ConDiSim: Conditional Diffusion Models for Simulation Based Inference

TL;DR

ConDiSim introduces a conditional diffusion model for efficient and accurate simulation-based inference of complex, multi-modal posterior distributions in systems with intractable likelihoods, demonstrating strong performance on benchmarks and real-world problems.

Contribution

It proposes a novel diffusion-based framework, ConDiSim, for posterior approximation in simulation-based inference, capturing complex dependencies and multi-modality effectively.

Findings

Achieves high accuracy in posterior approximation across benchmark problems.

Maintains computational efficiency and training stability.

Demonstrates robustness on real-world inference tasks.

Abstract

We present a conditional diffusion model - ConDiSim, for simulation-based inference of complex systems with intractable likelihoods. ConDiSim leverages denoising diffusion probabilistic models to approximate posterior distributions, consisting of a forward process that adds Gaussian noise to parameters, and a reverse process learning to denoise, conditioned on observed data. This approach effectively captures complex dependencies and multi-modalities within posteriors. ConDiSim is evaluated across ten benchmark problems and two real-world test problems, where it demonstrates effective posterior approximation accuracy while maintaining computational efficiency and stability in model training. ConDiSim offers a robust and extensible framework for simulation-based inference, particularly suitable for parameter inference workflows requiring fast inference methods.