pith. sign in

arxiv: 2605.21698 · v1 · pith:UYGEEWUHnew · submitted 2026-05-20 · 📊 stat.CO

A Gaussian Sum Filter for Unifying Gaussian and Particle Filters

Kostas Tsampourakis , V\'ictor Elvira This is my paper

Pith reviewed 2026-05-22 07:46 UTC · model grok-4.3

classification 📊 stat.CO
keywords Gaussian sum filterparticle filterBayesian filteringstate-space modelsadaptive filteringnonlinear filteringunification
0
0 comments X

The pith

An augmented Gaussian sum filter unifies Gaussian sum filters and particle filters through continuous interpolation via tunable covariances.

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

The paper proposes the Augmented Gaussian Sum Filter to combine the strengths of Gaussian sum filters, which use local Gaussian approximations, and particle filters, which rely on Monte Carlo sampling for nonlinear state-space models. By introducing an augmented Gaussian approximation that includes latent variables and adjustable covariance parameters, the method creates a continuous spectrum of behaviors between the two families. Both standard Gaussian sum filters and particle filters emerge as limiting cases when the covariances are set to specific values. An adaptive variant automatically adjusts its operation based on the local strength of nonlinearities, favoring efficiency where Gaussians suffice and robustness elsewhere. The approach is illustrated in a target-tracking task where it avoids common breakdowns of either parent method.

Core claim

The Augmented Gaussian Sum Filter represents the posterior via an augmented Gaussian parameterized by latent variables and tunable covariance parameters; adjusting these covariances produces continuous interpolation between Gaussian-sum and particle-filter regimes, with both recovered exactly as special cases, and an adaptive version that switches behavior according to local nonlinearity.

What carries the argument

Augmented Gaussian approximation parameterized by latent variables and tunable covariance parameters, which enables continuous interpolation between GSF-like and PF-like behavior.

If this is right

  • When local nonlinearities are mild the filter reverts to efficient Gaussian-sum behavior.
  • When nonlinearities intensify it automatically adopts particle-filter robustness.
  • Both endpoint methods are recovered exactly by fixing the covariance parameters to particular values.
  • The framework avoids the numerical instabilities typical of pure Gaussian sum filters and the high particle counts typical of pure particle filters.

Where Pith is reading between the lines

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

  • The same covariance-tuning idea could be applied to unify other pairs of approximate filters in sequential Monte Carlo settings.
  • The adaptive switching rule might be ported to higher-dimensional or non-Gaussian base approximations.
  • Empirical validation on additional benchmark tracking problems would clarify how quickly the method detects and responds to changes in nonlinearity strength.

Load-bearing premise

An augmented Gaussian approximation with latent variables and tunable covariances can represent the posterior well enough to support stable interpolation and adaptive switching without new instabilities or large errors in strongly nonlinear settings.

What would settle it

In a strongly nonlinear target-tracking simulation where a standard Gaussian sum filter diverges, check whether the adaptive AGSF maintains bounded error while using far fewer particles than a standard particle filter.

Figures

Figures reproduced from arXiv: 2605.21698 by Kostas Tsampourakis, V\'ictor Elvira.

Figure 1
Figure 1. Figure 1: (a) Geometric illustration of the linear approximation [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Splitting, prediction and update steps for a component [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Experiment B. (Top) MSE plotted in time for L￾GSF, U-GSF(M = 1000), L-AGSF, U-AGSF(M = 100, N = 5, L = 5) and BPF (M = 100). (Bottom) Plot of the proportionality parameters of the update step ρ2. We see that when the model is linear the AGSF algorithms behave like the GSF and when it becomes nonlinear (MSV) the proportionality decreases, leading the AGSF algorithms to behave more like a particle filter. AG… view at source ↗
Figure 4
Figure 4. Figure 4: Plots of samples from the posteriors of different algorithms for the target tracking problem with [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Plots of samples from the posteriors of different algorithms for the target tracking problem with [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Plots of samples from the posteriors of different algorithms for the target tracking problem with [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Plots of samples from the posteriors of different algorithms for the target tracking problem with [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Plots of samples from the posteriors of different algorithms for the target tracking problem with [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Plots of samples from the posteriors of different algorithms for the target tracking problem with [PITH_FULL_IMAGE:figures/full_fig_p019_9.png] view at source ↗
read the original abstract

