pith. machine review for the scientific record. sign in

arxiv: 2604.18319 · v1 · submitted 2026-04-20 · 📊 stat.ML · cs.LG· stat.ME

Recognition: unknown

Overcoming Selection Bias in Statistical Studies With Amortized Bayesian Inference

Jonas Arruda , Sophie Chervet , Paula Staudt , Andreas Wieser , Michael Hoelscher , Isabelle Sermet-Gaudelus , Nadine Binder , Lulla Opatowski
show 1 more author
Jan Hasenauer
Authors on Pith no claims yet

Pith reviewed 2026-05-10 03:41 UTC · model grok-4.3

classification 📊 stat.ML cs.LGstat.ME
keywords selection biasamortized Bayesian inferencesimulation-based inferenceneural posterior estimationgenerative simulatorsbias correctionBayesian statistics
0
0 comments X

The pith

Embedding the data selection process directly into generative simulators enables debiased amortized Bayesian inference without needing tractable likelihoods.

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

The paper shows that selection bias, where the chance an observation enters a dataset depends on outcomes or covariates of interest, can be corrected by treating the selection rule as part of the simulation that generates synthetic data for inference. Classical fixes require tractable likelihoods that many complex models lack, while standard simulation-based methods assume missingness at random and break when selection involves unobserved factors. By baking the selection mechanism into the simulator and using neural posterior estimation, the approach produces well-calibrated posteriors and lets users test for bias presence via diagnostics that compare simulated and real data. This recasts bias correction as a simulation problem that works for high-dimensional or latent-dynamic models common in epidemiology and surveys.

Core claim

By embedding the selection mechanism directly into the generative simulator, the approach enables amortized Bayesian inference without requiring tractable likelihoods. This recasting of selection bias as part of the simulation process allows both debiased estimates and explicit tests for the presence of bias, with integrated diagnostics for simulated-versus-observed discrepancies and posterior calibration. The method recovers well-calibrated posterior distributions across three statistical applications with diverse selection mechanisms, including cases where likelihood-based approaches yield biased estimates.

What carries the argument

Bias-aware neural posterior estimation that incorporates the selection mechanism directly into the generative simulator.

If this is right

  • Debiased parameter estimates become available for models with intractable likelihoods or latent dynamics.
  • Users can explicitly test whether selection bias is present by comparing posteriors with and without the embedded mechanism.
  • Diagnostics detect mismatches between simulated and observed data as well as miscalibrated uncertainty.
  • The framework succeeds in applications where inverse-probability weighting or explicit likelihood models produce biased results.

Where Pith is reading between the lines

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

  • The same embedding strategy could be used to handle other data-generating distortions such as confounding or measurement error by simulating them explicitly.
  • In survey or epidemiological practice this reduces reliance on post-hoc weighting and allows joint inference on both parameters and the selection process itself.
  • Sensitivity checks on the selection model specification become a standard diagnostic step before trusting the posteriors.

Load-bearing premise

The selection process can be fully and accurately specified and embedded into the generative simulator without introducing new unmodeled biases or needing knowledge of unobserved variables.

What would settle it

Apply the method to a controlled dataset where the true selection rule is known and varied, then check whether the recovered posteriors remain calibrated and unbiased when the embedded selection model matches the true rule versus when it is deliberately misspecified.

Figures

Figures reproduced from arXiv: 2604.18319 by Andreas Wieser, Isabelle Sermet-Gaudelus, Jan Hasenauer, Jonas Arruda, Lulla Opatowski, Michael Hoelscher, Nadine Binder, Paula Staudt, Sophie Chervet.

