Frugal, Flexible, Faithful: Causal Data Simulation via Frengression

2025-08-01

AI Summary

Robust benchmarking of causal inference methods on real-world data has long been hindered by the scarcity of high-fidelity, controllable interventional simulation datasets. To address this, we propose Frengression—a novel framework that integrates parsimonious parameterization with deep generative modeling to directly learn the joint distribution over covariates, interventions, and outcomes. This enables precise, consistent, and extrapolation-guaranteed estimation of causal marginals. Frengression supports high-fidelity generation of multivariate time-series data and permits direct sampling under arbitrary intervention distributions, substantially improving simulation flexibility and controllability. Empirical evaluation on real clinical trial data demonstrates that Frengression-synthesized data yield accurate causal effect estimates. The framework thus significantly enhances the practicality, scalability, and reproducibility of causal simulation studies.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Web Mining and Content Analysis: Web data generation and simulation
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Machine learning has revitalized causal inference by combining flexible models and principled estimators, yet robust benchmarking and evaluation remain challenging with real-world data. In this work, we introduce frengression, a deep generative realization of the frugal parameterization that models the joint distribution of covariates, treatments and outcomes around the causal margin of interest. Frengression provides accurate estimation and flexible, faithful simulation of multivariate, time-varying data; it also enables direct sampling from user-specified interventional distributions. Model consistency and extrapolation guarantees are established, with validation on real-world clinical trial data demonstrating frengression's practical utility. We envision this framework sparking new research into generative approaches for causal margin modelling.

Problem

• Robust benchmarking and evaluation in causal inference with real-world data
• Flexible and faithful simulation of multivariate, time-varying data
• Direct sampling from user-specified interventional distributions

Innovation

• Deep generative model for causal data simulation
• Flexible faithful multivariate time-varying data simulation
• Direct sampling from interventional distributions