State-space models (SSMs) are a broad class of probabilistic models for dynamical systems with many applications in engineering and science. Bayesian filtering is analytically tractable only in the linear-Gaussian setting, where the Kalman filter yields exact posterior distributions. For nonlinear or non-Gaussian SSMs, approximations are required. Two prominent families of approximate methods are Gaussian sum filters (GSFs), which rely on local Gaussian approximations and numerical integration schemes, and particle filters (PFs), which use sequential Monte Carlo sampling. Despite their success, GSFs can suffer from numerical instabilities and severe failures in strongly nonlinear regimes, while PFs are flexible and robust but often demand substantial computational resources to achieve accurate estimates. In this work, we propose the Augmented Gaussian Sum Filter (AGSF), a novel filtering framework that unifies GSFs and PFs through an augmented Gaussian approximation parameterized by latent variables and tunable covariance parameters. By adjusting these covariances, the AGSF interpolates continuously between GSF-like and PF-like behavior, recovering both as special cases. Building on this view, we develop an adaptive AGSF that automatically shifts its behavior according to the local nature of the nonlinearities, acting more like a GSF when Gaussian approximations are reliable and more like a PF when they are not. In a target-tracking application, we demonstrate that AGSF is efficient and robust to common failure modes of both GSFs and PFs. We empirically validate the switching behavior of the adaptive mechanism in a toy example.

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 / 1 minor

Summary. The manuscript proposes the Augmented Gaussian Sum Filter (AGSF) for state-space models, unifying Gaussian sum filters (GSFs) and particle filters (PFs) via an augmented Gaussian approximation parameterized by latent variables and tunable covariance parameters. By adjusting the covariances, the filter is claimed to interpolate continuously between GSF-like and PF-like behavior, recovering both as special cases. An adaptive variant automatically shifts behavior according to local nonlinearities. The method is illustrated in a target-tracking application showing efficiency and robustness, with empirical validation of the switching mechanism in a toy example.

Significance. If the unification and adaptive mechanism can be shown to hold with stable interpolation and without introducing new instabilities, the work would provide a useful bridge between two established filtering families, potentially improving robustness in nonlinear regimes while controlling computational cost. The empirical demonstrations in target tracking and the toy example for adaptive switching offer initial evidence of practical value, though the absence of detailed error bounds or convergence analysis limits the assessed impact at present.

major comments (2)
  1. [Abstract / AGSF proposal paragraph] Abstract, paragraph on AGSF proposal: the central claim that adjusting tunable covariance parameters produces a continuous interpolation recovering GSF (finite local covariances) and PF (covariances driven toward zero) as exact special cases lacks any derivation showing moment preservation or absence of discontinuities under the nonlinear map. This assumption is load-bearing for both the unification and the adaptive switching mechanism.
  2. [Target-tracking application] Target-tracking application section: the reported robustness to common failure modes of GSFs and PFs is presented without quantitative error metrics, baseline comparisons, or analysis of approximation error in strongly nonlinear regimes, leaving the practical advantage of the interpolation unsubstantiated.
minor comments (1)
  1. [Toy example] The toy example description would benefit from explicit specification of the nonlinearity diagnostic used for adaptive switching and the exact parameter settings that recover the GSF and PF limits.

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, indicating where we agree that revisions will strengthen the presentation and where we provide additional clarification based on the existing material.

read point-by-point responses

  1. Referee: Abstract, paragraph on AGSF proposal: the central claim that adjusting tunable covariance parameters produces a continuous interpolation recovering GSF (finite local covariances) and PF (covariances driven toward zero) as exact special cases lacks any derivation showing moment preservation or absence of discontinuities under the nonlinear map. This assumption is load-bearing for both the unification and the adaptive switching mechanism.

