed-point solution ζ ‹ of an operator F, satisfying ζ ‹ “ Fpζ ‹ q. We collect the esti￾mates and their covariances into the tuple: ζ “ ` θp nl, Σθp nl , Φpnl, ΣΦp nl ˘ for DB-P and ζ “ ` θp nl, Σθp nl , … view at source ↗
Figure 6
Figure 6. Figure 6: Parameter estimation squared-error distributions for Experiment setup 1 [PITH_FULL_IMAGE:figures/full_fig_p022_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Parameter estimation squared-error distributions for Experiment setup 2 Benchmark 2.2 (Trajectory horizon effects). This experiment considers 10,000 simulations with trajectories of varying lengths, T P t0, 4, 10, 13, 16, 20, 25, 32, 40, 50, 63, 79, 100u, to examine how parameter estimation error evolves as the number of measurements increases [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Parameter estimation squared-error vs. trajectory length T for Experiment setup 2 Benchmark 2.3 (Dual state-parameter vs. SMC methods). In this final benchmark, DS-P is compared with two well-known SMC-based approaches, PGAS and SMC-EM. Both methods alternate between updating the latent states given the current parameter estimates and updating the parameters given sampled state trajectories, whereas DS-P a… view at source ↗
Figure 9
Figure 9. Figure 9: Squared-error distributions: dual state-parameter vs. SMC-based methods with that of DS-P but also exhibits a substantial number of simulations with noticeably larger errors, raising its overall MSE. PGAS shows a similar pattern: one concentration of errors near DS-P and another at much larger values, again leading to a higher MSE. When the process noise increases to σ 2 w “ 0.01, the squared-error distrib… view at source ↗
read the original abstract

This paper presents a nonlinear parameter estimator for Wiener-type state-space models obtained as a fixed-point architecture that couples two affine minimum mean-squared error (MMSE) estimators: one for the unknown parameters and one for latent variables. The architecture retains the functional structure of the optimal affine MMSE parameter estimator while incorporating Dynamic Basis Statistics (DBS) estimates that summarize nonlinear basis-function evaluations. Two DBS construction strategies are developed, leading to two nonlinear estimator frameworks. The dual basis-parameter estimator combines an affine basis estimator with the affine parameter estimator, whereas the dual state-parameter estimator first computes affine state estimates and their covariances, then maps these state-estimate statistics through a Gaussian DBS operator to obtain DBS estimates. Both dual estimators admit fixed-point characterizations that alternate between estimating each component using the updated prior of the other, obtained from that component's plug-in estimate statistics from the previous iteration. The efficacy of the proposed methods is examined via extensive Monte Carlo experiments, showing that the dual basis-parameter estimator attains parameter mean-squared errors comparable to those of the purely affine parameter estimator, while the dual state-parameter estimator achieves the lowest parameter mean-squared error, outperforming both the dual basis-parameter and purely affine parameter estimators, as well as sequential Monte Carlo variants of classical Particle Gibbs and Expectation-Maximization schemes.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The manuscript proposes nonlinear parameter estimators for Wiener-type state-space models constructed by coupling two affine MMSE estimators (one for parameters, one for latent states or basis functions) through Dynamic Basis Statistics (DBS) operators. Two frameworks are developed: the dual basis-parameter estimator and the dual state-parameter estimator. Both are realized via fixed-point iterations that alternate between the component estimators using plug-in statistics from the prior iterate. Monte Carlo experiments are reported to show that the dual state-parameter estimator attains the lowest parameter mean-squared error, outperforming the dual basis-parameter estimator, the purely affine parameter estimator, and SMC variants of Particle Gibbs and EM.

Significance. If the fixed-point iterations are shown to converge reliably to points that improve upon the affine baseline, the approach could supply an efficient nonlinear estimator that reuses existing affine MMSE components rather than requiring full particle methods. The reported empirical ordering is potentially useful for parameter learning in nonlinear state-space models, but the lack of supporting analysis limits the strength of the performance claims.

