Counterfactual Probabilistic Diffusion with Expert Models

2025-08-18

AI Summary

Addressing the challenge of counterfactual distribution forecasting under data scarcity in complex dynamical systems, this paper proposes ODE-Diff: a framework that incorporates an imperfect mechanistic expert model as a structured prior into a temporal diffusion generative process. By integrating ordinary differential equation (ODE) modeling with probability-guided inference, ODE-Diff enables causal-aware counterfactual reasoning. The method synergistically unifies mechanism-driven and data-driven paradigms, preserving interpretability while enhancing few-shot generalization. Evaluated on COVID-19 transmission simulation, pharmacokinetic modeling, and real-world clinical time-series data, ODE-Diff consistently outperforms mainstream point-estimation and purely data-driven baselines—achieving superior accuracy in both point prediction and full-distribution forecasting. This work establishes a reliable, interpretable counterfactual modeling paradigm for decision support in public health and precision medicine.

Technology Category

• Reasoning under Uncertainty: Causality
• Computer Vision: Diffusion Models for Vision
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

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.

Problem

• Predicting counterfactual distributions in complex dynamical systems
• Overcoming data scarcity limitations in scientific modeling
• Bridging mechanistic and data-driven approaches for causal inference

Innovation

• Diffusion-based framework with expert guidance
• Bridges mechanistic and data-driven approaches
• Extracts high-level signals as structured priors