Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

This paper claims that a neural simulation-based estimator and a pseudo-marginal particle MCMC sampler both deliver accurate, robust Bayesian inference for stochastic compartmental epidemic models—the neural estimator roughly ten times fast

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

A head-to-head benchmark shows particle-filter MCMC and conditional normalizing flows both infer parameters of stochastic SIR/SEIR epidemic models reliably, with CNF broader and ~10x faster per inference and PF tighter but slower.

T0 review reviewed 2026-08-03 challenge →

load-bearing objection A solid, honest benchmark of PF vs CNF for stochastic epidemic models; the main conclusion holds, but the paper hand-waves CNF's calibration failures and the speed comparison is on unequal hardware. the 3 major comments →

arxiv 2512.02528 v4 pith:Q75UUA2K submitted 2025-12-02 q-bio.QM

Assessment of Simulation-based Inference Methods for Stochastic Compartmental Models in Epidemiological Research

Vincent Wieland , Nils Wassmuth , Lorenzo Contento , Martin K\"uhn , Jan Hasenauer This is my paper
classification q-bio.QM
keywords simulation-based inferenceBayesian parameter inferencestochastic compartmental modelsparticle Markov chain Monte Carloconditional normalizing flowsepidemic nowcastingpractical identifiabilityCOVID-19 modeling
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

The paper sets out to establish that two likelihood-free Bayesian inference methods—conditional normalizing flows, a trained neural network that directly approximates the posterior, and pseudo-marginal particle Markov chain Monte Carlo, which uses an unbiased particle-filter likelihood estimate inside an MCMC sampler—are both viable for fitting stochastic compartmental epidemic models. Across synthetic and real-world data on an SIR model and a two-variant SEIR model, both methods recover the latent dynamics, produce similar predictive fits, and give accurate point estimates. The practical payoff would be that epidemic modelers can choose between a once-trained, roughly 10x faster neural estimator for nowcasting and a slower but formally exact sampler, depending on whether speed or posterior fidelity matters. The paper reports a caveat: the neural flow's formal calibration diagnostics deviate in several experiments, which the authors argue does not degrade point estimates or predictive accuracy.

Core claim

The paper's central claim is that both conditional normalizing flows (CNF) and particle filters inside pseudo-marginal MCMC (PF) are viable inference strategies for complex stochastic epidemic models, with CNF focusing on broader parameter-space exploration and PF on tighter data fitting. The supporting evidence comes from a controlled comparison on stochastic differential equation versions of SIR and two-variant SEIR models, including dense, sparse, and partially missing observation schedules, plus a real Ethiopian COVID-19 cohort dataset. In the identifiable SIR case the posteriors from both methods almost perfectly overlap; in the non-identifiable SEIR case PF yields narrower posteriors l

What carries the argument

The comparison is carried by two inference engines: (1) pseudo-marginal particle MCMC, where a bootstrap particle filter with 200 particles produces an unbiased estimate of the intractable likelihood, embedded in an adaptive Metropolis-Hastings sampler; and (2) conditional normalizing flows, invertible neural networks trained offline on 100,000 simulator samples by maximizing the change-of-variables log-density, with learned LSTM summary statistics to handle irregular and missing data. A third mechanism is the reparametrization (r0 = beta*gamma^-1, e0 = kappa^-1 + gamma^-1, s0 = s*gamma^-1) that collapses the model's practical non-identifiability manifold and improves posterior alignment. Th

Load-bearing premise

The load-bearing premise is that the normalizing-flow estimator's formal calibration deviations (seen in Supplementary Figures S22–S23) are benign enough not to bias the credible intervals used in forecasts, and that 'accuracy' on real data can be measured against the published parameter means from the same group's deterministic ODE analysis; if either fails, the conclusion that CNF is a viable stand-alone inference strategy is weakened.

What would settle it

Run the trained CNF on, say, 2,000 simulated datasets and compute the empirical coverage of its 90% credible intervals for the parameters and for future infection counts; if coverage falls substantially below 90% in the regions that drive nowcasts—while particle MCMC intervals on the same datasets stay near nominal—then the miscalibration is not benign and the claim that CNF is a viable stand-alone inference strategy for epidemic nowcasting would be falsified.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • On an identifiable stochastic SIR model, CNF and PF produce nearly identical marginal and joint posteriors, MAP estimates close to the true parameters, and equal predictive RMSE; practitioners can use either method.
  • On the non-identifiable two-variant SEIR model, both methods fit the data equally well, but PF gives tighter posteriors with low effective sample sizes while CNF gives broader posteriors with imperfect calibration; the difference largely reflects posterior geometry, not an error in either method.
  • Sparse, irregularly sampled, and partially missing data do not degrade inference quality for either method; PF handles such data naturally, while CNF requires per-missingness-pattern amortization design.
  • On real Ethiopian cohort data, both methods produce posterior predictive trajectories with RMSEs comparable to or better than the reference deterministic analysis, with PF closer on infection counts.
  • CNF is about 10 times faster than PF under comparable CPU settings and the same number of posterior samples, making it attractive for time-sensitive nowcasting once the network is trained.