major comments (2)
  1. [Abstract] Abstract (paragraph on fixed-point characterizations): The central empirical claim—that the dual state-parameter estimator achieves the lowest parameter MSE—rests on fixed-point iterations that alternate between the affine state and parameter estimators using plug-in DBS statistics from the previous iterate. No contraction argument, Lyapunov function, convergence rate, or even empirical diagnostics (e.g., iteration trajectories or stability checks) are supplied to establish that the map reaches a useful fixed point that improves on the purely affine estimator. Without such analysis the reported Monte Carlo superiority cannot be guaranteed to follow from the stated procedure.
  2. [Monte Carlo experiments] Monte Carlo experiments section: The performance ordering is supported only by Monte Carlo experiments whose details (number of trials, data-generation parameters, error-bar reporting, and exclusion criteria) are not fully specified. This makes it impossible to assess whether the observed advantage of the dual state-parameter estimator is robust or sensitive to implementation choices in the fixed-point loop.
minor comments (2)
  1. The definition and construction of the DBS operators (both strategies) should be stated with explicit equations early in the manuscript so that the plug-in usage in the fixed-point iterations is immediately verifiable.
  2. Notation for the two dual estimators and the Gaussian DBS operator should be made consistent between the abstract and the main text to avoid reader confusion.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on convergence and experimental reproducibility. We address each major point below and commit to revisions that strengthen the presentation without altering the core claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract (paragraph on fixed-point characterizations): The central empirical claim—that the dual state-parameter estimator achieves the lowest parameter MSE—rests on fixed-point iterations that alternate between the affine state and parameter estimators using plug-in DBS statistics from the previous iterate. No contraction argument, Lyapunov function, convergence rate, or even empirical diagnostics (e.g., iteration trajectories or stability checks) are supplied to establish that the map reaches a useful fixed point that improves on the purely affine estimator. Without such analysis the reported Monte Carlo superiority cannot be guaranteed to follow from the stated procedure.

    Authors: We agree that the manuscript provides no theoretical convergence analysis (contraction mapping, Lyapunov function, or rate) for the fixed-point iteration. The iteration is constructed by alternating the two affine MMSE estimators with plug-in DBS statistics from the prior iterate, which preserves the optimality of each component under the current statistics. To strengthen the empirical claim, we will add iteration-trajectory plots and stability checks across random initializations in the revised Monte Carlo section. These diagnostics will show rapid convergence to points that improve on the affine baseline. A full contraction argument would require additional model assumptions and is beyond the present scope. revision: partial

  2. Referee: [Monte Carlo experiments] Monte Carlo experiments section: The performance ordering is supported only by Monte Carlo experiments whose details (number of trials, data-generation parameters, error-bar reporting, and exclusion criteria) are not fully specified. This makes it impossible to assess whether the observed advantage of the dual state-parameter estimator is robust or sensitive to implementation choices in the fixed-point loop.

    Authors: We acknowledge that the experimental protocol was not described at the required level of detail. In the revision we will explicitly state: 500 independent Monte Carlo trials, the precise Wiener-model parameters (nonlinearity, process and measurement noise variances, basis-function dimension), reporting of mean and standard-deviation error bars across trials, and confirmation that no trials were excluded on the basis of convergence failure. These additions will permit direct assessment of robustness to the fixed-point implementation. revision: yes

Circularity Check

0 steps flagged

No circularity: construction and empirical comparison are independent of target quantities.

full rationale

The paper defines the dual estimators explicitly as fixed-point couplings of standard affine MMSE estimators with newly introduced DBS operators; the fixed-point characterization is a definitional property of the iteration, not a reduction of the performance claim. Parameter MSE superiority is asserted only via Monte Carlo comparison against Particle Gibbs, EM, and purely affine baselines, with no equations that fit parameters to the target MSE or rename fitted statistics as predictions. No self-citation load-bearing steps, uniqueness theorems, or ansatz smuggling appear in the provided derivation chain. The architecture is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on standard assumptions for affine MMSE optimality and the validity of the Gaussian DBS operator for summarizing nonlinear evaluations; no free parameters are explicitly fitted in the abstract description, and DBS is introduced as a methodological construct rather than a new physical entity.

axioms (1)
  • domain assumption Affine MMSE estimators are optimal under the linear-Gaussian assumptions implicit in the Wiener model components
    The architecture retains the functional structure of the optimal affine MMSE parameter estimator.
invented entities (1)
  • Dynamic Basis Statistics (DBS) no independent evidence
    purpose: Summarize nonlinear basis-function evaluations within an affine estimation framework
    Two DBS construction strategies are developed to enable the dual estimators.

pith-pipeline@v0.9.1-grok · 5784 in / 1299 out tokens · 20227 ms · 2026-06-27T17:32:29.332431+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

53 extracted references · 1 linked inside Pith

  1. [1]

    Particle Markov chain Monte Carlo methods.Journal of the Royal Statistical Society Series B: Statistical Methodology, 72:269–342, 2010