    Authors: The AGSF construction defines the augmented Gaussian approximation such that the covariance parameters enter the representation linearly in the exponent of the Gaussian density. As these parameters vary continuously from finite positive values (recovering the local linearization and quadrature steps of a GSF) to values approaching zero (collapsing each component to a Dirac measure whose locations are propagated by the nonlinear dynamics, recovering the empirical measure of a PF), the resulting predictive and filtering distributions vary continuously in the weak topology. Moment preservation follows because the mean and covariance of each component are computed exactly from the previous step before the covariance scaling is applied, and the nonlinear map acts only on the mean locations. We acknowledge that an explicit lemma establishing continuity of the filter recursion under this parameterization would make the argument more self-contained. We will add this derivation, together with a short proof that the special cases are recovered exactly, as a new subsection in the revised manuscript. revision: yes

  2. Referee: Target-tracking application section: the reported robustness to common failure modes of GSFs and PFs is presented without quantitative error metrics, baseline comparisons, or analysis of approximation error in strongly nonlinear regimes, leaving the practical advantage of the interpolation unsubstantiated.

    Authors: The target-tracking experiments illustrate that the AGSF avoids both the divergence occasionally observed in GSFs under strong nonlinearity and the particle-depletion issues of PFs at modest sample sizes. While the manuscript already contains qualitative trajectory plots and a statement of computational cost, we agree that quantitative support is needed to substantiate the claimed advantage of the interpolation. In the revision we will add tables reporting RMSE (averaged over 100 Monte Carlo runs) for position and velocity estimates, direct comparisons against a standard GSF with the same number of components and a bootstrap PF with the same number of particles, and a brief analysis of the approximation error (measured by the Kullback-Leibler divergence to a high-fidelity reference filter) in the most nonlinear segments of the trajectory. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper proposes the AGSF as a new modeling framework whose unification property is achieved directly by introducing tunable covariance parameters in an augmented Gaussian approximation; the continuous interpolation between GSF-like and PF-like regimes is therefore a definitional feature of the construction rather than a derived prediction that reduces to prior inputs. No load-bearing self-citations, fitted parameters renamed as predictions, or uniqueness theorems imported from the authors' own prior work appear in the abstract or proposal description. The adaptive switching mechanism is presented as an additional design choice based on local nonlinearity diagnostics, and the target-tracking validation is empirical rather than tautological. The overall derivation chain remains self-contained as an original approximation technique.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on the existence of an augmented Gaussian representation whose covariance parameters can be adjusted to recover both limiting behaviors; the abstract does not enumerate explicit free parameters or new entities beyond the latent variables and tunable covariances mentioned.

free parameters (1)
  • tunable covariance parameters
    These are the adjustable parameters that control the interpolation between GSF-like and PF-like regimes; their specific values or fitting procedure are not detailed in the abstract.
axioms (1)
  • domain assumption State-space models admit a Markovian structure allowing sequential filtering updates
    Standard background assumption for all Bayesian filtering methods invoked implicitly when discussing SSMs and posterior approximations.
invented entities (1)
  • augmented Gaussian approximation with latent variables no independent evidence
    purpose: To parameterize a continuous family of approximations that includes both Gaussian sum and particle filter behaviors as special cases
    New representational device introduced in the abstract to achieve the unification; no independent evidence outside the paper is provided.

pith-pipeline@v0.9.0 · 5802 in / 1523 out tokens · 40375 ms · 2026-05-22T07:46:34.159795+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages

  1. [1]

    Hierarchical recurrent state space models reveal discrete and continuous dynamics of neural activity in c. elegans,

    S. Linderman, A. Nichols, D. Blei, M. Zimmer, and L. Paninski, “Hierarchical recurrent state space models reveal discrete and continuous dynamics of neural activity in c. elegans,”BioRxiv, p. 621540, 2019

    work page 2019

  2. [2]

    State-space models for the dynamics of wild animal populations,

    S. Buckland, K. Newman, L. Thomas, and N. Koesters, “State-space models for the dynamics of wild animal populations,”Ecological modelling, vol. 171, no. 1-2, pp. 157–175, 2004

    work page 2004

  3. [3]