Where Pith is reading between the lines

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

  • The paper trains a fresh flow per experiment rather than a globally amortized model; a truly amortized CNF across time grids and missingness patterns would be needed to realize the 10x speed advantage for repeated nowcasts, and the paper's own discussion flags this conditioning challenge.
  • Because the reparametrization so strongly improved posterior alignment, a natural testable extension is to apply the same identifiability-aware reparametrization to other multi-variant epidemic models and check whether neural and MCMC posterior differences shrink correspondingly.
  • The authors report CNF posteriors that are broader than PF's, in contrast to the common pattern where amortized neural inference underestimates uncertainty; if replicated, this suggests the flow is exploring the non-identifiable manifold rather than collapsing onto it, and one could test this by measuring CNF coverage separately for identifiable versus weakly identified parameters.
  • The natural next step implied by the trade-off is a hybrid: use a trained CNF to propose parameters inside a pseudo-marginal MCMC kernel, combining the speed and tail coverage of neural proposals with the asymptotic exactness of particle MCMC; the paper gestures toward this but does not test it.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper benchmarks two Bayesian inference approaches for stochastic compartmental epidemic models: pseudo-marginal particle MCMC (PF; a bootstrap particle filter inside adaptive Metropolis-Hastings) and conditional normalizing flows (CNF; BayesFlow with learned summary statistics). It evaluates them on a stochastic SIR model and a two-variant SEIR model with Gaussian observation noise, using synthetic dense/sparse/missing data and real longitudinal COVID-19 data from Ethiopia. The authors compare posterior shapes, MAP estimates, RMSE, convergence diagnostics, SBC calibration, and computational cost. They conclude that both methods provide accurate and robust inference, with CNF giving broader posteriors and roughly 10x faster inference, while PF gives tighter posteriors and better sampling diagnostics in identifiable settings, but suffers from low ESS in the full SEIR model.

Significance. If the results hold after the calibration question is resolved, the paper is a useful practical benchmark for epidemic nowcasting. Its strengths include a systematic experimental design (identifiable and non-identifiable models, dense/sparse/missing data, real-world noise), a broad diagnostic suite (R-hat, ESS, SBC, recovery plots, posterior predictive intervals), and public code and synthetic data on Zenodo/GitHub. The comparison of an amortized neural posterior estimator against a pseudo-marginal MCMC baseline on realistic stochastic compartmental models is timely and of interest to the q-bio and SBI communities. The central claim is conditional on whether CNF's SBC miscalibration can be shown to be benign for decision-relevant quantities; the paper currently asserts this without a coverage check.

major comments (3)
  1. [§5.1, §5.3; Figs S22–S23] The claim that CNF's SBC deviations are benign for point estimates and predictive accuracy is not established. SBC ECDF diagnostics probe the full posterior, while recovery R² and posterior-predictive RMSE are averaged summaries that can hide systematic bias in credible intervals. The deviations in Fig. S23 are large relative to the 95% bands, so this is not a negligible artifact. Since the abstract and discussion promise 'accurate and robust inference capabilities' for both methods, please add a coverage evaluation on held-out synthetic datasets: for nominal 50/90/95% credible intervals, report empirical coverage for each parameter and for decision-relevant quantities (peak infection, cumulative incidence/seroprevalence). If CNF intervals are not calibrated, either demonstrate that the bias is confined to weakly identifiable directions with no impact on forecasts, or soften the claims a
  2. [§5.1, Table S27, §6] The 'roughly 10-times faster' conclusion is a central practical claim, but the timing comparison is not clearly normalized. Table S27 reports 'Accumulated CPU-time in minutes' while noting that PF ran on 4 cores and CNF on 20 cores. If these are wall-clock times, the comparison is not iso-core and the speedup is inflated; if they are aggregate CPU times, 'faster' should be rephrased as 'lower total compute time' and wall-clock speed should also be reported. Please provide timings with matched core allocations or core-normalized values, and state the metric unambiguously in the main text.
  3. [§5.2, Tables S13–S14; §5.4, Table S25] In the full SEIR model, PF's ESS values for γ⁻¹, κ, β, and s are roughly 400–600, which is close to the 100·nchains = 400 guideline, and R-hat reaches 1.025. The text describes this as 'limited exploration of posterior tails' but still concludes that PF provides 'more exact and tighter posterior estimates.' Tighter posteriors from chains with marginal ESS and R-hat near these values require caution; please quantify Monte Carlo error of the PF posterior widths (e.g., repeated seeds or per-chain standard errors) or temper the 'more exact' wording.
minor comments (6)
  1. [Figure S19] The caption says 'Results of the SIR model with dataset SIR 2,' but the panels display r0, e0, s0, tvar, and I0, which correspond to the reparametrized SEIR model. The caption should be corrected.
  2. [Abstract / Section 1] The front matter abstract lists SIS among the models, while the full-text abstract and Section 1 treat only SIR and SEIR. Please align the model list between the abstract and the body.
  3. [§2.1, Table 1 / Table S3] For the SIR model, I0 is listed as an initialization parameter but its value and whether it is estimated or fixed are not stated in Section 2.1 or Table S3. Please clarify.
  4. [§4 / Tables S13, S22] The text cites the guideline that ESS should exceed 100·nchains, but some full-SEIR ESS values are close to this threshold. State explicitly whether each reported ESS is considered acceptable or not, rather than only calling values 'low.'
  5. [§5.2] Typo: 'Supplemetary Tables' should be 'Supplementary Tables.' Also, when stating that CNF showed 'good calibration' in Supplementary Figure S20, specify whether the ECDF differences remain within the 95% bounds for all parameters; the y-axis range in that figure makes small excursions potentially visible.
  6. [§5.4, Table 7] The real-data 'public parameters' are the published deterministic mean estimates from [1]. This should be stated clearly in the main text (currently the supplement/caption only implies it), so readers do not mistake the real-data RMSE for a comparison against ground truth.