Figure 1
Figure 1. Figure 1: Different sources of selection bias in statistical studies. The forward process generates outcomes from model parameters and covariates, with selection acting as an additional component that determines which individuals are observed. Selection may depend on covariates alone (e.g., a non-representative sample), on latent states such as survival and hence parameters (e.g., individuals must be alive to remain… view at source ↗
Figure 2
Figure 2. Figure 2: Estimation of prevalence from biased population sample on simulated data and the KoCo19 Study. (A) Visualization of the infection model and selection bias based on covariates. (B) Covariate distribution in the KoCo19 cohort for all 5 rounds compared to Munich, with indication for whom the infection status y is missing. (C) Absolute error of estimated prevalence across 1000 simulated datasets (including sel… view at source ↗
Figure 3
Figure 3. Figure 3: Correcting for bias due to missing disease information because of death on simulations and the Framingham Heart Study. (A) Visualization of the time-to-event model and bias due to missing dementia because of death. (B) Observable and unobservable transitions between states from over 2200 patients in the Framingham Heart Study summed over all 4 epochs. Parts of the transitions to death are interval-censored… view at source ↗
Figure 4
Figure 4. Figure 4: Validating bias correction on the Framingham Heart Study. (A) A classifier is trained on parameter-data pairs to predict if the pair belongs to the posterior or the joint. (B) Marginal posteriors of the bias-aware NPE for the baseline transition scale parameter a01 (healthy to dementia) for all 4 epochs. Posterior samples were tested using a classifier-based diagnostic, where a C2ST score of 0.5 means that… view at source ↗
Figure 5
Figure 5. Figure 5: Correcting for bias under multiple selection procedures on simulated data and the PedCovid Study. (A) Visualization of selection bias based on outcome and covariates. (B) Fraction of household members who were the first to test positive in each age group, shown for real data and for different simulated selection mechanisms (infections occurring on the same day are assigned to the younger age group). A tota… view at source ↗
read the original abstract

Selection bias arises when the probability that an observation enters a dataset depends on variables related to the quantities of interest, leading to systematic distortions in estimation and uncertainty quantification. For example, in epidemiological or survey settings, individuals with certain outcomes may be more likely to be included, resulting in biased prevalence estimates with potentially substantial downstream impact. Classical corrections, such as inverse-probability weighting or explicit likelihood-based models of the selection process, rely on tractable likelihoods, which limits their applicability in complex stochastic models with latent dynamics or high-dimensional structure. Simulation-based inference enables Bayesian analysis without tractable likelihoods but typically assumes missingness at random and thus fails when selection depends on unobserved outcomes or covariates. Here, we develop a bias-aware simulation-based inference framework that explicitly incorporates selection into neural posterior estimation. By embedding the selection mechanism directly into the generative simulator, the approach enables amortized Bayesian inference without requiring tractable likelihoods. This recasting of selection bias as part of the simulation process allows us to both obtain debiased estimates and explicitly test for the presence of bias. The framework integrates diagnostics to detect discrepancies between simulated and observed data and to assess posterior calibration. The method recovers well-calibrated posterior distributions across three statistical applications with diverse selection mechanisms, including settings in which likelihood-based approaches yield biased estimates. These results recast the correction of selection bias as a simulation problem and establish simulation-based inference as a practical and testable strategy for parameter estimation under selection bias.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper proposes a bias-aware simulation-based inference framework that embeds the selection mechanism directly into the generative simulator to enable amortized neural posterior estimation (NPE) without requiring tractable likelihoods. It claims this recasts selection bias correction as a simulation problem, allows debiased estimates, explicit bias testing, and recovers well-calibrated posteriors across three statistical applications with diverse selection mechanisms (including cases where likelihood-based methods are biased). Diagnostics for data discrepancy and posterior calibration are integrated.

Significance. If the empirical results hold under the stated assumptions, the work would meaningfully extend simulation-based inference to handle selection bias in complex models common to epidemiology and surveys. It provides a practical alternative to inverse-probability weighting or explicit likelihood corrections when likelihoods are intractable. Credit is due for integrating calibration diagnostics and for demonstrating the approach on multiple applications. However, significance is limited by the untested requirement that the selection process (including dependence on unobserved variables) be known exactly and correctly specified in the simulator.

major comments (2)
  1. [§3 (Framework)] §3 (Framework): The central claim that embedding a known selection process P(select | y, x, z) into the simulator yields well-calibrated posteriors via amortized NPE holds only under correct specification. The manuscript provides no sensitivity analysis or robustness checks to misspecification of the selection model, which is load-bearing because any mismatch reintroduces bias; the three applications use hand-specified rules and thus do not test this case.
  2. [§5 (Applications and Results)] §5 (Applications and Results): While well-calibrated posteriors are asserted, specific quantitative validation (e.g., posterior coverage rates, calibration plots, or comparison metrics against likelihood-based baselines) is not detailed sufficiently to evaluate whether the data support superiority or calibration in the presence of selection on unobserved variables.
minor comments (2)
  1. [Abstract] Abstract: The phrase 'superiority over likelihood-based methods' is stated without accompanying metrics or conditions, reducing clarity.
  2. [§2 or §3] Notation for the generative model including the selection indicator should be introduced with an explicit equation early in the methods to aid readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed review. The comments identify important areas where additional analysis and quantitative detail will strengthen the manuscript. We address each major comment point-by-point below, agree that revisions are warranted, and outline the specific changes we will make.

