PriorGuide: Test-Time Prior Adaptation for Simulation-Based Inference

ICLR 2026 Conference Submission13405 Authors

Keywords: Simulation-based inference, Amortized inference, Test-time adaptation, Bayesian workflow, Neural posterior estimation, Diffusion models

TL;DR: PriorGuide enables efficient incorporation of arbitrary priors at inference time for amortized diffusion-based simulation-based inference, without retraining the model.

Abstract: Amortized simulator-based inference offers a powerful framework for tackling Bayesian inference in computational fields such as engineering or neuroscience, increasingly leveraging modern generative methods like diffusion models to map observed data to model parameters or future predictions. These approaches yield posterior or posterior-predictive samples for new datasets without requiring further simulator calls after training on simulated parameter-data pairs. However, their applicability is often limited by the prior distribution(s) used to generate model parameters during this training phase. To overcome this constraint, we introduce *PriorGuide*, a technique specifically designed for diffusion-based amortized inference methods. PriorGuide leverages a novel guidance approximation that enables flexible adaptation of the trained diffusion model to new priors at test time, crucially without costly retraining. This allows users to readily incorporate updated information or expert knowledge post-training, enhancing the versatility of pre-trained inference models.

Primary Area: probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)

Submission Number: 13405