Circularity Check

0 steps flagged

No significant circularity: an empirical benchmark with independent synthetic recovery; self-citations are minor and not load-bearing.

full rationale

This is an empirical comparison study, not a derivation. The central claims are that CNF and PF both produce accurate posteriors and predictions for stochastic compartmental models; these claims are tested mostly on synthetic data with known ground truth, where the posterior estimates and RMSEs are compared against data-generating parameters rather than against any fitted quantity. The SIR, SEIR, and reparametrized SEIR experiments use independent simulators and priors, and the PF baseline is a standard pseudo-marginal PMMH implementation. No equation in the paper reduces to another by construction, and no fitted parameter is renamed as a prediction. Some self-citations appear: the real-data baseline uses published parameter means from [1], which shares authors with the present paper, and missing-data encoding follows [59], also from the same group. However, these references are not load-bearing for the main comparison: [1] is used only as a reference point for real-data RMSE, not to define the inference methodology or the outcome metric, and [59] supplies an implementation detail. The paper also openly reports CNF SBC miscalibration in Sections 5.1, 5.3 and the discussion, and the conclusion that the practical impact is minimal is an assertion rather than a derivation; that is a validation weakness but not circularity. Therefore no specific circular step exists, and the appropriate score is low, reflecting only minor self-citation without dependence.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 0 invented entities

No new particles, forces, or entities are introduced. The paper's central claim rests on standard inference theory, a diffusion approximation, and a set of manually chosen algorithmic hyperparameters. The reparametrized quantities r0, e0, s0 are derived combinations of existing parameters, not new entities.

free parameters (3)
  • Number of particles N in bootstrap filter = 100 (SIR), 200 (SEIR)
    Chosen by hand so that the variance of the log-likelihood estimate remains below one and particle ancestor diversity is maintained (Section A.4.1). Affects PMMH mixing and posterior quality.
  • CNF architecture hyperparameters = summary dim 2*n_theta+2, LSTM units 64, 8 coupling layers, 100 epochs, batch size 32, 100k simulations
    Fixed by a preliminary hyperparameter search and kept identical across experiments. These choices affect CNF calibration and recovery, which are central to the comparison.
  • Fixed kappa^-1 in reparametrized SEIR model = 5 (from the full model true parameter)
    The reparametrization in Section 2.2 fixes kappa^-1, and the alignment result depends on this fixed value. The paper does not explicitly state the value in the main text.
axioms (5)
  • domain assumption The SDE diffusion approximation with Euler-Maruyama integration and positivity clipping adequately represents the epidemic process.
    Section 2 and A.1 define the model as an SDE approximation to a CTMC; the Discussion acknowledges discretization error and numerical instabilities near boundaries.
  • standard math The bootstrap particle filter gives an unbiased likelihood estimator, so PMMH targets the exact posterior of the discretized model.
    Section 3.1 relies on the standard pseudo-marginal theory of Andrieu, Doucet, and Holenstein. This is valid under the usual assumptions on the unbiased estimator.
  • domain assumption CNF trained by maximum likelihood on simulator samples approximates the true posterior well enough that SBC deviations can be treated as practically irrelevant.
    Sections 5.1-5.4 observe SBC miscalibration for CNF (Figures S22, S23) but conclude practical impact is minimal based on point estimates and RMSE.
  • domain assumption Gaussian additive observation noise with binomial-derived standard deviations is an adequate noise model for both synthetic and real data.
    Section 2.3 and A.2 specify the noise model; the Discussion explicitly concedes that overdispersed counts may be better handled by other distributions.
  • domain assumption Priors from the cited literature and the published parameter estimates from [1] are appropriate reference points for the SEIR model.
    Section 5.2 uses priors from [61,62]; Section 5.4 uses published means from [1] as the real-data baseline. Ref [1] shares authors with the present paper.

reviewed 2026-08-03 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Assessment of Simulation-based Inference Methods for Stochastic Compartmental Models in Epidemiological Research." pith.science (2026). https://pith.science/paper/Q75UUA2K

@misc{pith2026251202528,
  author       = {Pith},
  title        = {Pith review of: Assessment of Simulation-based Inference Methods for Stochastic Compartmental Models in Epidemiological Research},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Q75UUA2K}},
  note         = {Machine review of arXiv:2512.02528}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Global pandemics, such as the recent COVID-19 crisis, highlight the need for stochastic epidemic models that can capture the randomness inherent in the spread of disease. Such models must be accompanied by methods for estimating parameters in order to generate fast nowcasts and short-term forecasts that can inform public health decisions. This paper presents a comparison of two advanced Bayesian inference methods: 1) pseudo-marginal particle Markov chain Monte Carlo, using an unbiased likelihood estimate obtained by Particle Filter (PF), and 2) Conditional Normalizing Flows (CNF). We investigate their performance on three commonly used compartmental models: A classical Susceptible-Infected-Susceptible (SIS), a Susceptible-Infected-Recovered (SIR) model and a two-variant Susceptible-Exposed-Infected-Recovered (SEIR) model, complemented by an observation model that maps latent trajectories to empirical data. Addressing the challenges of intractable likelihoods for parameter inference in stochastic settings, our analysis highlights how these likelihood-free methods provide accurate and robust inference capabilities. The results of our simulation study further underscore the effectiveness of these approaches in capturing the stochastic dynamics of epidemics, providing prediction capabilities for the control of epidemic outbreaks. Results on an Ethiopian cohort study demonstrate operational robustness under real-world noise and irregular data sampling. To facilitate reuse and to enable building pipelines that ultimately contribute to better informed decision making in public health, we make code and synthetic datasets publicly available.