    Christophe Andrieu, Arnaud Doucet, and Roman Holenstein. Particle Markov chain Monte Carlo methods.Journal of the Royal Statistical Society Series B: Statistical Methodology, 72:269–342, 2010

    2010

  2. [2]

    Wainwright, and Bin Yu

    Sivaraman Balakrishnan, Martin J. Wainwright, and Bin Yu. Statistical guarantees for the EM algorithm: From population to sample-based analysis.The Annals of Statistics, 45:77–120, 2017

    2017

  3. [3]

    On Markov chain Monte Carlo methods for tall data.Journal of Machine Learning Research, 18:1–43, 2017

    Rémi Bardenet, Arnaud Doucet, and Chris Holmes. On Markov chain Monte Carlo methods for tall data.Journal of Machine Learning Research, 18:1–43, 2017

    2017

  4. [4]

    Bauschke and Patrick L

    Heinz H. Bauschke and Patrick L. Combettes.Convex Analysis and Monotone Operator Theory in Hilbert Spaces. Springer, Cham, 2nd edition, 2017

    2017

  5. [5]

    Beal and Zoubin Ghahramani

    Matthew J. Beal and Zoubin Ghahramani. The variational Bayesian EM algorithm for incomplete data: with application to scoring graphical model structures.Bayesian Statistics, 7:453–463, 2003

    2003

  6. [6]

    Curse-of-dimensionality revisited: Collapse of the particle filter in very large scale systems

    Thomas Bengtsson, Peter Bickel, and Bo Li. Curse-of-dimensionality revisited: Collapse of the particle filter in very large scale systems. InProbability and Statistics: Essays in Honor of David A. Freedman, pages 316–335. Institute of Mathematical Statistics, 2008

    2008

  7. [7]

    Bishop.Pattern Recognition and Machine Learning

    Christopher M. Bishop.Pattern Recognition and Machine Learning. Springer, 2006. NONLINEAR ESTIMATOR: DUAL BAYESIAN AFFINE ESTIMATORS FOR PARAMETER LEARNING 31

    2006

  8. [8]

    Blei, Alp Kucukelbir, and Jon D

    David M. Blei, Alp Kucukelbir, and Jon D. McAuliffe. Variational inference: A review for statisticians.Journal of the American statistical Association, 112:859–877, 2017

    2017

  9. [9]

    Improved particle approximations to the joint smoothing distribution using Markov chain Monte Carlo.IEEE Transactions on Signal Processing, 61:956–963, 2012

    Pete Bunch and Simon Godsill. Improved particle approximations to the joint smoothing distribution using Markov chain Monte Carlo.IEEE Transactions on Signal Processing, 61:956–963, 2012

    2012

  10. [10]

    Burden and J

    Richard L. Burden and J. Douglas Faires.Numerical Analysis. Brooks/Cole, 9th edition, 2010

    2010

  11. [11]

    Springer, 2005

    Olivier Cappé, Eric Moulines, and Tobias Rydén.Inference in Hidden Markov Models. Springer, 2005

    2005

  12. [12]

    Carlin and Thomas A

    Bradley P. Carlin and Thomas A. Louis.Bayes and Empirical Bayes Methods for Data Analysis. Chapman and Hall/CRC, 2nd edition, 2000

    2000

  13. [13]

    Carlton and Jay L

    Matthew A. Carlton and Jay L. Devore.Probability with Applications in Engineering, Science, and Technology. Springer, 2017

    2017

  14. [14]

    Springer International Publishing, 2020

    Nicolas Chopin and Omiros Papaspiliopoulos.An Introduction to Sequential Monte Carlo. Springer International Publishing, 2020

    2020

  15. [15]

    Nicolas Chopin and Sumeetpal S. Singh. On particle Gibbs sampling.Bernoulli, 21:1855–1883, 2015

    2015

  16. [16]

    Wills, Thomas B

    Jarrad Courts, Adrian G. Wills, Thomas B. Schön, and Brett Ninness. Variational system identification for nonlinear state-space models.Automatica, 147:110687, 2023

    2023

  17. [17]

    On backward smoothing algorithms.The Annals of Statistics, 51:2145—-2169, 2023

    Hai-Dang Dau and Nicolas Chopin. On backward smoothing algorithms.The Annals of Statistics, 51:2145—-2169, 2023

    2023

  18. [18]

    Convergence of a stochastic approximation version of the EM algorithm.The Annals of Statistics, 27:94–128, 1999

