ConDiSim: Conditional Diffusion Models for Simulation Based Inference
This provides a robust framework for parameter inference in complex systems, particularly benefiting workflows requiring fast inference methods, though it appears incremental as it builds on existing diffusion model techniques.
The authors tackled the problem of simulation-based inference for complex systems with intractable likelihoods by developing ConDiSim, a conditional diffusion model that approximates posterior distributions, and demonstrated its effectiveness across multiple benchmarks and real-world problems with improved accuracy and computational efficiency.
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.