Figures

Figures reproduced from arXiv: 2512.02528 by Jan Hasenauer, Lorenzo Contento, Martin K\"uhn, Nils Wassmuth, Vincent Wieland.

Figure 1
Figure 1. Figure 1: Schematic representations of the models and methods. (a) Graph of compart￾ments and possible transitions with corresponding rate parameters for the SIR model. (b) Graph of compartments and possible transitions with corresponding rate parameters for the two-variant SEIR model. (c) Workflow for the assessment of the Bayesian inference. 5 [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Evaluation of the posterior approximations for the SIR model A Posterior approximations based on 10,000 samples, true parameters are indicated by bold black lines and joint MAP estimates appear in a darker shade. B Model fit with percentiles (50%, 90%, 95%) based on simulating the 10,000 samples. Beyond these direct comparisons, the methods differed in sampling quality. PF achieved excellent convergence, w… view at source ↗
Figure 3
Figure 3. Figure 3: Marginal posterior approximations and prior distribution. Histograms of poste￾rior approximations and prior distribution for the two-variant SEIR model using a dense dataset. mates differ under CNF and PF. We first examined posterior distributions and predictive performance using four dense simu￾lated observational time series. Posterior distributions showed markedly lower variance compared to the correspo… view at source ↗
Figure 4
Figure 4. Figure 4: Evaluation of the posterior approximations for the two-variant SEIR model with a dense dataset. A Posterior approximations based on 10,000 samples, true parameters are indicated by bold black lines and joint MAP estimates appear in a darker shade. B Model fit with percentiles (50%, 90%, 95%) based on simulating the 10,000 samples. relatively low (∼500–600) for several parameters (Supplementary Table S13), … view at source ↗
Figure 5
Figure 5. Figure 5: Evaluation of the posterior approximations for the reparametrized two-variant SEIR model with dense data. A Posterior approximations based on 10,000 samples, true parameters are indicated by bold black lines and joint MAP estimates appear in a darker shade. B Model fit with percentiles (50%, 90%, 95%) based on simulating the 10,000 samples. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Evaluation of the posterior approximations for the two-variant SEIR model with real data. A Posterior approximations based on 10,000 samples, true parameters are indicated by bold black lines and joint MAP estimates appear in a darker shade. B Model fit with percentiles (50%, 90%, 95%) based on simulating the 10,000 samples. 20 [PITH_FULL_IMAGE:figures/full_fig_p020_6.png] view at source ↗

Review history (2 revisions) →

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. MEmilio -- A high performance Modular EpideMIcs simuLatIOn software for multi-scale and comparative simulations of infectious disease dynamics

    q-bio.PE 2026-02 conditional novelty 6.0

    MEmilio is a unified, open-source, high-performance epidemic-modeling framework that implements compartmental, metapopulation, agent-based, hybrid, and ML-surrogate models in one C++/Python codebase.

Reference graph

Works this paper leans on

79 extracted references · 1 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Esayas Kebede Gudina, Solomon Ali, Eyob Girma, Addisu Gize, Birhanemeskel Tegene, Gadissa Bedada Hundie, Wondewosen Tsegaye Sime, Rozina Ambachew, Alganesh Gebrey- ohanns, Mahteme Bekele, Abhishek Bakuli, Kira Elsbernd, Simon Merkt, Lorenzo Contento, Michael Hoelscher, Jan Hasenauer, Andreas Wieser, and Arne Kroidl. Seroepidemiology and model-based predic...

    2021

  2. [2]

    Assessing the use of influenza forecasts and epidemiological modeling in public health decision making in the united states.Scientific reports, 8(1):12406, 2018

    Colin Doms, Sarah C Kramer, and Jeffrey Shaman. Assessing the use of influenza forecasts and epidemiological modeling in public health decision making in the united states.Scientific reports, 8(1):12406, 2018

    2018

  3. [3]

    The changing face of pandemic risk: 2024 report, 2025

    Global Preparedness Monitoring Board. The changing face of pandemic risk: 2024 report, 2025

    2024

  4. [4]

    William Ogilvy Kermack, A. G. McKendrick, and Gilbert Thomas Walker. A contribution to the mathematical theory of epidemics.Proceedings of the Royal Society of London, Series A, 115(772):700–721, 1927

    1927

  5. [5]

    The mathematics of infectious diseases.SIAM review, 42(4):599–653, 2000

    Herbert W Hethcote. The mathematics of infectious diseases.SIAM review, 42(4):599–653, 2000

    2000

  6. [6]

    Agent-based models in empirical social research.Socio- logical Methods & Research, 44(2):186–221, 2015

    Elizabeth Bruch and Jon Atwell. Agent-based models in empirical social research.Socio- logical Methods & Research, 44(2):186–221, 2015

    2015

  7. [7]

    Stochastic epidemic models: a survey.Mathematical Biosciences, 225(1):24– 35, 2010

    Tom Britton. Stochastic epidemic models: a survey.Mathematical Biosciences, 225(1):24– 35, 2010

    2010

  8. [8]

    Fakult¨ at f¨ ur Mathematik, Informatik und Statistik der Ludwig-Maximilians-Universit¨ at M¨ unchen, 2010

    Christiane Dragatz.Bayesian inference for diffusion processes. Fakult¨ at f¨ ur Mathematik, Informatik und Statistik der Ludwig-Maximilians-Universit¨ at M¨ unchen, 2010

    2010

  9. [9]

