Aggregated Posterior Predictive Checks for Generative Modeling

Abstract

Latent variable generative models are commonly fit using simple priors over latent variables, but draws from these priors often fail to produce realistic data. This failure is due to a mismatch between the prior and the aggregated posterior, the distribution of latent variables induced by the fitted model and the data. This mismatch is often viewed as evidence that the prior is misspecified and should be replaced. Alternatively, in modern generative models, a two-stage strategy is increasingly used where first, the model is fit, and second, the aggregated posterior is estimated (van den Oord et al.,2017; Rombach et al., 2022.). Synthetic data are then obtained by sampling from this aggregated posterior instead of the prior. To check such procedures, we introduce the aggregated posterior predictive check (APPC). Theoretically, we establish sufficient conditions under which the APPC is asymptotically calibrated. For probabilistic principal component analysis, we show that the APPC can remain calibrated under a misspecified latent prior when pervasive factors permit recovery of the signal space. Experiments with variational autoencoders show that aggregated posterior sampling improves generation for heavy-tailed and clustered data relative to Gaussian prior sampling while performing comparably to models with more flexible latent priors.

Generative sequence modeling for infinite memory processes via predictive states
Michael Wieck-Sosa, Cosma Rohilla Shalizi
arXiv29 Sep 2026PreprintarXiv licenceFull textStatisticsarXiv
Score-Based Generative Data Assimilation for Integrating Aggregated Surveillance Data into Agent-Based Models in Epidemic Tracking
Siming Liang, Jacob Hauck, Minglei Yang +3
arXiv1 Sep 2026PreprintarXiv licenceFull textBioinformatics & computational biology Numerical analysis & scientific computingarXiv
SCOPE Before You Sample: Conformal Scores Guide Posterior Exploration for Fast and Calibrated Bayesian Predictive Inference
Zahra Moslemi, Ziyi Liang, Babak Shahbaba
arXiv12 Oct 2026PreprintarXiv licenceFull textStatisticsarXiv
See You at the Posterior Line: Learning Bayesian Modeling Through a Car Racing Game
Federica Zoe Ricci, Mine Dogucu
arXiv19 Sep 2026PreprintCC BY-NC-ND 4.0Full textStatisticsarXiv
Contextual Decision Making: Predictive or Generative
L. Jeff Hong, Weihuan Huang, Xinyao Li
arXiv23 Oct 2026PreprintarXiv licenceFull textOptimisation & control StatisticsarXiv

Cite

BibTeX RIS

BibTeX

@misc{dutta2026aggregated,
  title = {{Aggregated Posterior Predictive Checks for Generative Modeling}},
  author = {Shweta Dutta and Gemma E. Moran},
  howpublished = {arXiv},
  year = {2026},
  month = sep,
  doi = {10.48550/arxiv.2609.20999},
  eprint = {2609.20999},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.20999},
}

RIS

TY  - UNPB
TI  - Aggregated Posterior Predictive Checks for Generative Modeling
AU  - Shweta Dutta
AU  - Gemma E. Moran
PY  - 2026
DA  - 2026/09/17
PB  - arXiv
DO  - 10.48550/arxiv.2609.20999
UR  - https://arxiv.org/abs/2609.20999
AB  - Latent variable generative models are commonly fit using simple priors over latent variables, but draws from these priors often fail to produce realistic data. This failure is due to a mismatch between the prior and the aggregated posterior, the distribution of latent variables induced by the fitted model and the data. This mismatch is often viewed as evidence that the prior is misspecified and should be replaced. Alternatively, in modern generative models, a two-stage strategy is increasingly used where first, the model is fit, and second, the aggregated posterior is estimated (van den Oord et al.,2017; Rombach et al., 2022.). Synthetic data are then obtained by sampling from this aggregated posterior instead of the prior. To check such procedures, we introduce the aggregated posterior predictive check (APPC). Theoretically, we establish sufficient conditions under which the APPC is asymptotically calibrated. For probabilistic principal component analysis, we show that the APPC can remain calibrated under a misspecified latent prior when pervasive factors permit recovery of the signal space. Experiments with variational autoencoders show that aggregated posterior sampling improves generation for heavy-tailed and clustered data relative to Gaussian prior sampling while performing comparably to models with more flexible latent priors.
ER  -

Abstract

Related

Cite