    Aoki,State space modeling of time series, Springer Science & Business Media, 2013

    M. Aoki,State space modeling of time series, Springer Science & Business Media, 2013

    work page 2013

  4. [4]

    S ¨arkk¨a and L

    S. S ¨arkk¨a and L. Svensson,Bayesian filtering and smoothing, vol. 17, Cambridge university press, 2023

    work page 2023

  5. [5]

    A new approach to linear filtering and prediction problems,

    R. E. Kalman, “A new approach to linear filtering and prediction problems,” 1960

    work page 1960

  6. [6]

    New results in linear filtering and prediction theory,

    R. E. Kalman and R. S. Bucy, “New results in linear filtering and prediction theory,” 1961

    work page 1961

  7. [7]

    G. L. Smith, S. F. Schmidt, and L. A. McGee,Application of statistical filter theory to the optimal estimation of position and velocity on board a circumlunar vehicle, National Aeronautics and Space Administration, 1962

    work page 1962

  8. [8]

    The unscented kalman filter for nonlinear estimation,

    E. A. Wan and R. Van Der Merwe, “The unscented kalman filter for nonlinear estimation,” inProceedings of the IEEE 2000 Adaptive Sys- tems for Signal Processing, Communications, and Control Symposium (Cat. No. 00EX373). Ieee, 2000, pp. 153–158

    work page 2000

  9. [9]

    Discrete-time nonlinear filtering algorithms using gauss–hermite quadrature,

    I. Arasaratnam, S. Haykin, and R. J. Elliott, “Discrete-time nonlinear filtering algorithms using gauss–hermite quadrature,”Proceedings of the IEEE, vol. 95, no. 5, pp. 953–977, 2007

    work page 2007

  10. [10]

    Gaussian filters for nonlinear filtering problems,

    K. Ito and K. Xiong, “Gaussian filters for nonlinear filtering problems,” IEEE Transactions on Automatic Control, vol. 45, no. 5, pp. 910–927, 2000

    work page 2000

  11. [11]

    Cubature kalman filters,

    I. Arasaratnam and S. Haykin, “Cubature kalman filters,”IEEE Transactions on automatic control, vol. 54, no. 6, pp. 1254–1269, 2009

    work page 2009

  12. [12]

    Nonlinear bayesian estimation using gaussian sum approximations,

    D. Alspach and H. Sorenson, “Nonlinear bayesian estimation using gaussian sum approximations,”IEEE Transactions on Automatic Con- trol, vol. 17, no. 4, pp. 439–448, 1972

    work page 1972

  13. [13]

    Recursive bayesian estimation using gaussian sums,

    H. Sorenson and D. Alspach, “Recursive bayesian estimation using gaussian sums,”Automatica, vol. 7, no. 4, pp. 465–479, 1971

    work page 1971

  14. [14]

    Gaussian mixture nonlinear filtering with resampling for mixand narrowing,

    M. L. Psiaki, “Gaussian mixture nonlinear filtering with resampling for mixand narrowing,”IEEE Transactions on Signal Processing, vol. 64, no. 21, pp. 5499–5512, 2016

    work page 2016

  15. [15]

    Novel approach to nonlinear/non-gaussian bayesian state estimation,

    N. J. Gordon, D. J. Salmond, and A. F. Smith, “Novel approach to nonlinear/non-gaussian bayesian state estimation,” inIEE proceedings F (radar and signal processing). IET, 1993, vol. 140, pp. 107–113

    work page 1993

  16. [16]

    Doucet, N

    A. Doucet, N. De Freitas, N. J. Gordon, et al.,Sequential Monte Carlo methods in practice, vol. 1, Springer, 2001

    work page 2001

  17. [17]

    Chopin, O

    N. Chopin, O. Papaspiliopoulos, et al.,An introduction to sequential Monte Carlo, vol. 4, Springer, 2020

    work page 2020

  18. [18]

    Generalized Multiple Importance Sampling,

    V . Elvira, L. Martino, D. Luengo, and M. F. Bugallo, “Generalized Multiple Importance Sampling,”Statistical Science, vol. 34, no. 1, pp. 129 – 155, 2019

    work page 2019

  19. [19]