    Linda J. S. Allen. A primer on stochastic epidemic models: Formulation, numerical simu- lation, and analysis.Infectious Disease Modelling, 2:128–142, 2017

    2017

  10. [10]

    Perspective: Stochastic algorithms for chemical kinetics.Journal of Chemical Physics, 138, 2013

    Dan Gillespie, Andreas Hellander, and Linda Petzold. Perspective: Stochastic algorithms for chemical kinetics.Journal of Chemical Physics, 138, 2013

    2013

  11. [11]

    A diffusion approximation for an epidemic model, 2006

    Christiane Dragatz. A diffusion approximation for an epidemic model, 2006

    2006

  12. [12]

    Springer Berlin Heidelberg, 2008

    Fred Brauer.Compartmental Models in Epidemiology, chapter 6, pages 19–79. Springer Berlin Heidelberg, 2008

    2008

  13. [13]

    N. G. van Kampen. The expansion of the master equation.Advance in Chemical Physics, 19, 1976

    1976

  14. [14]

    Inference for reaction networks using the linear noise approximation.Biometrics, 70(2):457–466, 2014

    Paul Fearnhead, Vasilieos Giagos, and Chris Sherlock. Inference for reaction networks using the linear noise approximation.Biometrics, 70(2):457–466, 2014

    2014

  15. [15]

    Solutions of ordinary differential equations as limits of pure jump markov processes.Journal of Applied Probability, 7(1):49–58, 1970

    Thomas G Kurtz. Solutions of ordinary differential equations as limits of pure jump markov processes.Journal of Applied Probability, 7(1):49–58, 1970. 24

    1970

  16. [16]

    Cambridge Studies in Mathematical Biology

    Hakan Andersson and Tom Britton.Stochastic epidemic models and their statistical analy- sis. Cambridge Studies in Mathematical Biology. Springer Verlag, 2001

    2001

  17. [17]

    Estimating parameters in stochastic compartmental models using Markov chain methods.Mathematical Medicine and Biology: A Journal of the IMA, 15(1):19–40, 1998

    Gavin J Gibson and Eric Renshaw. Estimating parameters in stochastic compartmental models using Markov chain methods.Mathematical Medicine and Biology: A Journal of the IMA, 15(1):19–40, 1998

    1998

  18. [18]

    Bayesian inference for partially observed stochas- tic epidemics.Journal of the Royal Statistical Society Series A: Statistics in Society, 162(1):121–129, 1999

    Philip D O’Neill and Gareth O Roberts. Bayesian inference for partially observed stochas- tic epidemics.Journal of the Royal Statistical Society Series A: Statistics in Society, 162(1):121–129, 1999

    1999

  19. [19]

    Challenges in estimation, uncertainty quantification and elicitation for pandemic modelling.Epidemics, 38:100547, 2022

    Ben Swallow, Paul Birrell, Joshua Blake, Mark Burgman, Peter Challenor, Luc E Coffeng, Philip Dawid, Daniela De Angelis, Michael Goldstein, Victoria Hemming, et al. Challenges in estimation, uncertainty quantification and elicitation for pandemic modelling.Epidemics, 38:100547, 2022

    2022

  20. [20]

    A linear noise approximation for stochastic epidemic models fit to partially observed incidence counts.Biometrics, 78(4):1530–1541, 2022

    Jonathan Fintzi, Jon Wakefield, and Vladimir N Minin. A linear noise approximation for stochastic epidemic models fit to partially observed incidence counts.Biometrics, 78(4):1530–1541, 2022

    2022

  21. [21]

    Simulation-based Bayesian inference for epidemic models.Computational Statistics & Data Analysis, 71:434– 447, 2014

    Trevelyan J McKinley, Joshua V Ross, Rob Deardon, and Alex R Cook. Simulation-based Bayesian inference for epidemic models.Computational Statistics & Data Analysis, 71:434– 447, 2014

    2014

  22. [22]

    Bayesian inference for nonlinear stochastic SIR epidemic model.Journal of Statistical Computation and Simulation, 86(11):2229–2240, 2016

    Hamid El Maroufy, Tewfik Kernane, Sidali Becheket, and Abdellah Ouddadj. Bayesian inference for nonlinear stochastic SIR epidemic model.Journal of Statistical Computation and Simulation, 86(11):2229–2240, 2016

    2016

  23. [23]

    On sequential Monte Carlo sam- pling methods for Bayesian filtering.Statistics and Computing, 2000

    Arnaud Doucet, Simon Godsill, and Christophe Andrieu. On sequential Monte Carlo sam- pling methods for Bayesian filtering.Statistics and Computing, 2000

    2000

  24. [24]

    Chopin, P

    N. Chopin, P. E. Jacob, and O. Papaspiliopoulos. SMC2: An efficient algorithm for se- quential analysis of state space models.Journal of the Royal Statistical Society Series B: Statistical Methodology, 75, 2012

    2012

  25. [25]

    Christophe Andrieu and Gareth O. Roberts. The pseudo-marginal approach for efficient Monte Carlo computations.The Annals of Statistics, 37(2), 2009

    2009

  26. [26]

    Particle Markov chain Monte Carlo methods.Journal of the Royal Statistical Society: Series B (Statistical Methodology), 72(3):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(3):269–342, 2010

    2010

  27. [28]

    McLean, and Umberto Picchini

    Samuel Wiqvist, Andrew Golightly, Ashleigh T. McLean, and Umberto Picchini. Efficient inference for stochastic differential equation mixed-effects models using correlated particle pseudo-marginal algorithms.Computational Statistics & Data Analysis, 157, 2021

