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

2025-10-15

AI Summary

Existing amortized inference methods rely on fixed priors during training, limiting adaptability to novel prior distributions at test time. This work introduces a diffusion-based framework for test-time prior adaptation, enabling dynamic alignment with arbitrary new priors without retraining or additional simulations. Our approach comprises three key innovations: (1) the first integration of diffusion models into amortized Bayesian inference, yielding prior-agnostic pretraining; (2) a novel guided approximation mechanism that allows users to incorporate domain-specific prior knowledge or constraints; and (3) lightweight test-time optimization to ensure high-fidelity posterior sampling. We validate the framework across diverse scientific computing tasks, demonstrating substantial improvements in posterior estimation accuracy and generalization. The method provides a flexible, efficient inference tool for real-world applications—including engineering modeling and computational neuroscience—where priors are uncertain or evolve post-deployment.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Probabilistic Inference

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

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.

Problem

• Adapting simulation-based inference to new prior distributions
• Enabling test-time prior updates without model retraining
• Enhancing pre-trained models' versatility with expert knowledge

Innovation

• Test-time prior adaptation for simulation-based inference
• Guidance approximation enables flexible prior adaptation
• No retraining required for new prior distributions