Pith. sign in

REVIEW 4 major objections 5 minor 75 references

Guided filtering and smoothing for infinite-dimensional diffusions

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The smoothing and filtering distributions of an infinite-dimensional diffusion can be sampled exactly by weighting draws from a tractable guided process whose Radon-Nikodym derivative is explicit.

desk verdict Genuine extension of guided proposals to SPDEs with a real gap in stated regularity conditions and single-run numerics; both fixable, and the paper deserves peer review. read the letter →

arxiv 2507.06786 v2 pith:FEIKMFH2 submitted 2025-07-09 math.PR

Thorben Pieper-Sethmacher , Daniele Avitabile , Frank van der Meulen This is my paper · ORCID
classification math.PR MSC 60H1560G3562M2065C0565C40
keywords dataassimilationfilteringsmoothingguidedprocessDoobh-transforminfinite-dimensionaldiffusionstochasticneural-fieldequationsequentialMonteCarlo
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper establishes that conditioning an infinite-dimensional diffusion on discrete, finite-dimensional noisy observations can be handled by a tractable surrogate: the guided distribution. The smoothing law $L^h(X)$ and the guided law $L^g(X)$ are absolutely continuous, with Radon-Nikodym derivative $\Phi(X)=\frac{C_g}{C_h}\exp(\int_0^T \langle F(s,X_s),G(s,X_s)\rangle\,ds)$, where $G$ is a computable Gaussian score. Weighted samples from $L^g(X)$ therefore give Monte Carlo estimates of the filtering and smoothing distributions that are exact up to sampling error, and the changes of measure are defined on the Hilbert-space path space, so the construction stays well-defined as the spatial and temporal mesh is refined. The paper packages this into a guided particle filter for online filtering and into Metropolis-Hastings and Gibbs samplers for offline smoothing and Bayesian parameter estimation, and demonstrates the full pipeline on the stochastic neural-field equation.

What carries the argument

The load-bearing object is the guided distribution $L^g(X)$: the law of $X$ under the change of measure whose density process $E^g_t$ is built from the Gaussian surrogate $g$ defined through the backwards information filter. It replaces the intractable Doob $h$-transform of the true process with a tractable version based on an Ornstein-Uhlenbeck kernel $\nu$, so the score $G(t,x)=D_x\log g(t,x)$ has the affine form $V_t-U_tx$; $U$ and $V$ solve the backward Riccati and linear evolution equations (28)-(29), and $c$ solves (30), giving the full log-density. This $G$ simultaneously defines the guided SDE (23) used to generate proposals and appears inside the Radon-Nikodym weight $\Phi$, so the same precomputation drives both sampling and exact weighting. The construction is discretisation-free, and computing $G$ via the Riccati route has cost independent of the number of observations.

What would settle it

Run the guided particle filter on a semilinear SPDE with a superlinear drift (for example a cubic reaction term) under a fixed observation scheme, and compare the empirical mean of $\Phi(X^g)$ over many independent runs with $1$ while refining the time mesh. If $\mathbb{E}^g[\Phi(X^g)]$ deviates from $1$, or the effective sample size collapses exactly when the drift leaves the Lipschitz class, the imported martingale condition fails and the weight recursion is biased; the paper's identity predicts $\mathbb{E}^g[\Phi(X^g)]=1$ whenever the construction is valid.

Watch

Extended reading notes

Core claim

The paper's central claim is that filtering and smoothing for mild solutions of semilinear SPDEs observed through finite-rank linear measurements can be solved by importance sampling from a guided process. For observations $Y_i \mid X_{t_i}\sim \mathcal{N}(LX_{t_i},\Sigma)$, the authors define $h$ through the backwards information filter using the intractable transition kernel $\mu$, and define $g$ by the same recursion with the transition kernel $\nu$ of an Ornstein-Uhlenbeck process; $g$ is a Gaussian density with explicit score $G(t,x)=D_x\log g(t,x)$. Theorem 2.7 identifies the law of $X$ under a martingale change of measure $P^g$ as the unique mild solution of $dX_t^g=[AX_t^g+F(t,X_t^g)+QG(t,X_t^g)]\,dt+Q^{1/2}\,dW_t^g$, and the ratio of the two changes of measure collapses to the explicit weight $\Phi(X)$ above. Hence unweighted draws from the guided process, weighted by $\Phi$, are Monte Carlo samples from the smoothing law; restricting $g$ to a one-step-ahead window gives the particle-filter proposal. The same identity supports a pCN-accelerated Metropolis-Hastings sampler on Wiener space and a Gibbs sampler for an unknown initial state and model parameters, with non-Gaussian observations handled by an extra likelihood-ratio factor.

Load-bearing premise

The construction relies on the two exponential martingales that define the smoothing and guided measures being genuine martingales and on the guided SDE having a unique mild solution; these conditions are imported from earlier results rather than restated as assumptions on the drift and noise, so for superlinear drifts outside the Lipschitz class the guided measure may fail to be a probability measure, breaking the weight recursion.

Editorial extensions

If this is right

  • Filtering and smoothing posteriors of an infinite-dimensional diffusion can be estimated exactly up to Monte Carlo error by importance-weighting samples from the guided process, because the Radon-Nikodym derivative between the smoothing and guided laws is the explicit functional $\Phi(X)$.
  • The guided particle filter uses a likelihood-informed proposal: particles evolve under the guided SDE, and the weight update (33) only needs the surrogate likelihood $g_i$ and the integral of $\langle F,G\rangle$, so the next observation steers particles before resampling.
  • The smoothing and parameter-estimation samplers target the full joint posterior of the path and unknown parameters using guided proposals localised by pCN in Wiener space; when only the nonlinearity depends on the parameter, the backward Riccati equations are solved once.
  • The framework is discretisation-free: the changes of measure are defined on the Hilbert-space path space, so filtering and smoothing remain well-defined as spatial and temporal meshes are refined, unlike methods that discretise the SPDE before deriving a proposal.
  • Non-Gaussian observation densities are accommodated by replacing $k$ with $l_i$ in the smoothing measure and adjusting the weight by the product $\prod_i l_i/k_i$, preserving the exact weighted-sampling structure.