    2021

  28. [29]

    Bayesianly justifiable and relevant frequency calculations for the applied statistician.The Annals of Statistics, pages 1151–1172, 1984

    Donald B Rubin. Bayesianly justifiable and relevant frequency calculations for the applied statistician.The Annals of Statistics, pages 1151–1172, 1984. 25

    1984

  29. [30]

    Approximate Bayesian computation for infectious disease modelling.Epidemics, 29, 2019

    Amanda Minter and Renata Retkute. Approximate Bayesian computation for infectious disease modelling.Epidemics, 29, 2019

    2019

  30. [31]

    The frontier of simulation-based in- ference.Proceedings of the National Academy of Sciences, 117(48):30055–30062, 2020

    Kyle Cranmer, Johann Brehmer, and Gilles Louppe. The frontier of simulation-based in- ference.Proceedings of the National Academy of Sciences, 117(48):30055–30062, 2020

    2020

  31. [32]

    Radev, Ulf K

    Stefan T. Radev, Ulf K. Mertens, Andreas Voss, Lynton Ardizzone, and Ullrich K¨ othe. BayesFlow: Learning complex stochastic models with invertible neural networks.IEEE transactions on neural networks and learning systems, 33(4):1452–1466, 2020

    2020

  32. [33]

    Stefan T Radev, Frederik Graw, Simiao Chen, Nico T Mutters, Vanessa M Eichel, Till B¨ arnighausen, and Ullrich K¨ othe. OutbreakFlow: Model-based Bayesian inference of disease outbreak dynamics with invertible neural networks and its application to the COVID-19 pandemics in Germany.PLoS computational biology, 17(10):e1009472, 2021

    2021

  33. [34]

    Geir Storvik, Alfonso Diz-Lois Palomares, Solveig Engebretsen, Gunnar Oyvind Isaksson Ro, Kenth Engo-Monsen, Anja Brathen Kristoffersen, Birgitte Freiesleben de Blasio, and Arnoldo Frigessi. A sequential monte carlo approach to estimate a time-varying reproduction number in infectious disease models: the covid-19 case*.Journal of the Royal Statistical Soc...

    2023

  34. [35]

    Tiberiu Harko, Francisco SN Lobo, and Man Kwong Mak. Exact analytical solutions of the susceptible-infected-recovered (sir) epidemic model and of the sir model with equal death and birth rates.Applied Mathematics and Computation, 236:184–194, 2014

    2014

  35. [36]

    Cambridge University Press, 1999

    Daryl J Daley and Joseph Mark Gani.Epidemic modelling: an introduction. Cambridge University Press, 1999

    1999

  36. [37]

    Princeton University Press, 2008

    Matt J Keeling and Pejman Rohani.Modeling infectious diseases in humans and animals. Princeton University Press, 2008

    2008

  37. [38]

    Wiley Online Library, 1994

    Jos´ e M Bernardo, Adrian FM Smith, and Mark Berliner.Bayesian theory, volume 586. Wiley Online Library, 1994

    1994

  38. [40]

    Wilkinson

    Darren J. Wilkinson. Parameter inference for stochastic kinetic models of bacterial gene regulation: A Bayesian approach to systems biology. InBayesian Statistics 9, pages 679–

  39. [41]

    Logos Verlag Berlin GmbH, 2013

    Jan Hasenauer.Modeling and parameter estimation for heterogeneous cell populations. Logos Verlag Berlin GmbH, 2013

    2013

  40. [42]

    Wilkinson

    Andrew Golightly and Darren J. Wilkinson. Bayesian parameter inference for stochastic biochemical network models using particle Markov chain Monte Carlo.Interface Focus, 1(6):807–820, 2011. Publisher: Royal Society

    2011

  41. [43]

    Springer, 2004

    Pierre Moral.Feynman-Kac formulae: Genealogical and interacting particle systems with applications. Springer, 2004. 26

    2004

  42. [45]

    Variational inference with normalizing flows

    Danilo Rezende and Shakir Mohamed. Variational inference with normalizing flows. In International conference on machine learning, pages 1530–1538. PMLR, 2015

    2015

  43. [46]

    Learning likeli- hoods with conditional normalizing flows.arXiv preprint, 2019

    Christina Winkler, Daniel Worrall, Emiel Hoogeboom, and Max Welling. Learning likeli- hoods with conditional normalizing flows.arXiv preprint, 2019

    2019

  44. [47]

    Normalizing flows: An introduc- tion and review of current methods.IEEE transactions on pattern analysis and machine intelligence, 43(11):3964–3979, 2020

    Ivan Kobyzev, Simon JD Prince, and Marcus A Brubaker. Normalizing flows: An introduc- tion and review of current methods.IEEE transactions on pattern analysis and machine intelligence, 43(11):3964–3979, 2020

    2020

  45. [48]

    Guided image generation with conditional invertible neural networks.arXiv preprint, 2019

    Lynton Ardizzone, Carsten L¨ uth, Jakob Kruse, Carsten Rother, and Ullrich K¨ othe. Guided image generation with conditional invertible neural networks.arXiv preprint, 2019

    2019

  46. [49]

    Structured con- ditional continuous normalizing flows for efficient amortized inference in graphical models

    Christian Weilbach, Boyan Beronov, William Harvey, and Frank Wood. Structured con- ditional continuous normalizing flows for efficient amortized inference in graphical models. InProceedings of the 23rd International Conference on Artificial Intelligence and Statistics, volume 108, pages 4441–4451, 2020

