Counterfactual Probabilistic Diffusion with Expert Models
This work addresses the challenge of reliable and interpretable causal inference under data scarcity for domains like public health and medicine, representing an incremental improvement by bridging mechanistic and data-driven approaches.
The paper tackles the problem of predicting counterfactual distributions in complex dynamical systems, such as public health and medicine, by proposing ODE-Diff, a time series diffusion-based framework that incorporates expert models as structured priors, and it demonstrates consistent outperformance over baselines in point prediction and distributional accuracy across COVID-19 simulations, pharmacological dynamics, and real-world cases.
Predicting counterfactual distributions in complex dynamical systems is essential for scientific modeling and decision-making in domains such as public health and medicine. However, existing methods often rely on point estimates or purely data-driven models, which tend to falter under data scarcity. We propose a time series diffusion-based framework that incorporates guidance from imperfect expert models by extracting high-level signals to serve as structured priors for generative modeling. Our method, ODE-Diff, bridges mechanistic and data-driven approaches, enabling more reliable and interpretable causal inference. We evaluate ODE-Diff across semi-synthetic COVID-19 simulations, synthetic pharmacological dynamics, and real-world case studies, demonstrating that it consistently outperforms strong baselines in both point prediction and distributional accuracy.