Reading between the lines

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

  • Editorial inference: the same construction should extend to other semilinear SPDEs whose drift splits as $AX+F(X)$ with an Ornstein-Uhlenbeck auxiliary process, making the guided distribution a general template for data assimilation beyond the neural-field case study.
  • Editorial inference: because the guiding score $G$ is computed from all future observations, the smoothing sampler can be read as a principled, exact alternative to nudging, which typically uses only the next observation and supplies no Radon-Nikodym weight.
  • Editorial inference: a testable extension would estimate the observation operator $L$ or the noise covariance $\Sigma$ alongside the state, since the backward recursions depend on them only through terminal conditions.
  • Editorial inference: the closed-form scalar Riccati solution in the diagonalisable case suggests spectral discretisations can reduce the cost of computing $G$ below the general $O(NM^3)$ estimate, enabling higher-resolution neural-field inversions.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper develops a guided-measure framework for filtering, smoothing, and parameter estimation of infinite-dimensional diffusions that are mild solutions of semilinear SPDEs and are observed through finite-dimensional Gaussian functionals at discrete times. It constructs an intractable smoothing law L^h(X) via an h-transformed measure and a tractable guided law L^g(X) based on an auxiliary Ornstein-Uhlenbeck process, and claims that the two laws are absolutely continuous with explicit Radon-Nikodym derivative Phi(X) = (C_g/C_h) exp( integral_0^T <F(s,X_s), G(s,X_s)> ds ), Eqs. (24)-(25). The paper derives closed-form Gaussian representations for g, alternative Riccati equations for the score G, and uses these in a sequential Monte Carlo filter and in Metropolis-Hastings/Gibbs samplers for smoothing and parameter estimation. The methodology is illustrated on the stochastic Amari equation, with code provided.

Significance. If the central theorems hold, this is a valuable contribution: it extends guided-process methodology from finite-dimensional diffusions to Hilbert-space-valued SPDEs, provides explicit and computable guidance terms via Riccati equations, and treats filtering, smoothing, and parameter estimation in a unified framework that is formulated before spatial-temporal discretisation. The paper is transparent about the intractability of the true filter, ships source code, and gives closed-form formulas (Theorem 2.6, Theorem 2.9, Proposition 2.12) that practitioners can implement. However, the exactness claims rest on imported regularity conditions that are not stated as assumptions, and the numerical evidence is single-run and mixed in the dense-observation regime. The significance is therefore conditional on closing the scope gap and correcting the case-study formulas.

major comments (4)
  1. [Section 5.1, Proposition 5.1] Proposition 5.1 contains a sign error in the semigroup factors. For A = -I on L^2(D), the semigroup is S_t = e^{-t} I, so S_{t_i - t} = e^{-(t_i - t)} I and the one-step guided score should be G_i(t,x) = e^{-(t_i - t)} L^* (Sigma + (1 - e^{-2(t_i - t)})/2 L Q L^*)^{-1} (y_i - e^{-(t_i - t)} L x). The printed formula instead uses exp(-(t - t_i)) = e^{t_i - t} and (1 - exp(-2(t - t_i)))/2 = (1 - e^{2(t_i - t)})/2, which for t < t_i are respectively the reciprocal and the negative of the correct quantities. This contradicts the convention Delta_i := t_i - t introduced in Appendix B and would make the proposal (32) steer away from the observation with a negative covariance matrix. The numerical results for GPF-I in Figures 3-6 should be re-examined after this correction.
  2. [Section 2.2 / Appendix A.2, Theorem 2.7] The standing assumptions in Section 1.1 only require F to be continuous and Equation (1) to have a unique mild solution. The proof of Theorem 2.7 in Appendix A.2 invokes [47, Theorem 4.1] for the generator identity (55) and [53, Lemma C.1] for the true-martingale property of E(M^g), and it asserts uniqueness of the mild solution to the guided equation (23). The hypotheses of these imported results are not restated, and they are not implied by the standing assumptions: the drift in (23) is F(t,x) + QG(t,x) with G affine in x, so uniqueness requires more than continuity of F, and the generator computation (55) needs regularity of F beyond what is assumed. Since Theorem 2.7 is the basis for the exact importance weights (24)-(25), the class of SPDEs for which the method is exact is left undelimited. Please state explicit sufficient conditions on F, A, Q, and the observation scheme, and verify them for the Amari case study.
  3. [Section 2.3, Theorems 2.9 and Proposition 2.12] Theorem 2.9 asserts existence and uniqueness of mild solutions U and V to the backwards Riccati and evolution equations (28)-(29), and Proposition 2.12 uses the trace tr(U_t Q). These statements require explicit operator assumptions on A, Q, and the domains of U_t; for arbitrary unbounded A and trace-class Q the infinite-dimensional Riccati equation is not automatically well-posed. The cited references [10,6] contain conditions, but they are not stated in the paper. This matters because the algorithms (Algorithm 3-5) rely on these objects being well defined independently of the discretisation, and the paper's 'discretisation-free' claim depends on the continuous-level formulation being rigorous.
  4. [Section 5.2, Figures 3-6] The filtering comparison is based on a single simulated dataset for each of the two parameter regimes, with fixed hyperparameters (J = 100, N = 30, beta = 0.1, alpha = 0.75) and no Monte Carlo repetitions or standard errors. In the dense-observation version of Experiment 1, the UKF outperforms both guided particle filters and GPF-II outperforms GPF-I; the favourable conclusions for GPF-I rest on the sparse-observation settings and on Experiment 2. This is acceptable as an illustration, but it is not sufficient to support comparative statements such as 'the proposed GPF-I outperforms GPF-II over all three observation schemes' in Experiment 2. Please add repeated simulations with error bars or temper the comparative claims.
minor comments (5)
  1. [Figure 6 caption] The caption of Figure 6 says 'based on the dataset in 1'; this should refer to Figure 2, since the displayed experiment is Experiment 2 with delta = 0.5.
  2. [Algorithm 5, step 3] In Algorithm 5, step 3 contains two sub-items labelled '(iii)'; the second should be '(iv)'.
  3. [Section 2.3, Eq. (28)] The sentence introducing Theorem 2.9 contains a stray '1' before '(t_{i-1}, t_i]'.
  4. [Section 4.2, Assumption 4.3] Assumption 4.3 should specify that the Gaussian measure nu_0 is centered: the pCN update for X_0 in Algorithm 4 (X_0' = sqrt(1-beta_0^2) X_0 + beta_0 z with z ~ nu_0) is reversible with respect to nu_0 only when nu_0 has zero mean. If nu_0 has nonzero mean, the update should be centered around that mean.
  5. [Section 5.2, relative errors] The relative error |Xhat_ti - X_ti| / |X_ti| can be unstable when |X_ti| is small; a short remark on this choice, or the use of an absolute error alongside it, would help interpretation of Figures 3 and 5.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the RN-derivative identity (24)-(25) follows algebraically from the defining densities, and the nontrivial martingale step is imported from cited prior results rather than assumed; minor self-citation noted but not counted as circular.