    2020

  47. [50]

    Witte, and Felix J

    Ali Siahkoohi, Gabrio Rizzuti, Philipp A. Witte, and Felix J. Herrmann. Faster uncertainty quantification for inverse problems with conditional normalizing flows.arXiv preprint, 2020

    2020

  48. [51]

    BayesFlow: Amortized Bayesian workflows with neural networks, 2023

    Stefan T Radev, Marvin Schmitt, Lukas Schumacher, Lasse Elsem¨ uller, Valentin Pratz, Yan- nik Sch¨ alte, Ullrich K¨ othe, and Paul-Christian B¨ urkner. BayesFlow: Amortized Bayesian workflows with neural networks, 2023

    2023

  49. [52]

    Val- idating bayesian inference algorithms with simulation-based calibration.arXiv preprint, 2018

    Sean Talts, Michael Betancourt, Daniel Simpson, Aki Vehtari, and Andrew Gelman. Val- idating bayesian inference algorithms with simulation-based calibration.arXiv preprint, 2018

    2018

  50. [53]

    Adaptive methods for stochastic differential equa- tions via natural embeddings and rejection sampling with memory.Discrete and continuous dynamical systems

    Christopher Rackauckas and Qing Nie. Adaptive methods for stochastic differential equa- tions via natural embeddings and rejection sampling with memory.Discrete and continuous dynamical systems. Series B, 22(7):2731, 2017

    2017

  51. [54]

    Modelingtoolkit: A composable graph transformation system for equation- based modeling, 2021

    Yingbo Ma, Shashi Gowda, Ranjan Anantharaman, Chris Laughman, Viral Shah, and Chris Rackauckas. Modelingtoolkit: A composable graph transformation system for equation- based modeling, 2021

    2021

  52. [55]

    An adaptive metropolis algorithm

    Heikki Haario, Eero Saksman, and Johanna Tamminen. An adaptive metropolis algorithm. Bernoulli, 7(2):223 – 242, 2001

    2001

  53. [56]

    pyPESTO - Parameter EStimation TOolbox for python (v0.2.16), 2023

    Yannik Sch¨ alte, Fabian Fr¨ ohlich, Paul Stapor, Jakob Vanhoefer, Daniel Weindl, Paul Jonas Jost, Dantong Wang, Polina Lakrisenko, Elba Raim´ undez, Dilan Pathirana, Leonard Schmi- ester, Philipp St¨ adter, Lorenzo Contento, Simon Merkt, Erika Dudkin, Stephan Grein, and Jan Hasenauer. pyPESTO - Parameter EStimation TOolbox for python (v0.2.16), 2023

    2023

  54. [57]

    Tor Erlend Fjelde, Kai Xu, David Widmann, Mohamed Tarek, Cameron Pfiffer, Martin Trapp, Seth D Axen, Xianda Sun, Markus Hauru, Penelope Yong, et al. Turing. jl: a general-purpose probabilistic programming language.ACM Transactions on Probabilistic Machine Learning, 2025. 27

    2025

  55. [58]

    Rank-normalization, folding, and localization: An improved R for assessing convergence of MCMC (with discussion).Bayesian analysis, 16(2), 2021

    Aki Vehtari, Andrew Gelman, Daniel Simpson, Bob Carpenter, and Paul-Christian B¨ urkner. Rank-normalization, folding, and localization: An improved R for assessing convergence of MCMC (with discussion).Bayesian analysis, 16(2), 2021

    2021

  56. [59]

    Missing data in amortized simulation- based neural posterior estimation.PLOS Computational Biology, 20(6):e1012184, 2024

    Zijian Wang, Jan Hasenauer, and Yannik Sch¨ alte. Missing data in amortized simulation- based neural posterior estimation.PLOS Computational Biology, 20(6):e1012184, 2024

    2024

  57. [60]

    Routledge, 1998

    Bernard W Silverman.Density estimation for statistics and data analysis. Routledge, 1998

    1998

  58. [61]

    Incubation period of COVID-19: A rapid systematic review and meta-analysis of observational research.BMJ open, 10(8):e039652, 2020

    Conor McAloon, ´Aine Collins, Kevin Hunt, Ann Barber, Andrew W Byrne, Francis Butler, Miriam Casey, John Griffin, Elizabeth Lane, David McEvoy, et al. Incubation period of COVID-19: A rapid systematic review and meta-analysis of observational research.BMJ open, 10(8):e039652, 2020

    2020

  59. [62]

    Comparisons of viral shedding time of SARS-CoV-2 of different samples in ICU and non- ICU patients.The Journal of infection, 81(1):147, 2020

    Zhixiong Fang, Yi Zhang, Changfa Hang, Jingwen Ai, Shaojie Li, and Wenhong Zhang. Comparisons of viral shedding time of SARS-CoV-2 of different samples in ICU and non- ICU patients.The Journal of infection, 81(1):147, 2020

    2020

  60. [63]

    COVID-19 outbreak in Wuhan demonstrates the limitations of publicly available case numbers for epidemiological modeling.Epidemics, 34:100439, 2021

    Elba Raim´ undez, Erika Dudkin, Jakob Vanhoefer, Emad Alamoudi, Simon Merkt, Lara Fuhrmann, Fan Bai, and Jan Hasenauer. COVID-19 outbreak in Wuhan demonstrates the limitations of publicly available case numbers for epidemiological modeling.Epidemics, 34:100439, 2021

    2021

  61. [64]