read point-by-point responses

  1. Referee: §3 (Framework): The central claim that embedding a known selection process P(select | y, x, z) into the simulator yields well-calibrated posteriors via amortized NPE holds only under correct specification. The manuscript provides no sensitivity analysis or robustness checks to misspecification of the selection model, which is load-bearing because any mismatch reintroduces bias; the three applications use hand-specified rules and thus do not test this case.

    Authors: We agree that correct specification of the selection mechanism is a necessary assumption for the framework to produce well-calibrated posteriors, as is the case for any generative model in simulation-based inference. The manuscript states this modeling assumption explicitly. We did not include sensitivity analyses in the original submission, as the primary goal was to establish the approach under known selection processes. In the revision we will add a new subsection to §3 that discusses the consequences of misspecification and report a sensitivity study on one application (e.g., by perturbing selection probabilities in the epidemiological example) to quantify the resulting bias in posterior estimates. This will directly address the load-bearing nature of the assumption. revision: yes

  2. Referee: §5 (Applications and Results): While well-calibrated posteriors are asserted, specific quantitative validation (e.g., posterior coverage rates, calibration plots, or comparison metrics against likelihood-based baselines) is not detailed sufficiently to evaluate whether the data support superiority or calibration in the presence of selection on unobserved variables.

    Authors: We thank the referee for noting that the quantitative evidence could be presented more explicitly. The original manuscript reports calibration diagnostics and shows that likelihood-based methods are biased while our approach recovers the ground-truth parameters in the presence of selection on unobserved variables. To improve rigor and transparency, the revised version will include explicit posterior coverage rates (50%, 90%, and 95% credible intervals) computed over repeated simulations, additional calibration plots for all three applications, and tabulated metrics (bias, RMSE, and interval coverage) comparing our method to likelihood-based baselines where feasible. These additions will appear in the main text of §5 and the supplementary material. revision: yes

Circularity Check

0 steps flagged

No significant circularity; method is a direct application of standard simulation-based inference

full rationale

The paper proposes embedding a known selection mechanism into the generative simulator and then applying amortized neural posterior estimation (NPE) to recover posteriors. This is a straightforward methodological extension of existing SBI techniques rather than a derivation that reduces to fitted inputs or self-referential definitions. Validation rests on three independent empirical applications with hand-specified selection rules and calibration diagnostics, none of which are shown to be tautological by construction in the provided text. No load-bearing equations, uniqueness theorems, or ansatzes are imported via self-citation in a way that collapses the central claim. The framework is therefore self-contained against external simulation benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Based solely on the abstract, the central claim rests on the assumption that the selection mechanism is known and can be faithfully simulated; no specific free parameters or invented entities are mentioned.

axioms (1)
  • domain assumption The selection process can be explicitly and accurately incorporated into the generative simulator
    This is the core recasting described in the abstract that enables the bias-aware inference.

pith-pipeline@v0.9.0 · 5595 in / 1196 out tokens · 30385 ms · 2026-05-10T03:41:48.975732+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

300 extracted references · 124 canonical work pages · 2 internal anchors

  1. [1]

    Streeter and Nan Xuan Lin and Louise Crathorne and Marcela Haasova and Christopher Hyde and David Melzer and William E

    Adam J. Streeter and Nan Xuan Lin and Louise Crathorne and Marcela Haasova and Christopher Hyde and David Melzer and William E. Henley , keywords =. Adjusting for unmeasured confounding in nonrandomized longitudinal studies: a methodological review , journal =. 2017 , issn =. doi:https://doi.org/10.1016/j.jclinepi.2017.04.022 , url =

    work page doi:10.1016/j.jclinepi.2017.04.022 2017

  2. [2]