full rationale

Walking the derivation chain, the smoothing density h in (10) and the guided density g in (16) are defined independently from the true transition kernel and the OU transition kernel, and Theorem 2.6 obtains the closed form of g by Gaussian convolution; the Riccati and backwards-ODE forms in Theorems 2.9 and Proposition 2.12 are mild-solution identities, not disguised assumptions. The central absolute-continuity statement (24)-(25) is direct algebra from the defining densities E^h_T in (11) and E^g_T in (22): after substituting the terminal equalities h(T,·)=k(·,y_n)=g(T,·), the product of observation kernels cancels and leaves C_g/C_h times the exponential of the integral of <F,G>. This is a constructional identity, not a fitted or predicted quantity. The nontrivial content is the martingale property of E^g, whose proof in Appendix A.2 invokes [47, Theorem 4.1] for the generator action on g and [53, Lemma C.1] for the Doleans-Dade exponential; these are cited published mathematical results with stated hypotheses, and the required Lipschitz property of G is verified in the text, so they constitute legitimate support rather than a circular reduction. Remark 2.5 even observes that the hard-to-verify differentiability condition in Theorem 2.4(ii) is not needed for sampling. The numerical section benchmarks against the UKF and the [44] guided particle filter and checks posterior estimates against the known parameters used to generate the data, so no fitted quantity is relabelled as a prediction. The main substantive caveat is an unstated regularity scope: the true-martingale property of E^g and the uniqueness of a mild solution to (23) are imported without restating all hypotheses of [53, Lemma C.1] and [47, Theorem 4.1], and these are not implied by the standing assumptions of continuity of F and unique mild solution of (1). That is a completeness or rigor gap, not circularity. I therefore find no circular step; score 2 reflects only the minor self-citation note, not a circularity finding.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central construction rests on standard SPDE well-posedness, the authors' earlier exponential change-of-measure machinery ([53]), and operator Riccati well-posedness. The numerical case study adds hand-chosen hyperparameters. No fitted parameters are used in the derivation of the method, and no new physical entities are postulated.

free parameters (4)
  • pCN step size beta = 0.1 in all experiments
    Hand-chosen step size for preconditioned Crank-Nicolson proposals in Algorithms 2, 3, 4, and 5; affects mixing but not the validity of the guiding construction.
  • ESS threshold alpha = 0.75
    Tempering parameter controlling the schedule of intermediate distributions in Algorithm 2; a user choice.
  • Particle count J and MCMC moves N = J=100, N=30
    Algorithmic hyperparameters in Section 5.2.
  • Initial adaptive Metropolis step sizes S0 = S0=1 for all parameters
    Adaptive scaling Metropolis proposals in Section 5.3; the scaling function r_j=j^(-2/3) and target acceptance alpha*=0.234 are also chosen by hand.
assumptions (5)
  • domain assumption Equation (1) admits a unique mild solution X with semigroup S_t and transition kernel mu; F is continuous.
    Stated in Section 1.1; the entire construction presumes this well-posedness.
  • standard math The exponential change-of-measure results of [53] (Lemma 3.3, Lemma C.1) and the Ito formula of [47] (Theorem 4.1) hold for h and g.
    Invoked in proofs of Theorems 2.4 and 2.7 (Appendix A). These are prior published results by the same group for the h-transform of SPDEs; correctness of the current framework hinges on them.
  • standard math The backwards Riccati equation (28) has a unique mild solution in the space of bounded operators; the linear equation (29) has a unique mild solution.
    Used in Theorem 2.9; uniqueness is cited to [10, Theorem 3.6] and [6, Chapter 2.2]. Needed for the efficient computation of G.
  • domain assumption For unknown X0, the prior mu_0 is dominated by a Gaussian measure nu_0 with density rho (Assumption 4.3).
    Required for the Gibbs sampler in Algorithm 4 and for the joint parameter posterior in Algorithm 5; not satisfied by Dirac priors, which are handled separately in Section 4.1.
  • ad hoc to paper Frechet differentiability of h in x (Theorem 2.4(ii)), with D_x h in C^m((t_{i-1},t_i);H).
    The paper itself notes in Remark 2.5 that this is hard to verify and is not needed for sampling under P^g; the differential form of X^h is therefore a conditional result.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Guided filtering and smoothing for infinite-dimensional diffusions." pith.science (2026). https://pith.science/paper/FEIKMFH2