    Posterior collapse and latent variable non-identifiability.Advances in neural information processing systems, 34:5443–5455, 2021

    Yixin Wang, David Blei, and John P Cunningham. Posterior collapse and latent variable non-identifiability.Advances in neural information processing systems, 34:5443–5455, 2021

    2021

  62. [65]

    A trust crisis in simulation-based inference? Your posterior approximations can be unfaithful.Transactions on Machine Learning Research, 2022

    Joeri Hermans, Arnaud Delaunoy, Fran¸ cois Rozet, Antoine Wehenkel, Volodimir Begy, and Gilles Louppe. A trust crisis in simulation-based inference? Your posterior approximations can be unfaithful.Transactions on Machine Learning Research, 2022

    2022

  63. [66]

    Particle filters, the “optimal” proposal and high-dimensional systems

    Chris Snyder. Particle filters, the “optimal” proposal and high-dimensional systems. In Proceedings of the ECMWF Seminar on Data Assimilation for Atmosphere and Ocean., pages 1–10, 2011

    2011

  64. [67]

    On particle Gibbs sampling.Bernoulli, 21(3), 2015

    Nicolas Chopin and Sumeetpal S Singh. On particle Gibbs sampling.Bernoulli, 21(3), 2015

    2015

  65. [68]

    Newcastle University, 2023

    Thomas Edward Lowe.Accelerating pseudo-marginal Metropolis-Hastings schemes for stochastic kinetic models. Newcastle University, 2023

    2023

  66. [69]

    Sequential neural likelihood: Fast likelihood-free inference with autoregressive flows

    George Papamakarios, David Sterratt, and Iain Murray. Sequential neural likelihood: Fast likelihood-free inference with autoregressive flows. InThe 22nd international conference on artificial intelligence and statistics, pages 837–848. PMLR, 2019

    2019

  67. [70]

    Likelihood-free MCMC with amortized approximate ratio estimators

    Joeri Hermans, Volodimir Begy, and Gilles Louppe. Likelihood-free MCMC with amortized approximate ratio estimators. InInternational conference on Machine Learning, pages 4239–4248. PMLR, 2020

    2020

  68. [71]

    Flow matching for generative modeling.arXiv preprint arXiv:2210.02747, 2022

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling.arXiv preprint arXiv:2210.02747, 2022

    Pith/arXiv arXiv 2022

  69. [72]

    Free-form flows: Make any architecture a normalizing flow

    Felix Draxler, Peter Sorrenson, Lea Zimmermann, Armand Rousselot, and Ullrich K¨ othe. Free-form flows: Make any architecture a normalizing flow. InInternational Conference on Artificial Intelligence and Statistics, pages 2197–2205. PMLR, 2024. 28

    2024

  70. [73]

    Jonas Arruda, Niels Bracher, Ullrich K¨ othe, Jan Hasenauer, and Stefan T. Radev. Diffusion models in simulation-based inference: A tutorial review.arXiv preprint, 2025

    2025

  71. [74]

    De Moivre on the law of normal probability.Smith, David Eugene

    Helen M Walker and M Helen. De Moivre on the law of normal probability.Smith, David Eugene. A Source Book in Mathematics, Dover, pages 64690–4, 1985

    1985

  72. [75]

    Doucet, M

    A. Doucet, M. K. Pitt, G. Deligiannidis, and R. Kohn. Efficient implementation of Markov chain Monte Carlo when using an unbiased likelihood estimator.Biometrika, 102(2):295– 313, 2015

    2015

  73. [76]

    Springer Series in Statistics

    Nicolas Chopin and Omiros Papaspiliopoulos.An Introduction to sequential Monte Carlo. Springer Series in Statistics. Springer International Publishing, 2020

    2020

  74. [77]

    Rank-normalization, folding, and localization: An improved ˆRfor assessing convergence of MCMC (with discussion).Bayesian analysis, 16(2), 2021

    Aki Vehtari, Andrew Gelman, Daniel Simpson, Bob Carpenter, and Paul-Christian B¨ urkner. Rank-normalization, folding, and localization: An improved ˆRfor assessing convergence of MCMC (with discussion).Bayesian analysis, 16(2), 2021

    2021

  75. [78]

    Johansen

    Arnaud Doucet and Adam M. Johansen. A tutorial on particle filtering and smoothing: Fifteen years later.Oxford Handbook of Nonlinear filtering, 2008

    2008

  76. [79]

    Doucet, Nando de Freitas, and Neil J

    A. Doucet, Nando de Freitas, and Neil J. Gordon.Sequential Monte Carlo methods in practice. Statistics for Engineering and Information Science. Springer New York, 2001

    2001

  77. [80]

    Douc and O

    R. Douc and O. Cappe. Comparison of resampling schemes for particle filtering.Proceedings of the 4th International Symposium on Image and Signal Processing and Analysis, pages 64–69, 2005

    2005

  78. [81]

    Esayas Kebede Gudina, Solomon Ali, Eyob Girma, Addisu Gize, Birhanemeskel Tegene, Gadissa Bedada Hundie, Wondewosen Tsegaye Sime, Rozina Ambachew, Alganesh Gebrey- ohanns, Mahteme Bekele, Abhishek Bakuli, Kira Elsbernd, Simon Merkt, Lorenzo Contento, Michael Hoelscher, Jan Hasenauer, Andreas Wieser, and Arne Kroidl. Seroepidemiology and model-based predic...

    2021

  79. [706]

    Oxford University Press, 2011

    2011

This paper was first reviewed by deepseek-v4-flash on August 3, 2026.