    Multivariate Behavioral Research , volume=

    An introduction to propensity score methods for reducing the effects of confounding in observational studies , author=. Multivariate Behavioral Research , volume=. 2011 , publisher=

    2011

  3. [3]

    American Journal of Epidemiology , volume=

    Using sensitivity analyses for unobserved confounding to address covariate measurement error in propensity score methods , author=. American Journal of Epidemiology , volume=. 2018 , publisher=

    2018

  4. [4]

    Johan Wiley & Sons Inc , year=

    Sampling techniques , author=. Johan Wiley & Sons Inc , year=

  5. [5]

    Econometrica: Journal of the Econometric Society , pages=

    Sample selection bias as a specification error , author=. Econometrica: Journal of the Econometric Society , pages=. 1979 , publisher=

    1979

  6. [6]

    2019 , publisher=

    Statistical analysis with missing data , author=. 2019 , publisher=

    2019

  7. [7]

    Statistics in Medicine , volume=

    The design of simulation studies in medical statistics , author=. Statistics in Medicine , volume=. 2006 , publisher=

    2006

  8. [8]

    Linero, Antonio R and Daniels, Michael J , journal=

  9. [9]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Inference for non-random samples , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 1997 , publisher=

    1997

  10. [10]

    A robust

    Bai, Ray and Lin, Lifeng and Boland, Mary R and Chen, Yong , journal=. A robust

  11. [11]

    Biostatistics , volume=

    Incorporating prior beliefs about selection bias into the analysis of randomized trials with missing outcomes , author=. Biostatistics , volume=. 2003 , publisher=

    2003

  12. [12]

    American journal of epidemiology , volume=

    Selection bias in epidemiologic studies , author=. American journal of epidemiology , volume=. 1981 , publisher=

    1981

  13. [13]

    International Journal of Epidemiology , volume=

    A systematic review of quantitative bias analysis applied to epidemiological research , author=. International Journal of Epidemiology , volume=. 2021 , publisher=

    2021

  14. [14]

    2008 , publisher=

    Modern epidemiology , author=. 2008 , publisher=

    2008

  15. [15]

    The Thirteenth International Conference on Learning Representations , year=

    Robust Simulation-Based Inference under Missing Data via Neural Processes , author=. The Thirteenth International Conference on Learning Representations , year=

  16. [16]

    Proceedings of the 41st International Conference on Machine Learning , pages=

    All-in-one simulation-based inference , author=. Proceedings of the 41st International Conference on Machine Learning , pages=

  17. [17]

    Verifying probabilistic forecasts: Calibration and sharpness

    Probabilistic. Journal of the Royal Statistical Society Series B: Statistical Methodology , author =. 2007 , pages =. doi:10.1111/j.1467-9868.2007.00587.x , number =

    work page doi:10.1111/j.1467-9868.2007.00587.x 2007

  18. [18]

    Journal of statistical software , volume=

    Stan: A probabilistic programming language , author=. Journal of statistical software , volume=

  19. [19]

    Advances in Neural Information Processing Systems , volume=

    Deep sets , author=. Advances in Neural Information Processing Systems , volume=

  20. [20]

    Flow Matching for Generative Modeling

    Flow matching for generative modeling , author=. arXiv preprint arXiv:2210.02747 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  21. [21]

    Advances in Neural Information Processing Systems , volume=

    Flow matching for scalable simulation-based inference , author=. Advances in Neural Information Processing Systems , volume=

  22. [22]

    Advances in Neural Information Processing Systems , volume=

    Consistency models for scalable and fast simulation-based inference , author=. Advances in Neural Information Processing Systems , volume=

  23. [23]

    Proceedings of the 40th International Conference on Machine Learning , pages=

    Consistency models , author=. Proceedings of the 40th International Conference on Machine Learning , pages=

  24. [24]

    Advances in Neural Information Processing Systems , volume=

    Flexible statistical inference for mechanistic models of neural dynamics , author=. Advances in Neural Information Processing Systems , volume=

  25. [25]

    Statistics and Computing , volume=

    Graphical test for discrete uniformity and its applications in goodness-of-fit evaluation and multiple sample comparison , author=. Statistics and Computing , volume=. 2022 , publisher=

