GDR-learners: Orthogonal Learning of Generative Models for Potential Outcomes

2025-09-26

AI Summary

Existing deep generative models lack Neyman orthogonality in estimating conditional potential outcome distributions, thus failing to guarantee quasi-oracle efficiency and double robustness. To address this, we propose the GDR-learners framework—the first systematic integration of Neyman orthogonality into generative modeling. GDR-learners supports diverse architectures, including conditional normalizing flows (GDR-CNFs), conditional GANs (GDR-CGANs), conditional VAEs (GDR-CVAEs), and conditional diffusion models (GDR-CDMs). It enables doubly robust estimation with respect to both treatment mechanism and outcome model, achieving asymptotically optimal convergence rates. We establish theoretical guarantees of quasi-oracle efficiency under standard regularity conditions. Empirically, GDR-learners significantly outperforms state-of-the-art methods across multiple semi-synthetic benchmarks, delivering accurate, high-fidelity characterization of conditional potential outcome distributions.

Technology Category

• Machine Learning: Causal Learning
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Various deep generative models have been proposed to estimate potential outcomes distributions from observational data. However, none of them have the favorable theoretical property of general Neyman-orthogonality and, associated with it, quasi-oracle efficiency and double robustness. In this paper, we introduce a general suite of generative Neyman-orthogonal (doubly-robust) learners that estimate the conditional distributions of potential outcomes. Our proposed GDR-learners are flexible and can be instantiated with many state-of-the-art deep generative models. In particular, we develop GDR-learners based on (a) conditional normalizing flows (which we call GDR-CNFs), (b) conditional generative adversarial networks (GDR-CGANs), (c) conditional variational autoencoders (GDR-CVAEs), and (d) conditional diffusion models (GDR-CDMs). Unlike the existing methods, our GDR-learners possess the properties of quasi-oracle efficiency and rate double robustness, and are thus asymptotically optimal. In a series of (semi-)synthetic experiments, we demonstrate that our GDR-learners are very effective and outperform the existing methods in estimating the conditional distributions of potential outcomes.

Problem

• Estimating potential outcomes distributions from observational data
• Achieving Neyman-orthogonality for quasi-oracle efficiency
• Developing doubly-robust generative models for causal inference

Innovation

• Orthogonal learning for generative potential outcomes models
• Neyman-orthogonal doubly-robust learners framework
• Flexible integration with multiple deep generative architectures