Automatic, Debiased, and Invariant Counterfactual Generation under General Interventions

2026-06-05

AI Summary

Existing counterfactual generation methods are highly sensitive to model misspecification under complex interventions, resulting in poor generalization and unstable estimates. This work proposes ADIGen, a novel framework that uniquely integrates Riesz regression, causal invariance, and orthogonal statistical learning to enable automatic, debiased, and environment-invariant counterfactual generation in high-dimensional intervention and outcome settings. By leveraging Riesz regression, ADIGen circumvents the instability inherent in density ratio estimation; causal invariance enhances robustness to distributional shifts; and orthogonal learning confers double robustness against nuisance model misspecification. Theoretical analysis establishes an excess risk bound with a product-form remainder term, demonstrating that ADIGen effectively controls counterfactual risk under general interventions.

Abstract

Generative models for counterfactual outcomes have great potential to support decision-making under complex interventions, but existing approaches are limited by unstable estimation, poor generalization across environments, and bias from nuisance model misspecification. We introduce ADIGen, a framework for automatic, debiased, and invariant counterfactual generation under general interventions, including high-dimensional interventions and outcomes. ADIGen combines Riesz regression to avoid unstable density-ratio estimation, causal invariance to improve generalization under distribution shift, and orthogonal statistical learning to obtain doubly robust guarantees against nuisance model misspecification. We provide excess-risk bounds showing that ADIGen controls counterfactual risk under general interventions, with a product-bias nuisance remainder and an invariant risk bound across environments.

Problem

• counterfactual generation
• general interventions
• distribution shift
• nuisance model misspecification
• estimation instability

Innovation

• counterfactual generation
• causal invariance
• Riesz regression
• orthogonal learning
• distribution shift