@misc{pith2026250706786,
  author       = {Pith},
  title        = {Pith review of: Guided filtering and smoothing for infinite-dimensional diffusions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FEIKMFH2}},
  note         = {Machine review of arXiv:2507.06786}
}
read the original abstract

We consider the filtering and smoothing problems for an infinite-dimensional diffusion process X, observed through a finite-dimensional representation at discrete points in time. At the heart of our proposed methodology lies the construction of a path measure, termed the guided distribution of X, that is absolutely continuous with respect to the law of X, conditioned on the observations. We show that this distribution can be incorporated as a potent proposal measure for both sequential Monte Carlo as well as Markov Chain Monte Carlo schemes to tackle the filtering and smoothing problems respectively. In the offline setting, we extend our approach to incorporate parameter estimation of unknown model parameters. The proposed methodology is numerically illustrated in a case study for the stochastic Amari equation.

Figures

Figures reproduced from arXiv: 2507.06786 by the authors.

Figure 1
Figure 1. Data generating process of Experiment 1. Colour intensity represents the value of the process and observations, whereas time t is plotted along the y-axis and spatial coordinate ξ along the x-axis. Left: Heatmap of the sample path X(t, ξ) of Eq. (42) with parameters as in (47) and δ0 = 0. Right: Observations Yi at n = 20 observation times with m = 15 local averages of X(ti , ξ) per observation [PITH_FULL_IMAGE:figu… view at source ↗
Figure 2
Figure 2. Data generating process of Experiment 2. Colour intensity represents the value of the process and observations, whereas time t is plotted along the y-axis and spatial coordinate ξ along the x-axis. Left: Heatmap of the sample path X(t, ξ) of Eq. (42) with parameters as in (47) and δ0 = 0.5. Right: Observations Yi at n = 20 observation times with m = 15 local averages of X(ti , ξ) per observation [PITH_FULL_IMAGE:fi… view at source ↗
Figure 3
Figure 3. Relative errors |Xˆ ti − Xti |/|Xti | at observation times ti for the filtering estimates Xˆ ti based on the data set of [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Filtering estimates for the two guided particle filter methods based on the dataset in 1. Left column: Heatmaps of the filtered path estimate. Right column: Heatmaps of the absolute estimation error. The first and third rows show GPF-I (our proposed method), while the …
Figure 5
Figure 5. Figure 5: Relative errors |Xˆ ti − Xti |/|Xti | at observation times ti for the filtering estimates Xˆ ti based on the data set of [PITH_FULL_IMAGE:figures/full_fig_p021_5.png]
Figure 6
Figure 6. Figure 6: Filtering estimates for the two guided particle filter methods based on the dataset in 1. Left column: Heatmaps of the filtered path estimate. Right column: Heatmaps of the absolute estimation error. The first and third rows show GPF-I (our proposed method), while the …
Figure 7
Figure 7. Figure 7: Trace-plots of the MCMC outputs returned by Algorithm 5 for various model parameters. The rows show, in order, the trace-plots of η, ζ, A and δ. The dark blue solid line represents the sample mean after a burn-in period of 5000 samples, whereas the dashed lines represe…
Figure 8
Figure 8. Figure 8: Heatmap of the smoothing estimate of X. Left: Path estimate Xˆ given by the mean of the final 1000 MCMC iterations returned by Algorithm 5. Right: Absolute error with respect to the true signal X. Indeed, from plugging in (10) and using the Chapman-Kolmogorov equation,…
Figure 9
Figure 9. Figure 9: Spatially localised sample paths of the smoothing distribution of X. Spatial localisations are computed by applying the observation operator L. Green represents early samples in the Markov chain, whereas blue represents later samples. The orange line shows the true loc…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

75 extracted references · 64 canonical work pages

  1. [1]

    Kernel Recon- struction for Delayed Neural Field Equations.The Journal of Mathematical Neuroscience, 8(1): 3, December 2018

    Jehan Alswaihli, Roland Potthast, Ingo Bojak, Douglas Saddy, and Axel Hutt. Kernel Recon- struction for Delayed Neural Field Equations.The Journal of Mathematical Neuroscience, 8(1): 3, December 2018. ISSN 2190-8567. doi: 10.1186/s13408-018-0058-8

    work page doi:10.1186/s13408-018-0058-8 2018

  2. [2]

    Parameter estimation for semilinear spdes from local measurements.Bernoulli, 29(3):2035–2061, 2023

    Randolf Altmeyer, Igor Cialenco, and Gregor Pasemann. Parameter estimation for semilinear spdes from local measurements.Bernoulli, 29(3):2035–2061, 2023

    work page 2023

  3. [3]

    Dynamics of pattern formation in lateral-inhibition type neural fields.Biological Cybernetics, 27(2):77–87, 1977

    Shun-ichi Amari. Dynamics of pattern formation in lateral-inhibition type neural fields.Biological Cybernetics, 27(2):77–87, 1977

    work page 1977

  4. [4]

    A tutorial on adaptive mcmc.Statistics and comput- ing, 18:343–373, 2008

    Christophe Andrieu and Johannes Thoms. A tutorial on adaptive mcmc.Statistics and comput- ing, 18:343–373, 2008

    work page 2008

  5. [5]

    Bain and D

    A. Bain and D. Crisan.Fundamentals of Stochastic Filtering. Stochastic Modelling and Applied Probability. Springer New York, 2009. ISBN 9780387768960. URLhttps://books.google.de/ books?id=hE3KF5Wf6ecC

    work page 2009

  6. [6]