    Bernard Delyon, Marc Lavielle, and Eric Moulines. Convergence of a stochastic approximation version of the EM algorithm.The Annals of Statistics, 27:94–128, 1999

    1999

  19. [19]

    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 (Methodological), 39:1–38, 1977

    1977

  20. [20]

    Springer, 2001

    Arnaud Doucet, Nando De Freitas, and Neil Gordon.Sequential Monte Carlo Methods in Practice. Springer, 2001

    2001

  21. [21]

    On sequential Monte Carlo sampling methods for Bayesian filtering.Statistics and Computing, 10:197–208, 2000

    Arnaud Doucet, Simon Godsill, and Christophe Andrieu. On sequential Monte Carlo sampling methods for Bayesian filtering.Statistics and Computing, 10:197–208, 2000

    2000

  22. [22]

    Hidden Markov processes.IEEE Transactions on Information Theory, 48:1518– 1569, 2002

    Yariv Ephraim and Neri Merhav. Hidden Markov processes.IEEE Transactions on Information Theory, 48:1518– 1569, 2002

    2002

  23. [23]

    Lower and upper bounds on the minimum mean-square error in composite source signal estimation.IEEE transactions on Information Theory, 38:1709–1724, 2002

    Yariv Ephraim and Neri Merhav. Lower and upper bounds on the minimum mean-square error in composite source signal estimation.IEEE transactions on Information Theory, 38:1709–1724, 2002

    2002

  24. [24]

    Bayesian system ID: optimal management of parameter, model, and measurement uncertainty.Nonlinear Dynamics, 102:241–267, 2020

    Nicholas Galioto and Alex Arkady Gorodetsky. Bayesian system ID: optimal management of parameter, model, and measurement uncertainty.Nonlinear Dynamics, 102:241–267, 2020

    2020

  25. [25]

    Gelfand and Adrian F

    Alan E. Gelfand and Adrian F. M. Smith. Sampling-based approaches to calculating marginal densities.Journal of the American Statistical Association, 85:398–409, 1990

    1990

  26. [26]

    Carlin, Hal S

    Andrew Gelman, John B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin.Bayesian Data Analysis. CRC Press, 3rd edition, 2013

    2013

  27. [27]

    Parameter estimation for linear dynamical systems.Technical Report, University of Toronto:CRG–TR–96–2, 1996

    Zoubin Ghahramani and Geoffrey E Hinton. Parameter estimation for linear dynamical systems.Technical Report, University of Toronto:CRG–TR–96–2, 1996

    1996

  28. [28]

    Godsill, Arnaud Doucet, and Mike West

    Simon J. Godsill, Arnaud Doucet, and Mike West. Monte Carlo smoothing for nonlinear time series.Journal of the American Statistical Association, 99:156–168, 2004

    2004

  29. [29]

    Keith Hastings

    W. Keith Hastings. Monte Carlo sampling methods using Markov chains and their applications.Biometrika, 57:97– 109, 1970

    1970

  30. [30]

    John Wiley & Sons, 2004

    Simon Haykin.Kalman Filtering and Neural Networks. John Wiley & Sons, 2004

    2004

  31. [31]

    Stein’s lemma for elliptical random vectors.Journal of Multivariate Analysis, 99:912–927, 2008

    Zinoviy Landsman and Johanna Nešlehová. Stein’s lemma for elliptical random vectors.Journal of Multivariate Analysis, 99:912–927, 2008

    2008

  32. [32]

    Levy.Principles of Signal Detection and Parameter Estimation

    Bernard C. Levy.Principles of Signal Detection and Parameter Estimation. Springer, 2008

    2008

  33. [33]

    Yifang Li and Sujit K. Ghosh. Efficient sampling methods for truncated multivariate Normal and Student-t distri- butions subject to linear inequality constraints.Journal of Statistical Theory and Practice, 9:712–732, 2015

    2015

  34. [34]

    An efficient stochastic approximation EM algorithm using conditional particle filters

    Fredrik Lindsten. An efficient stochastic approximation EM algorithm using conditional particle filters. InIEEE International Conference on Acoustics, Speech and Signal Processing, pages 6274–6278, 2013

    2013

  35. [35]

    Jordan, and Thomas B

    Fredrik Lindsten, Michael I. Jordan, and Thomas B. Schön. Particle Gibbs with ancestor sampling.Journal of Machine Learning Research, 15:2145–2184, 2014. NONLINEAR ESTIMATOR: DUAL BAYESIAN AFFINE ESTIMATORS FOR PARAMETER LEARNING 32

    2014

  36. [36]

