Conditional Mean Flows: Few-Step Densities for Simulation-Based Inference
Abstract
Flow matching has emerged as a powerful generative family for simulation-based inference (SBI). However, sampling and density evaluation require numerical integration at inference time. This makes posterior sampling in neural posterior estimation costly. More importantly, it can make repeated likelihood evaluation prohibitively expensive and impractical for Markov chain Monte Carlo (MCMC) sampling, which may require thousands of repeated density evaluations. Here, we develop conditional mean flows for posterior and likelihood approximation, combining few-step sampling with tractable density evaluation. We show that the mean flow objective identifies conditional flow maps and that independent masking enables direct estimation of marginal likelihoods. We evaluate both posterior and likelihood estimation across several tasks and show that conditional mean flows retain the flexibility of flow-based density estimators while significantly reducing inference time. In particular, single-step conditional mean flow density evaluation yields estimation error comparable to 200-step flow matching while enabling about 200-fold faster downstream MCMC sampling.
Related
- A perspective note on likelihood approximation and inference for complex simulation models using a chain of aggregated normalizing flows
- Adaptive Bayesian Estimation of Conditional Densities Based on Orthogonal Polynomials
- Fitting Dynamic Conditional Densities to Cross-Sectional Returns with Persistent Tail Intensities
- Trend-concordance grey incidence with likelinesses closed-form small-sample inference without elicited prior densities
- Compact Mean Curvature Flows from Non-Compact Initial Data
BibTeX
@misc{arruda2026conditional,
title = {{Conditional Mean Flows: Few-Step Densities for Simulation-Based Inference}},
author = {Jonas Arruda and Jan Hasenauer and Stefan T. Radev},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.21657},
eprint = {2610.21657},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.21657},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Conditional Mean Flows: Few-Step Densities for Simulation-Based Inference
AU - Jonas Arruda
AU - Jan Hasenauer
AU - Stefan T. Radev
PY - 2026
DA - 2026/10/15
PB - arXiv
DO - 10.48550/arxiv.2610.21657
UR - https://arxiv.org/abs/2610.21657
AB - Flow matching has emerged as a powerful generative family for simulation-based inference (SBI). However, sampling and density evaluation require numerical integration at inference time. This makes posterior sampling in neural posterior estimation costly. More importantly, it can make repeated likelihood evaluation prohibitively expensive and impractical for Markov chain Monte Carlo (MCMC) sampling, which may require thousands of repeated density evaluations. Here, we develop conditional mean flows for posterior and likelihood approximation, combining few-step sampling with tractable density evaluation. We show that the mean flow objective identifies conditional flow maps and that independent masking enables direct estimation of marginal likelihoods. We evaluate both posterior and likelihood estimation across several tasks and show that conditional mean flows retain the flexibility of flow-based density estimators while significantly reducing inference time. In particular, single-step conditional mean flow density evaluation yields estimation error comparable to 200-step flow matching while enabling about 200-fold faster downstream MCMC sampling.
ER -