    Springer, 2007

    Alain Bensoussan, Giuseppe Da Prato, Michel C Delfour, and Sanjoy K Mitter.Representation and control of infinite dimensional systems, volume 1. Springer, 2007. GUIDED FILTERING AND SMOOTHING FOR INFINITE-DIMENSIONAL DIFFUSIONS 33

    work page 2007

  7. [7]

    Mcmc methods for diffu- sion bridges.Stochastics and Dynamics, 08(03):319–350, 2008

    Alexandros Beskos, Gareth Roberts, Andrew Stuart, and Jochen Voss. Mcmc methods for diffu- sion bridges.Stochastics and Dynamics, 08(03):319–350, 2008. doi: 10.1142/S0219493708002378. URLhttps://doi.org/10.1142/S0219493708002378

    work page doi:10.1142/s0219493708002378 2008

  8. [8]

    Mat´ ern gaussian processes on riemannian manifolds.Advances in Neural Information Processing Systems, 33:12426–12437, 2020

    Viacheslav Borovitskiy, Alexander Terenin, Peter Mostowsky, et al. Mat´ ern gaussian processes on riemannian manifolds.Advances in Neural Information Processing Systems, 33:12426–12437, 2020

    work page 2020

Show all 75 references
  1. [9]

    Bressloff.Waves in Neural Media

    Paul C. Bressloff.Waves in Neural Media. Lecture Notes on Mathematical Modelling in the Life Sciences. Springer New York, New York, NY, 2014. ISBN 978-1-4614-8865-1 978-1-4614-8866-8. doi: 10.1007/978-1-4614-8866-8

    2014 doi

  2. [10]

    Solutions and approximations to the riccati integral equation with values in a space of compact operators.SIAM Journal on Control and Optimiza- tion, 53(5):2846–2877, 2015

    John A Burns and Carlos N Rautenberg. Solutions and approximations to the riccati integral equation with values in a space of compact operators.SIAM Journal on Control and Optimiza- tion, 53(5):2846–2877, 2015

    2015

  3. [11]

    On the approximation of operator-valued riccati equations in hilbert spaces

    James Cheung. On the approximation of operator-valued riccati equations in hilbert spaces. Journal of Mathematical Analysis and Applications, 547(1):129250, 2025

    2025

  4. [12]

    Springer, 2020

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

    2020

  5. [13]

    Statistical inference for spdes: an overview.Statistical Inference for Stochastic Processes, 21(2):309–329, 2018

    Igor Cialenco. Statistical inference for spdes: an overview.Statistical Inference for Stochastic Processes, 21(2):309–329, 2018. doi: 10.1007/s11203-018-9175-4

    2018 doi

  6. [14]

    Parameter estimation for the stochastically perturbed navier–stokes equations.Stochastic Processes and their Applications, 121:701–724, 2011

    Igor Cialenco and Nathan Glatt-Holtz. Parameter estimation for the stochastically perturbed navier–stokes equations.Stochastic Processes and their Applications, 121:701–724, 2011

    2011

  7. [15]

    Igor Cialenco, Hyun-Jung Kim, and Gregor Pasemann. Statistical analysis of discretely sampled semilinear spdes: a power variation approach.Stochastics and Partial Differential Equations: Analysis and Computations, 12(1):326–351, 2024

    2024

  8. [16]

    J. M. C. Clark. The simulation of pinned diffusions. InDecision and Control, 1990., Proceedings of the 29th IEEE Conference on, pages 1418–1420. IEEE, 1990

    1990

  9. [17]

    On the physical nudging equations.Climate Dynamics, 58(5):1459–1476, 2022

    Giovanni Conti, Ali Aydo˘ gdu, Silvio Gualdi, Antonio Navarra, and Joe Tribbia. On the physical nudging equations.Climate Dynamics, 58(5):1459–1476, 2022

    2022

  10. [18]

    Model and observation-error covariance matrix information in the physical nudging equations.Quarterly Journal of the Royal Meteorological Society, page e4979, 2025

    Giovanni Conti, Peter Jan van Leeuwen, and Jeffrey Anderson. Model and observation-error covariance matrix information in the physical nudging equations.Quarterly Journal of the Royal Meteorological Society, page e4979, 2025

    2025

  11. [19]

    Stephen Coombes and Kyle C. A. Wedgwood.Neurodynamics: An Applied Mathematics Per- spective, volume 75 ofTexts in Applied Mathematics. Springer International Publishing, Cham,

  12. [20]

    Cotter, Dan Crisan, and Maneesh Kumar Singh

    Colin J. Cotter, Dan Crisan, and Maneesh Kumar Singh. Data assimilation for the stochastic camassa-holm equation using particle filtering: A numerical investigation. In Bertrand Chapron, Dan Crisan, Darryl D. Holm, Etienne M´ emin, and Jane-Lisa Coughlan, editors,Stochastic Tr...

    2025

  13. [21]

    S. L. Cotter, G. O. Roberts, A. M. Stuart, and D. White. MCMC Methods for Functions: Modifying Old Algorithms to Make Them Faster.Statistical Science, 28(3):424 – 446, 2013. doi: 10.1214/13-STS421. URLhttps://doi.org/10.1214/13-STS421

    2013 doi

  14. [22]

    Gunzburger, and Christian Vollmann.Nonlocal Integral Equation Contin- uum Models: Nonstandard Symmetric Interaction Neighborhoods and Finite Element Discretiza- tions

    Marta D’Elia, Max D. Gunzburger, and Christian Vollmann.Nonlocal Integral Equation Contin- uum Models: Nonstandard Symmetric Interaction Neighborhoods and Finite Element Discretiza- tions. Number 31 in Computational Science and Engineering. Society for Industrial and Applied M...

