High-Dimensional Simulation-Based Inference in Latent Spaces
Abstract
Neural simulation-based inference (SBI) has been widely successful in inferring a relatively small number of interpretable parameters from potentially high-dimensional observations, such as images or time series. Accordingly, representation learning in SBI has focused almost exclusively on compressing the observations used to condition the posterior. More recently, however, SBI has begun to target increasingly high-dimensional parameter spaces, raising the complementary question of whether the inference target itself should be compressed. Our answer is a practical merger of SBI and latent generative modeling, which learns a low-dimensional representation of the simulator parameters, performs posterior inference directly in this latent space, and maps posterior samples back to the original parameter space. We characterize the conditions under which latent-space inference recovers the desired target posterior and systematically study its empirical trade-offs. Across four case studies and three generative families, we compare latent and standard estimators while controlling for network capacity, regularization, optimization, and training compute. At matched training compute, latent-space inference achieves accuracy and marginal calibration comparable to direct target-space inference while sampling up to more than an order of magnitude faster.
Related
- Causal Mediation Analysis with High-Dimensional Surrogates: Identification, Estimation and Inference under Latent Factor Models
- Selective Inference for Deep Clustering in Latent Spaces
- A reproducible simulation dataset and benchmark for parametric competing-risks inference under censoring and latent dependence.
- Likelihood-Free Inference in High-Dimensional Models
- Observation-Specific Uncertainty Modeling for Latent-Variable Regression in High-Dimensional Analytical Data
BibTeX
@misc{kuhmichel2026highdimensional,
title = {{High-Dimensional Simulation-Based Inference in Latent Spaces}},
author = {Lars Kühmichel and Stefan T. Radev and Bhanu Prasanna Koppolu and Masoumeh Davoudi and Jerry M. Huang and Paul-Christian Bürkner},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.37381},
eprint = {2609.37381},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.37381},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - High-Dimensional Simulation-Based Inference in Latent Spaces
AU - Lars Kühmichel
AU - Stefan T. Radev
AU - Bhanu Prasanna Koppolu
AU - Masoumeh Davoudi
AU - Jerry M. Huang
AU - Paul-Christian Bürkner
PY - 2026
DA - 2026/09/29
PB - arXiv
DO - 10.48550/arxiv.2609.37381
UR - https://arxiv.org/abs/2609.37381
AB - Neural simulation-based inference (SBI) has been widely successful in inferring a relatively small number of interpretable parameters from potentially high-dimensional observations, such as images or time series. Accordingly, representation learning in SBI has focused almost exclusively on compressing the observations used to condition the posterior. More recently, however, SBI has begun to target increasingly high-dimensional parameter spaces, raising the complementary question of whether the inference target itself should be compressed. Our answer is a practical merger of SBI and latent generative modeling, which learns a low-dimensional representation of the simulator parameters, performs posterior inference directly in this latent space, and maps posterior samples back to the original parameter space. We characterize the conditions under which latent-space inference recovers the desired target posterior and systematically study its empirical trade-offs. Across four case studies and three generative families, we compare latent and standard estimators while controlling for network capacity, regularization, optimization, and training compute. At matched training compute, latent-space inference achieves accuracy and marginal calibration comparable to direct target-space inference while sampling up to more than an order of magnitude faster.
ER -