    Elvira and L

    V . Elvira and L. Martino,Advances in Importance Sampling, pp. 1–14, 02 2021

    work page 2021

  20. [20]

    A tutorial on particle filtering and smoothing: Fifteen years later,

    A. Doucet and A. M. Johansen, “A tutorial on particle filtering and smoothing: Fifteen years later,”Handbook of nonlinear filtering, vol. 12, no. 656-704, 2009

    work page 2009

  21. [21]

    A survey of convergence results on particle filtering methods for practitioners,

    D. Crisan and A. Doucet, “A survey of convergence results on particle filtering methods for practitioners,”IEEE Transactions on signal processing, vol. 50, no. 3, pp. 736–746, 2002

    work page 2002

  22. [22]

    Particle filters in robotics.,

    S. Thrun, “Particle filters in robotics.,” inUAI. Citeseer, 2002, vol. 2, pp. 511–518

    work page 2002

  23. [23]

    Particle filters and bayesian inference in financial econometrics,

    H. F. Lopes and R. S. Tsay, “Particle filters and bayesian inference in financial econometrics,”Journal of Forecasting, vol. 30, no. 1, pp. 168–209, 2011

    work page 2011

  24. [24]

    Particle filters for positioning, navigation, and tracking,

    F. Gustafsson, F. Gunnarsson, N. Bergman, U. Forssell, J. Jansson, R. Karlsson, and P.-J. Nordlund, “Particle filters for positioning, navigation, and tracking,”IEEE Transactions on signal processing, vol. 50, no. 2, pp. 425–437, 2002

    work page 2002

  25. [25]

    Curse of dimensionality and particle filters,

    F. Daum and J. Huang, “Curse of dimensionality and particle filters,” in2003 IEEE aerospace conference proceedings (Cat. No. 03TH8652). IEEE, 2003, vol. 4, pp. 4 1979–4 1993

    work page 2003

  26. [26]

    Nonlinear measurement update and prediction: Prior density splitting mixture estimator,

    A. Rauh, K. Briechle, and U. D. Hanebeck, “Nonlinear measurement update and prediction: Prior density splitting mixture estimator,” in2009 IEEE Control Applications,(CCA) & Intelligent Control,(ISIC). IEEE, 2009, pp. 1421–1426

    work page 2009

  27. [27]

    An adaptive derivative free method for bayesian posterior approximation,

    M. Raitoharju and S. Ali-Loytty, “An adaptive derivative free method for bayesian posterior approximation,”IEEE Signal Processing Letters, vol. 19, no. 2, pp. 87–90, 2012

    work page 2012

  28. [28]

    Mixture kalman filters,

    R. Chen and J. S. Liu, “Mixture kalman filters,”Journal of the Royal Statistical Society: Series B (Statistical Methodology), vol. 62, no. 3, pp. 493–508, 2000

    work page 2000

  29. [29]

    Gaussian sum particle filtering,

    J. H. Kotecha and P. M. Djuric, “Gaussian sum particle filtering,”IEEE Transactions on signal processing, vol. 51, no. 10, pp. 2602–2612, 2003

    work page 2003

  30. [30]

    Gaussian particle filtering,

    J. Kotecha and P. Djuric, “Gaussian particle filtering,”IEEE Transac- tions on Signal Processing, vol. 51, no. 10, pp. 2592–2601, 2003

    work page 2003

  31. [31]

    Entropy-based approach for uncertainty propagation of nonlinear dynamical systems,

    K. J. DeMars, R. H. Bishop, and M. K. Jah, “Entropy-based approach for uncertainty propagation of nonlinear dynamical systems,”Journal of Guidance, Control, and Dynamics, vol. 36, no. 4, pp. 1047–1057, 2013

    work page 2013

  32. [32]

    The split and merge unscented gaussian mixture filter,

    F. Faubel, J. McDonough, and D. Klakow, “The split and merge unscented gaussian mixture filter,”IEEE Signal Processing Letters, vol. 16, no. 9, pp. 786–789, 2009

    work page 2009

  33. [33]