    2022

  26. [26]

    arXiv preprint arXiv:2506.09258 , year=

    CFMI: Flow Matching for Missing Data Imputation , author=. arXiv preprint arXiv:2506.09258 , year=

    work page arXiv

  27. [27]

    Amortized

    Habermann, Daniel and Schmitt, Marvin and K. Amortized. Bayesian Analysis , volume=. 2025 , publisher=

    2025

  28. [28]

    American Journal of Epidemiology , volume=

    Estimating prevalence from the results of a screening test , author=. American Journal of Epidemiology , volume=. 1978 , publisher=

    1978

  29. [29]

    International conference on machine learning , pages=

    Set transformer: A framework for attention-based permutation-invariant neural networks , author=. International conference on machine learning , pages=. 2019 , organization=

    2019

  30. [30]

    Solving stochastic inverse problems with stochastic

    Zhang, Yi and Mikelsons, Lars , booktitle=. Solving stochastic inverse problems with stochastic. 2023 , organization=

    2023

  31. [31]

    Current Epidemiology Reports , volume=

    Are all biases missing data problems? , author=. Current Epidemiology Reports , volume=. 2015 , publisher=

    2015

  32. [32]

    BMC Medical Research Methodology , volume=

    Accounting for bias due to outcome data missing not at random: comparison and illustration of two approaches to probabilistic bias analysis: a simulation study , author=. BMC Medical Research Methodology , volume=. 2024 , publisher=

    2024

  33. [33]

    2025 , journal=

    Flow IV: Counterfactual Inference In Nonseparable Outcome Models Using Instrumental Variables , author=. 2025 , journal=

    2025

  34. [34]

    Advances in Neural Information Processing Systems , volume=

    Denoising diffusion probabilistic models , author=. Advances in Neural Information Processing Systems , volume=

  35. [35]

    Advances in Neural Information Processing Systems , volume=

    Generative modeling by estimating gradients of the data distribution , author=. Advances in Neural Information Processing Systems , volume=

  36. [36]

    arXiv e-prints , keywords =

    Diffusion Models in Simulation-Based Inference: A Tutorial Review , author=. arXiv preprint arXiv:2512.20685 , year=

    work page arXiv

  37. [37]

    European Journal of Epidemiology , volume=

    A multi-state model based reanalysis of the Framingham heart study , author=. European Journal of Epidemiology , volume=. 2019 , publisher=

    2019

  38. [38]

    The New England Journal of Medicine , author =

    Letter to ". The New England Journal of Medicine , author =. 2016 , keywords =. doi:10.1056/NEJMc1604823 , language =

    work page doi:10.1056/nejmc1604823 2016

  39. [39]

    Biometrical Journal , author =

    Estimating hazard ratios in cohort data with missing disease information due to death:. Biometrical Journal , author =. 2017 , keywords =. doi:10.1002/bimj.201500167 , language =

    work page doi:10.1002/bimj.201500167 2017

  40. [40]

    , volume =

    Missing information caused by death leads to bias in relative risk estimates. , volume =. Journal of Clinical Epidemiology , author =. 2014 , keywords =. doi:10.1016/j.jclinepi.2014.05.010 , language =

    work page doi:10.1016/j.jclinepi.2014.05.010 2014

  41. [41]

    International Journal of Epidemiology , author =

    Interval-censored time-to-event and competing risk with death: is the illness-death model more accurate than the. International Journal of Epidemiology , author =. 2013 , pages =. doi:10.1093/ije/dyt126 , language =

    work page doi:10.1093/ije/dyt126 2013

  42. [42]

    Biostatistics , volume=

    A penalized likelihood approach for an illness--death model with interval-censored data: application to age-specific incidence of dementia , author=. Biostatistics , volume=. 2002 , publisher=

    2002

  43. [43]

    Discriminative calibration: Check

    Yao, Yuling and Domke, Justin , journal=. Discriminative calibration: Check

  44. [44]

    Linhart, Julia and Gramfort, Alexandre and Rodrigues, Pedro , journal=

  45. [45]

    International Conference on Learning Representations , year=

    Revisiting Classifier Two-Sample Tests , author=. International Conference on Learning Representations , year=

  46. [46]