    Jun S. Liu. Siegel’s formula via Stein’s identities.Statistics & Probability Letters, 21:247–251, 1994

    1994

  37. [37]

    Prentice Hall PTR, 2nd edition, 1999

    Lennart Ljung.System Identification: Theory for the User. Prentice Hall PTR, 2nd edition, 1999

    1999

  38. [38]

    David J. C. MacKay.Information Theory, Inference and Learning Algorithms. Cambridge University Press, 2003

    2003

  39. [39]

    Rosenbluth, Marshall N

    Nicholas Metropolis, Arianna W. Rosenbluth, Marshall N. Rosenbluth, Augusta H. Teller, and Edward Teller. Equation of state calculations by fast computing machines.The Journal of Chemical Physics, 21:1087–1092, 1953

    1953

  40. [40]

    The Fubini theorem

    Jan Mikusiński. The Fubini theorem. InThe Bochner Integral, pages 91–105. Springer, 1978

    1978

  41. [41]

    Nested sequential Monte Carlo methods

    Christian Naesseth, Fredrik Lindsten, and Thomas Schön. Nested sequential Monte Carlo methods. InInternational Conference on Machine Learning, pages 1292–1301, 2015

    2015

  42. [42]

    Naesseth, Fredrik Lindsten, and Thomas B

    Christian A. Naesseth, Fredrik Lindsten, and Thomas B. Schön. Elements of sequential Monte Carlo.Foundations and Trends in Machine Learning, 12:307–392, 2019

    2019

  43. [43]

    Pitt, Ralph dos Santos Silva, Paolo Giordani, and Robert Kohn

    Michael K. Pitt, Ralph dos Santos Silva, Paolo Giordani, and Robert Kohn. On some properties of Markov chain Monte Carlo simulation methods based on the particle filter.Journal of Econometrics, 171:134–151, 2012

    2012

  44. [44]

    MCMC-based inference in the era of big data: A fundamental analysis of the convergence complexity of high-dimensional chains.arXiv preprint arXiv:1508.00947, 2015

    Bala Rajaratnam and Doug Sparks. MCMC-based inference in the era of big data: A fundamental analysis of the convergence complexity of high-dimensional chains.arXiv preprint arXiv:1508.00947, 2015

    Pith/arXiv arXiv 2015

  45. [45]

    Robert and George Casella.Monte Carlo Statistical Methods

    Christian P. Robert and George Casella.Monte Carlo Statistical Methods. Springer, 2nd edition, 2004

    2004

  46. [46]

    Roberts and Jeffrey S

    Gareth O. Roberts and Jeffrey S. Rosenthal. General state space Markov chains and MCMC algorithms.Probability Surveys, 1:20–71, 2004

    2004

  47. [47]

    Cambridge University Press, 2013

    Simo Särkkä.Bayesian Filtering and Smoothing. Cambridge University Press, 2013

    2013

  48. [48]

    Schön, Adrian Wills, and Brett Ninness

    Thomas B. Schön, Adrian Wills, and Brett Ninness. System identification of nonlinear state-space models.Auto- matica, 47:39–49, 2011

    2011

  49. [49]

    Identification of block-oriented nonlinear systems starting from linear approxi- mations: A survey.Automatica, 85:272–292, 2017

    Maarten Schoukens and Koen Tiels. Identification of block-oriented nonlinear systems starting from linear approxi- mations: A survey.Automatica, 85:272–292, 2017

    2017

  50. [50]

    Charles M. Stein. Estimation of the mean of a multivariate Normal distribution.The Annals of Statistics, 9:1135– 1151, 1981

    1981

  51. [51]

    Tokdar and Robert E

    Surya T. Tokdar and Robert E. Kass. Importance sampling: A review.Wiley Interdisciplinary Reviews: Computa- tional Statistics, 2:54–60, 2010

    2010

  52. [52]

    Optimal Bayesian affine estimator and active learning for the Wiener model.arXiv preprint arXiv:2504.05490, 2025

    Sasan Vakili, Manuel Mazo Jr, and Peyman Mohajerin Esfahani. Optimal Bayesian affine estimator and active learning for the Wiener model.arXiv preprint arXiv:2504.05490, 2025

    arXiv 2025

  53. [53]

    Dual Kalman filtering methods for nonlinear prediction, smoothing and estimation

    Eric Wan and Alex Nelson. Dual Kalman filtering methods for nonlinear prediction, smoothing and estimation. In Advances in Neural Information Processing Systems, volume 9, pages 793–799, 1996

    1996