    2024

  15. [23]

    Simulation of conditioned diffusion and application to parameter estimation.Stochastic Processes and their Applications, 116(11):1660 – 1675, 2006

    Bernard Delyon and Ying Hu. Simulation of conditioned diffusion and application to parameter estimation.Stochastic Processes and their Applications, 116(11):1660 – 1675, 2006. ISSN 0304-

    2006

  16. [24]

    Springer, 2001

    Arnaud Doucet, Nando De Freitas, Neil James Gordon, et al.Sequential Monte Carlo methods in practice, volume 1. Springer, 2001

    2001

  17. [25]

    Number 94 in CBMS-NSF Regional Conference Series in Applied Mathematics

    Qiang Du.Nonlocal Modeling, Analysis, and Computation. Number 94 in CBMS-NSF Regional Conference Series in Applied Mathematics. Society for Industrial and Applied Mathematics, Philadelphia, 2019. ISBN 978-1-61197-561-1

    2019

  18. [26]

    Bard Ermentrout and David H

    G. Bard Ermentrout and David H. Terman.Mathematical Foundations of Neuroscience, vol- ume 35 ofInterdisciplinary Applied Mathematics. Springer New York, New York, NY, 2010. ISBN 978-0-387-87707-5 978-0-387-87708-2. doi: 10.1007/978-0-387-87708-2. 34 GUIDED FILTERING AND SMOOTH...

    2010 doi

  19. [27]

    Faugeras and J

    O. Faugeras and J. Inglis. Stochastic neural field equations: A rigorous footing.Journal of Mathematical Biology, 71(2):259–300, August 2015. ISSN 0303-6812, 1432-1416. doi: 10.1007/ s00285-014-0807-6

    2015

  20. [28]

    Non-parametric estimation of the reaction term in semi-linear spdes with spatial ergodicity.arXiv preprint arXiv:2307.05457, 2023

    Sascha Gaudlitz. Non-parametric estimation of the reaction term in semi-linear spdes with spatial ergodicity.arXiv preprint arXiv:2307.05457, 2023

    2023 arXiv

  21. [29]

    Estimation for the reaction term in semi-linear spdes under small diffusivity.Bernoulli, 29(4):3033–3058, 2023

    Sascha Gaudlitz and Markus Reiß. Estimation for the reaction term in semi-linear spdes under small diffusivity.Bernoulli, 29(4):3033–3058, 2023

    2023

  22. [30]

    J. S. Gibson. Linear-quadratic optimal control of hereditary differential systems: Infinite dimen- sional riccati equations and numerical approximations.SIAM Journal on Control and Optimiza- tion, 21(1):95–139, 1983. doi: 10.1137/0321006

    1983 doi

  23. [31]

    An introduction to stochastic pdes.arXiv preprint arXiv:0907.4178, 2009

    Martin Hairer. An introduction to stochastic pdes.arXiv preprint arXiv:0907.4178, 2009

    2009 arXiv

  24. [32]

    Nonparametric calibration for stochastic reaction– diffusion equations based on discrete observations.Stochastic Processes and their Applications, 162:171–217, 2023

    Florian Hildebrandt and Mathias Trabs. Nonparametric calibration for stochastic reaction– diffusion equations based on discrete observations.Stochastic Processes and their Applications, 162:171–217, 2023

    2023

  25. [33]

    J. E. Hoke and R. A. Anthes. The initialization of numerical models by a dynamic-initialization technique.Monthly Weather Review, 104(12):1551–1556, 1976

    1976

  26. [34]

    Izhikevich.Dynamical Systems in Neuroscience: The Geometry of Excitability and Bursting

    Eugene M. Izhikevich.Dynamical Systems in Neuroscience: The Geometry of Excitability and Bursting. The MIT Press, 2006. ISBN 978-0-262-27607-8. doi: 10.7551/mitpress/2526.001.0001

    2006 doi

  27. [35]

    Wandering bumps in stochastic neural fields.SIAM Journal on Applied Dynamical Systems, 12(1):61–94, 2013

    Zachary P Kilpatrick and Bard Ermentrout. Wandering bumps in stochastic neural fields.SIAM Journal on Applied Dynamical Systems, 12(1):61–94, 2013

    2013

  28. [36]

    Kilpatrick and Gr´ egory Faye

    Zachary P. Kilpatrick and Gr´ egory Faye. Pulse Bifurcations in Stochastic Neural Fields.SIAM Journal on Applied Dynamical Systems, 13(2):830–860, January 2014. ISSN 1536-0040. doi: 10.1137/140951369

    2014 doi

  29. [37]

    Large Deviations for Nonlocal Stochastic Neural Fields.The Journal of Mathematical Neuroscience, 4(1):1, 2014

    Christian Kuehn and Martin G Riedler. Large Deviations for Nonlocal Stochastic Neural Fields.The Journal of Mathematical Neuroscience, 4(1):1, 2014. ISSN 2190-8567. doi: 10.1186/2190-8567-4-1

    2014 doi

  30. [38]

    M. V. Kulikova and G. Yu. Kulikov. Data-driven parameter estimation in stochastic dynamic neural fields by state-space approach and continuous-discrete extended Kalman filtering.Digital Signal Processing, page 104010, March 2023. ISSN 1051-2004. doi: 10.1016/j.dsp.2023.104010

    2023

  31. [39]

    Bayesian inference for fluid dynamics: a case study for the stochastic rotating shallow water model.Frontiers in Applied Mathematics and Statistics, 8:949354, 2022

    Oana Lang, Peter Jan Van Leeuwen, Dan Crisan, and Roland Potthast. Bayesian inference for fluid dynamics: a case study for the stochastic rotating shallow water model.Frontiers in Applied Mathematics and Statistics, 8:949354, 2022

    2022

  32. [40]

    Springer

    Kody Law, Andrew Stuart, and Konstantinos Zygalakis.Data Assimilation. Springer. Springer, September 2015. ISBN 3-319-20325-8