    New England Journal of Medicine , volume=

    Incidence of dementia over three decades in the Framingham Heart Study , author=. New England Journal of Medicine , volume=. 2016 , publisher=

    2016

  47. [47]

    Sensitivity-Aware Amortized

    Lasse Elsem. Sensitivity-Aware Amortized. Transactions on Machine Learning Research , issn=. 2024 , url=

    2024

  48. [48]

    Head-to-head evaluation of seven different seroassays including direct viral neutralisation in a representative cohort for

    Olbrich, Laura and Castelletti, Noemi and Schaelte, Yannik and Gari, Merce and Puetz, Peter and Bakuli, Abhishek and Pritsch, Michael and Kroidl, Inge and Saathoff, Elmar and Guggenbuehl Noller, Jessica Michelle and others , journal=. Head-to-head evaluation of seven different seroassays including direct viral neutralisation in a representative cohort for...

    2021

  49. [49]

    1994 , publisher=

    An introduction to the bootstrap , author=. 1994 , publisher=

    1994

  50. [50]

    International Conference on Learning Representations , year=

    Decoupled Weight Decay Regularization , author=. International Conference on Learning Representations , year=

  51. [51]

    2017 , url=

    Ilya Loshchilov and Frank Hutter , booktitle=. 2017 , url=

    2017

  52. [52]

    Fifth Symposium on Advances in Approximate

    Neural Score Estimation: Likelihood-Free Inference with Conditional Score Based Diffusion Models , author=. Fifth Symposium on Advances in Approximate. 2023 , url=

    2023

  53. [53]

    npj Systems Biology and Applications , year=

    Simulation-based inference of cell migration dynamics in complex spatial environments , author=. npj Systems Biology and Applications , year=

  54. [54]

    arXiv preprint arXiv:2502.06492 , year=

    An Overview and Recent Developments in the Analysis of Multistate Processes , author=. arXiv preprint arXiv:2502.06492 , year=

    work page arXiv

  55. [55]

    Artificial Intelligence Review , volume=

    Deep learning for survival analysis: a review , author=. Artificial Intelligence Review , volume=. 2024 , publisher=

    2024

  56. [56]

    Lars Kühmichel and Jerry M. Huang and Valentin Pratz and Jonas Arruda and Hans Olischläger and Daniel Habermann and Simon Kucharsky and Lasse Elsemüller and Aayush Mishra and Niels Bracher and Svenja Jedhoff and Marvin Schmitt and Paul-Christian Bürkner and Stefan T. Radev , year=

  57. [57]

    Current Epidemiology Reports , volume=

    Selection mechanisms and their consequences: understanding and addressing selection bias , author=. Current Epidemiology Reports , volume=. 2020 , publisher=

    2020

  58. [58]

    Clinical and Translational Science , volume=

    Big data and large sample size: a cautionary note on the potential for bias , author=. Clinical and Translational Science , volume=. 2014 , publisher=

    2014

  59. [59]

    BMJ Medicine , volume=

    Defining representativeness of study samples in medical and population health research , author=. BMJ Medicine , volume=

  60. [60]

    Nature Genetics , volume=

    Using large-scale population-based data to improve disease risk assessment of clinical variants , author=. Nature Genetics , volume=. 2025 , publisher=

    2025

  61. [61]

    Pharmacoepidemiology , pages=

    What is pharmacoepidemiology? , author=. Pharmacoepidemiology , pages=. 2019 , publisher=

    2019

  62. [62]

    Journal of the American statistical Association , volume=

    A generalization of sampling without replacement from a finite universe , author=. Journal of the American statistical Association , volume=. 1952 , publisher=

    1952

  63. [63]

    arXiv preprint arXiv:2602.11325 , year=

    Amortised and provably-robust simulation-based inference , author=. arXiv preprint arXiv:2602.11325 , year=

    work page arXiv

  64. [64]

    arXiv preprint arXiv:2602.09161 , year=

    Minimum Distance Summaries for Robust Neural Posterior Estimation , author=. arXiv preprint arXiv:2602.09161 , year=

    work page arXiv

  65. [65]

    Does Unsupervised Domain Adaptation Improve the Robustness of Amortized

    Lasse Elsem. Does Unsupervised Domain Adaptation Improve the Robustness of Amortized. Transactions on Machine Learning Research , issn=. 2025 , url=