    Discrete and continuous, probabilistic anticipation for autonomous robots in urban environments,

    F. Havlak and M. Campbell, “Discrete and continuous, probabilistic anticipation for autonomous robots in urban environments,”IEEE Transactions on Robotics, vol. 30, no. 2, pp. 461–474, 2013

    work page 2013

  34. [34]

    Automated splitting gaussian mixture nonlinear measurement update,

    K. Tuggle and R. Zanetti, “Automated splitting gaussian mixture nonlinear measurement update,”Journal of Guidance, Control, and Dynamics, vol. 41, no. 3, pp. 725–734, 2018

    work page 2018

  35. [35]

    Gaussian sum reapproximation for use in a nonlinear filter,

    M. L. Psiaki, J. R. Schoenberg, and I. T. Miller, “Gaussian sum reapproximation for use in a nonlinear filter,”Journal of Guidance, Control, and Dynamics, vol. 38, no. 2, pp. 292–303, 2015

    work page 2015

  36. [36]

    An augmented gaussian sum filter through a mixture decomposition,

    K. Tsampourakis and V . Elvira, “An augmented gaussian sum filter through a mixture decomposition,” inICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023, pp. 1–5

    work page 2023

  37. [37]

    Generalised particle filters with gaussian mixtures,

    D. Crisan and K. Li, “Generalised particle filters with gaussian mixtures,”Stochastic Processes and their Applications, vol. 125, no. 7, pp. 2643–2673, 2015

    work page 2015

  38. [38]

    A general method for approximating nonlinear transformations of probability distributions,

    S. Julier and J. K. Uhlmann, “A general method for approximating nonlinear transformations of probability distributions,” Tech. Rep., 1996

    work page 1996

  39. [39]

    The unscented kalman filter,

    E. A. Wan and R. Van Der Merwe, “The unscented kalman filter,” Kalman filtering and neural networks, pp. 221–280, 2001

    work page 2001

  40. [40]

    P. S. Maybeck,Stochastic models, estimation, and control, Academic press, 1982

    work page 1982

  41. [41]

    C. P. Robert and G. Casella,Monte Carlo statistical methods, vol. 2, Springer, 1999

    work page 1999

  42. [42]

    Monte carlo filter and smoother for non-gaussian nonlinear state space models,

    G. Kitagawa, “Monte carlo filter and smoother for non-gaussian nonlinear state space models,”Journal of computational and graphical statistics, vol. 5, no. 1, pp. 1–25, 1996

    work page 1996

  43. [43]

    Rethinking the effective sample size,

    V . Elvira, L. Martino, and C. P. Robert, “Rethinking the effective sample size,”International Statistical Review, vol. 90, no. 3, pp. 525–550, 2022

    work page 2022

  44. [44]

    Nonlinear programming,

    D. P. Bertsekas, “Nonlinear programming,”Journal of the Operational Research Society, vol. 48, no. 3, pp. 334–334, 1997

    work page 1997

  45. [45]

    Gaussian mixture reduction via clustering,

    D. Schieferdecker and M. F. Huber, “Gaussian mixture reduction via clustering,” in2009 12th international conference on information fusion. IEEE, 2009, pp. 1536–1543

    work page 2009

  46. [46]

    A look at gaus- sian mixture reduction algorithms,

    D. F. Crouse, P. Willett, K. Pattipati, and L. Svensson, “A look at gaus- sian mixture reduction algorithms,” in14th International Conference on Information Fusion. IEEE, 2011, pp. 1–8

    work page 2011

  47. [47]

    JAX: composable transformations of Python+NumPy programs,

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

    work page 2018

  48. [48]

    S. Chib, Y . Omori, and M. Asai,Multivariate Stochastic Volatility, pp. 365–400, Springer Berlin Heidelberg, Berlin, Heidelberg, 2009. 14 Supplementary Material APPENDIXB PROOF OFTHEOREM3 (i)For an overview of the BPF algorithm we refer the reader to Algorithm 11.9 of Ref. [4] in the text. We initialize the AGSF recursion with a particle approximation (i....

    work page 2009