    2015

  33. [41]

    Pedro M. Lima. Numerical Investigation of Stochastic Neural Field Equations. In Vinai K. Singh, David Gao, and Andreas Fischer, editors,Advances in Mathematical Methods and High Performance Computing, volume 41, pages 51–67. Springer International Publishing, Cham,

  34. [42]

    P.M. Lima, W. Erlhagen, M.V. Kulikova, and G.Yu. Kulikov. Numerical solution of the stochastic neural field equation with applications to working memory.Physica A: Statistical Mechanics and its Applications, 596:127166, June 2022. ISSN 03784371. doi: 10.1016/j.physa.2022.127166

    2022

  35. [43]

    Finn Lindgren, H ˚ avard Rue, and Johan Lindstr¨ om. An explicit link between gaussian fields and gaussian markov random fields: the stochastic partial differential equation approach.Journal of the Royal Statistical Society Series B: Statistical Methodology, 73(4):423–498, 2011

    2011

  36. [44]

    Particle filtering for stochastic navier–stokes signal observed with linear additive noise.SIAM Journal on Scientific Computing, 40(3):A1544–A1565, 2018

    Francesc Pons Llopis, Nikolas Kantas, Alexandros Beskos, and Ajay Jasra. Particle filtering for stochastic navier–stokes signal observed with linear additive noise.SIAM Journal on Scientific Computing, 40(3):A1544–A1565, 2018

    2018

  37. [45]

    Luenberger

    D.G. Luenberger. Observing the state of a linear system.IEEE Transactions on Military Elec- tronics, 8(2):74–80, 1964

    1964

  38. [46]

    MacLaurin and Paul C

    James N. MacLaurin and Paul C. Bressloff. Wandering bumps in a stochastic neural field: A variational approach.Physica D: Nonlinear Phenomena, 406:132403, 2020. ISSN 0167-2789. doi: 10.1016/j.physd.2020.132403

    2020

  39. [47]

    Fokker–planck equation for kolmogorov operators with unbounded coefficients

    Luigi Manca. Fokker–planck equation for kolmogorov operators with unbounded coefficients. Stochastic analysis and applications, 27(4):747–769, 2009

    2009

  40. [48]

    Continuous-discrete smoothing of diffusions.Electronic Journal of Statistics, 15(2):4295 – 4342, 2021

    Marcin Mider, Moritz Schauer, and Frank van der Meulen. Continuous-discrete smoothing of diffusions.Electronic Journal of Statistics, 15(2):4295 – 4342, 2021. doi: 10.1214/21-EJS1894. GUIDED FILTERING AND SMOOTHING FOR INFINITE-DIMENSIONAL DIFFUSIONS 35

    2021 doi

  41. [49]

    IOP Expanding Physics

    Gen Nakamura and Roland Potthast.Inverse Modeling: An Introduction to the Theory and Methods of Inverse Problems and Data Assimilation. IOP Expanding Physics. IOP Publishing, Bristol, UK, 2015. ISBN 978-0-7503-1218-9 978-0-7503-1219-6. doi: 10.1088/978-0-7503-1218-9

    2015 doi

  42. [50]

    Regression and Classification Using Gaussian Process Priors

    Radford M Neal. Regression and Classification Using Gaussian Process Priors. InBayesian Statistics 6: Proceedings of the Sixth Valencia International Meeting June 6-10, 1998. Oxford University Press, 08 1999. ISBN 9780198504856. doi: 10.1093/oso/9780198504856.003.0021. URLhttp...

    1998

  43. [51]

    Omiros Papaspiliopoulos and Gareth O. Roberts. Importance sampling techniques for estima- tion of diffusion models. In Mathieu Kessler, Alexander Linder, and Michael Sørensen, editors, Statistical Methods for Stochastic Differential Equations, volume 124 ofCRC Monographs on St...

    2012

  44. [52]

    Drift estimation for stochastic reaction-diffusion sys- tems.Electronic Journal of Statistics, 14:547 –579, 2020

    Gregor Pasemann and Wilhelm Stannat. Drift estimation for stochastic reaction-diffusion sys- tems.Electronic Journal of Statistics, 14:547 –579, 2020

    2020

  45. [53]

    On a class of exponential changes of measure for stochastic pdes.Stochastic Processes and their Applications, 185:104630, 2025

    Thorben Pieper-Sethmacher, Frank van der Meulen, and Aad van der Vaart. On a class of exponential changes of measure for stochastic pdes.Stochastic Processes and their Applications, 185:104630, 2025. ISSN 0304-4149. doi: https://doi.org/10.1016/j.spa.2025.104630

    2025

  46. [54]

    Simulation of infinite-dimensional diffusion bridges.arXiv preprint arXiv:2503.13177, 2025

    Thorben Pieper-Sethmacher, Frank van der Meulen, and Aad van der Vaart. Simulation of infinite-dimensional diffusion bridges.arXiv preprint arXiv:2503.13177, 2025

    2025 arXiv

  47. [55]

    Inverse Problems in Neural Field Theory.SIAM Journal on Applied Dynamical Systems, 8(4):1405–1433, January 2009

    Roland Potthast and Peter beim Graben. Inverse Problems in Neural Field Theory.SIAM Journal on Applied Dynamical Systems, 8(4):1405–1433, January 2009. ISSN 1536-0040. doi: 10.1137/080731220

    2009 doi

  48. [56]

    Cambridge University Press

    Sebastian Reich and Colin Cotter.Probabilistic Forecasting and Bayesian Data Assimilation. Cambridge University Press. Cambridge University Press, May 2015. ISBN 1-316-29942-2

    2015

  49. [57]