    2025

  66. [66]

    arXiv preprint arXiv:2512.22999 , year=

    Bracher, Niels and K. arXiv preprint arXiv:2512.22999 , year=

    work page arXiv

  67. [67]

    arXiv preprint arXiv:2503.24011 , year=

    Simulations in statistical workflows , author=. arXiv preprint arXiv:2503.24011 , year=

    work page arXiv

  68. [68]

    Protocol of a population-based prospective

    Radon, Katja and Saathoff, Elmar and Pritsch, Michael and Guggenb. Protocol of a population-based prospective. BMC Public Health , volume=. 2020 , publisher=

    2020

  69. [69]

    Serial interval in households infected with

    an der Heiden, Matthias and Buchholz, Udo , year=. Serial interval in households infected with. doi:10.1017/S0950268822001248 , journal=

    work page doi:10.1017/s0950268822001248

  70. [70]

    China CDC Weekly , volume =

    Uncovering the Impact of Control Strategies on the Transmission Pattern of. China CDC Weekly , volume =. 2022 , issn =. doi:10.46234/ccdcw2022.208 , author =

    work page doi:10.46234/ccdcw2022.208 2022

  71. [71]

    2023 , issn =

    The Lancet Microbe , volume =. 2023 , issn =. doi:10.1016/S2666-5247(23)00005-8 , author =

    work page doi:10.1016/s2666-5247(23)00005-8 2023

  72. [72]

    Environmental Microbiology , volume =

    Saliva for molecular detection of. Environmental Microbiology , volume =. doi:10.1111/1462-2920.16151 , author =

    work page doi:10.1111/1462-2920.16151

  73. [73]

    Preschool-age children maintain a distinct memory

    Manfroi, Beno. Preschool-age children maintain a distinct memory. Science Translational Medicine , volume=. 2024 , publisher=

    2024

  74. [74]

    Accounting for Selection Effects in Supernova Cosmology with Simulation-Based Inference and Hierarchical

    Boyd, Benjamin M and Grayling, Matthew and Thorp, Stephen and Mandel, Kaisey S , journal=. Accounting for Selection Effects in Supernova Cosmology with Simulation-Based Inference and Hierarchical

  75. [75]

    Partially Exchangeable Networks and Architectures for Learning Summary Statistics in Approximate

    Wiqvist, Samuel and Mattei, Pierre-Alexandre and Picchini, Umberto and Frellsen, Jes , booktitle =. Partially Exchangeable Networks and Architectures for Learning Summary Statistics in Approximate. 2019 , bdsk-url-1 =

    2019

  76. [76]

    Informative and adaptive distances and summary statistics in sequential approximate

    Sch. Informative and adaptive distances and summary statistics in sequential approximate. Plos one , number =

  77. [77]

    An amortized approach to non-linear mixed-effects modeling based on neural posterior estimation , url =

    Arruda, Jonas and Sch\". An amortized approach to non-linear mixed-effects modeling based on neural posterior estimation , url =. Proceedings of the 41st International Conference on Machine Learning , date-added =. 2024 , bdsk-url-1 =

    2024

  78. [78]

    Missing data in amortized simulation-based neural posterior estimation , volume =

    Zijian Wang and Jan Hasenauer and Yannik Sch. Missing data in amortized simulation-based neural posterior estimation , volume =. 2024 , bdsk-url-1 =. doi:10.1371/journal.pcbi.1012184 , journal =

    work page doi:10.1371/journal.pcbi.1012184 2024

  79. [79]

    arXiv preprint arXiv:1811.08723 , title =

    Durkan, Conor and Papamakarios, George and Murray, Iain , date-added =. arXiv preprint arXiv:1811.08723 , title =

    work page arXiv

  80. [80]

    Flow Matching for Scalable Simulation-Based Inference , volume =

    Dax, Maximilian and Wildberger, Jonas and Buchholz, Simon and Green, Stephen R and Macke, Jakob H and Sch. Flow Matching for Scalable Simulation-Based Inference , volume =. Advances in Neural Information Processing Systems 36 (NeurIPS 2023) , date-added =

    2023

Showing first 80 references.