Counterfactual Probabilistic Diffusion with Expert Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Mu, Wenhao, Cao, Zhi, Uludag, Mehmed, Rodríguez, Alexander
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912584426323968
author Mu, Wenhao
Cao, Zhi
Uludag, Mehmed
Rodríguez, Alexander
author_facet Mu, Wenhao
Cao, Zhi
Uludag, Mehmed
Rodríguez, Alexander
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2508_13355
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Counterfactual Probabilistic Diffusion with Expert Models
Mu, Wenhao
Cao, Zhi
Uludag, Mehmed
Rodríguez, Alexander
Machine Learning
Artificial Intelligence
Methodology
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.
title Counterfactual Probabilistic Diffusion with Expert Models
topic Machine Learning
Artificial Intelligence
Methodology
url https://arxiv.org/abs/2508.13355