    Cambridge University Press, 1 edition, July 2023

    Daniel Sanz-Alonso, Andrew Stuart, and Armeen Taeb.Inverse Problems and Data Assimilation. Cambridge University Press, 1 edition, July 2023. ISBN 978-1-009-41431-9 978-1-009-41432-6 978-1-009-41429-6. doi: 10.1017/9781009414319

    2023 doi

  50. [58]

    Simo Sarkka, Arno Solin, and Jouni Hartikainen. Spatiotemporal learning via infinite- dimensional bayesian filtering and smoothing: A look at gaussian process regression through kalman filtering.IEEE Signal Processing Magazine, 30(4):51–61, 2013

    2013

  51. [59]

    Sauer and Steven J

    Timothy D. Sauer and Steven J. Schiff. Data assimilation for heterogeneous networks: The consensus set.Physical Review E, 79(5):051909, May 2009. ISSN 1539-3755, 1550-2376. doi: 10.1103/PhysRevE.79.051909

    2009 doi

  52. [60]

    Guided proposals for simulat- ing multi-dimensional diffusion bridges.Bernoulli, 23(4A):2917 – 2950, 2017

    Moritz Schauer, Frank van der Meulen, and Harry van Zanten. Guided proposals for simulat- ing multi-dimensional diffusion bridges.Bernoulli, 23(4A):2917 – 2950, 2017. doi: 10.3150/ 16-BEJ833

    2017

  53. [61]

    Kalman filter control of a model of spatiotemporal cortical dynamics.Journal of neural engineering, 5(1):1, 2007

    Steven J Schiff and Tim Sauer. Kalman filter control of a model of spatiotemporal cortical dynamics.Journal of neural engineering, 5(1):1, 2007

    2007

  54. [62]

    K¨ unsch, and Werner A

    Fabio Sigrist, Hans R. K¨ unsch, and Werner A. Stahel. An spde-based spatio-temporal model for large data sets with an application to postprocessing precipitation forecasts.Journal of the Royal Statistical Society: Series C (Applied Statistics), 64(2):371–392, 2015. doi: 10.11...

    2015 doi

  55. [63]

    Cotter, and Dan Crisan

    Maneesh Kumar Singh, Joshua Hope-Collins, Colin J. Cotter, and Dan Crisan. Data assimilation using a global girsanov nudged particle filter, 2025. URLhttps://arxiv.org/abs/2507.17685

    2025 arXiv

  56. [64]

    Smith.Uncertainty Quantification: Theory, Implementation, and Applications

    Ralph C. Smith.Uncertainty Quantification: Theory, Implementation, and Applications. Num- ber 12 in Computational Science & Engineering. siam, Society for Industrial and Applied Math- ematics, Philadelphia, 2014. ISBN 978-1-61197-321-1

    2014

  57. [65]

    Inverse problems: a bayesian perspective.Acta numerica, 19:451–559, 2010

    Andrew M Stuart. Inverse problems: a bayesian perspective.Acta numerica, 19:451–559, 2010

    2010

  58. [66]

    Backward filtering forward guiding,

    Frank van der Meulen, Moritz Schauer, and Stefan Sommer. Backward filtering forward guiding,

  59. [67]

    Volume: 2 Reaction-Diffusion Equations, volume 104 ofMonographs in Mathematics

    Vitaly Volpert.Elliptic Partial Differential Equations. Volume: 2 Reaction-Diffusion Equations, volume 104 ofMonographs in Mathematics. Springer Basel, Basel, 2014. ISBN 978-3-0348-0812-5 978-3-0348-0813-2. doi: 10.1007/978-3-0348-0813-2

    2014 doi

  60. [68]

    The unscented kalman filter for nonlinear estimation

    Eric A Wan and Rudolph Van Der Merwe. The unscented kalman filter for nonlinear estimation. InProceedings of the IEEE 2000 adaptive systems for signal processing, communications, and control symposium (Cat. No. 00EX373), pages 153–158. Ieee, 2000

    2000

  61. [69]

    Chapman and Hall/CRC, 2019

    Christopher K Wikle, Andrew Zammit-Mangion, and Noel Cressie.Spatio-temporal statistics with R. Chapman and Hall/CRC, 2019. 36 GUIDED FILTERING AND SMOOTHING FOR INFINITE-DIMENSIONAL DIFFUSIONS

    2019

  62. [70]

    A mathematical theory of the functional dynamics of cortical and thalamic nervous tissue.Kybernetik, 13(2):55–80, 1973

    Hugh R Wilson and Jack D Cowan. A mathematical theory of the functional dynamics of cortical and thalamic nervous tissue.Kybernetik, 13(2):55–80, 1973

    1973

  63. [71]

    Severinsen, Christy Anna Hipsley, and Ste- fan Sommer

    Gefan Yang, Elizabeth Louise Baker, Michael L. Severinsen, Christy Anna Hipsley, and Ste- fan Sommer. Simulating infinite-dimensional nonlinear diffusion bridges, 2024. URLhttps: //arxiv.org/abs/2405.18353

    2024 arXiv

  64. [72]

    Chapman and Hall/CRC, 2002

    Valentin F Zaitsev and Andrei D Polyanin.Handbook of exact solutions for ordinary differential equations. Chapman and Hall/CRC, 2002

    2002

  65. [2019]

    doi: 10.1007/978-3-030-02487-1 2

    ISBN 978-3-030-02486-4 978-3-030-02487-1. doi: 10.1007/978-3-030-02487-1 2

    doi

  66. [2023]

    doi: 10.1007/978-3-031-21916-0

    ISBN 978-3-031-21915-3 978-3-031-21916-0. doi: 10.1007/978-3-031-21916-0

    doi

  67. [4149]

    URLhttp://www.sciencedirect.com/ science/article/pii/S0304414906000469

    doi: https://doi.org/10.1016/j.spa.2006.04.004. URLhttp://www.sciencedirect.com/ science/article/pii/S0304414906000469

    2006 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.