DoubleGen: Debiased Generative Modeling of Counterfactuals

2025-09-20

AI Summary

Counterfactual generation models are commonly afflicted by confounding bias—arising from unaccounted systematic differences between treatment and control groups—and misspecification bias due to inaccurate auxiliary models. This paper introduces DoubleGen, the first framework to integrate double robustness into generative modeling: it theoretically guarantees confounding bias elimination provided either the propensity score model or the outcome model is correctly specified. By reformulating training objectives for diffusion models, flow matching, and autoregressive language models, DoubleGen enables debiased counterfactual generation. The method offers finite-sample theoretical guarantees, achieving both oracle-optimal and minimax-optimal convergence rates. Empirical evaluations across diverse generative tasks demonstrate that DoubleGen significantly reduces estimation bias and enhances the robustness and accuracy of causal counterfactual predictions.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Causal Learning
• Computer Vision: Generative Adversarial Networks (GANs) for Vision

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
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Generative models for counterfactual outcomes face two key sources of bias. Confounding bias arises when approaches fail to account for systematic differences between those who receive the intervention and those who do not. Misspecification bias arises when methods attempt to address confounding through estimation of an auxiliary model, but specify it incorrectly. We introduce DoubleGen, a doubly robust framework that modifies generative modeling training objectives to mitigate these biases. The new objectives rely on two auxiliaries -- a propensity and outcome model -- and successfully address confounding bias even if only one of them is correct. We provide finite-sample guarantees for this robustness property. We further establish conditions under which DoubleGen achieves oracle optimality -- matching the convergence rates standard approaches would enjoy if interventional data were available -- and minimax rate optimality. We illustrate DoubleGen with three examples: diffusion models, flow matching, and autoregressive language models.

Problem

• Addresses confounding bias from systematic intervention group differences
• Mitigates misspecification bias from incorrect auxiliary model estimation
• Provides debiased generative modeling of counterfactual outcomes

Innovation

• Doubly robust framework modifies generative training objectives
• Uses propensity and outcome models to mitigate bias
• Achieves oracle optimality under specified conditions