Two-way Guidance: Simultaneous Learning of Summary Statistics for Generative Bayes
Abstract
Bayesian simulation-based inference (SBI) can struggle to recover increasingly concentrated posteriors as observation size grows. We consider learning sufficient summary statistics jointly with the posterior as a general principle for scalable inference in various SBI methods. A common measure taken in SBI is dimension reduction through predetermined summaries, which may partially discard posterior information. Our approach instead imposes structure on how the posterior learner depends on a learned summary (e.g., affinity), and learns the summary simultaneously with the posterior. We identify an important mechanism named two-way guidance: under a structural restriction on the summary, the posterior sufficiency of the learned summary simplifies posterior learning, whereas a suitable posterior objective guides the learned summary towards posterior sufficiency. This mechanism explains a counterintuitive finding that the structural restriction on the summary statistics improves posterior learning, rather than limit its performance. We establish theoretical guarantees on both posterior estimation error and the learned summary's departure from sufficiency, thus identifying conditions on the simulation budget for recovering increasingly concentrated posteriors as observation size grows.
Related
- Neural Estimation of Summary Statistics for Simulation-Based Inference with Non-Asymptotic Guarantees
- Interacting particle guidance for sampling reward-tilted generative priors
- Multi-Agent Flow Matching with Decoupled Generative Guidance
- A Framework for Teaching the Responsible and Effective Use of Generative AI in Statistics and Data Science
- Asymptotically Exact Negative Guidance of Diffusion Models via Positive-Unlabeled Learning
BibTeX
@misc{zhai2026twoway,
title = {{Two-way Guidance: Simultaneous Learning of Summary Statistics for Generative Bayes}},
author = {Percy S. Zhai and Max Pachman and Peverel D. Shipley and Jungeum Kim},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.12850},
eprint = {2610.12850},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.12850},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Two-way Guidance: Simultaneous Learning of Summary Statistics for Generative Bayes
AU - Percy S. Zhai
AU - Max Pachman
AU - Peverel D. Shipley
AU - Jungeum Kim
PY - 2026
DA - 2026/10/09
PB - arXiv
DO - 10.48550/arxiv.2610.12850
UR - https://arxiv.org/abs/2610.12850
AB - Bayesian simulation-based inference (SBI) can struggle to recover increasingly concentrated posteriors as observation size grows. We consider learning sufficient summary statistics jointly with the posterior as a general principle for scalable inference in various SBI methods. A common measure taken in SBI is dimension reduction through predetermined summaries, which may partially discard posterior information. Our approach instead imposes structure on how the posterior learner depends on a learned summary (e.g., affinity), and learns the summary simultaneously with the posterior. We identify an important mechanism named two-way guidance: under a structural restriction on the summary, the posterior sufficiency of the learned summary simplifies posterior learning, whereas a suitable posterior objective guides the learned summary towards posterior sufficiency. This mechanism explains a counterintuitive finding that the structural restriction on the summary statistics improves posterior learning, rather than limit its performance. We establish theoretical guarantees on both posterior estimation error and the learned summary's departure from sufficiency, thus identifying conditions on the simulation budget for recovering increasingly concentrated posteriors as observation size grows.
ER -