mlsanityData to 7 Oct 2026

Debiased Counterfactual Generation via Flow Matching from Observations

arXiv 2605.07665 ↗First posted 8 May 2026stat.ML cs.LGcited by 1

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

Estimating counterfactual distributions under interventions is central to treatment risk assessment and counterfactual generation tasks. Existing approaches model the counterfactual distribution as a standalone generative target, without exploiting its relationship to the observational data. In this work, we show that under standard assumptions, observational and counterfactual outcome distributions are tightly linked: they have identical support and tail behavior, remain statistically close under weak confounding, and share any features of high-dimensional outcomes which are invariant to confounders. These properties motivate learning counterfactual distributions not from scratch, but via a deconfounding flow from the observational distribution. We formulate this problem via flow-matching and derive a semiparametrically efficient estimator based on a novel efficient influence function correction. We subsequently extend our estimator to target minimal-energy flows in high-dimensions, which we show can be especially simple targets between observational and counterfactual distributions. In experiments, deconfounding flows outperform existing debiased counterfactual distribution estimators, while also mitigating known failure modes of flow-based methods.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models. Builds on 22 of its 33 references that are in the map. First posted on arXiv 8 May 2026.

Builds on

22 of 33 references.

Closest ancestors · 1

Same direction, newest first

  1. 2025DoubleGen: Debiased Generative Modeling of Counterfactuals20 Sep

Other references · 21

  1. 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,604
  2. 2022Flow Matching for Generative Modelingcited by 5,426
  3. 2017FiLM: Visual Reasoning with a General Conditioning Layercited by 2,286
  4. 2023Improving and generalizing flow-based generative models with minibatch optimal transportcited by 835
  5. 2017Causal Effect Inference with Deep Latent-Variable Modelscited by 408
  6. 2022Easily Accessible Text-to-Image Generation Amplifies Demographic Stereotypes at Large Scalecited by 262
  7. 2020Deep Structural Causal Models for Tractable Counterfactual Inferencecited by 210
  8. 2018FairGAN: Fairness-aware Generative Adversarial Networkscited by 164
  9. 2017CausalGAN: Learning Causal Implicit Generative Models with Adversarial Trainingcited by 152
  10. 2022Semiparametric doubly robust targeted double machine learning: a reviewcited by 129
  11. 2019Fair Generative Modeling via Weak Supervisioncited by 114
  12. 2017Automated versus do-it-yourself methods for causal inference: Lessons learned from a data analysis competitioncited by 106
9 more
  1. 2022Diffusion Causal Models for Counterfactual Estimationcited by 87
  2. 2023Causal normalizing flows: from theory to practicecited by 49
  3. 2021Semiparametric counterfactual density estimationcited by 42
  4. 2022Normalizing Flows for Interventional Density Estimationcited by 21
  5. 2021VACA: Design of Variational Graph Autoencoders for Interventional and Counterfactual Queriescited by 18
  6. 2018Causal effects based on distributional distancescited by 15
  7. 2025Mirror Flow Matching with Heavy-Tailed Priors for Generative Modeling on Convex Domainscited by 8
  8. 2023Counterfactual Density Estimation using Kernel Stein Discrepanciescited by 7
  9. 2024Counterfactual Cocycles: A Framework for Robust and Coherent Counterfactual Transportscited by 6

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Counterfactual Generation via Flow Matching: Coupling-Sensitive End-to-End Ratescited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2023Counterfactual Generative Models for Time-Varying Treatments
  2. 2025Flow-based Generative Modeling of Potential Outcomes and Counterfactuals
  3. 2026Are Good Generators Good Decision-Makers? Policy Learning for General Interventions via Retargeted Counterfactual Generation
  4. 2025Flow IV: Counterfactual Inference In Nonseparable Outcome Models Using Instrumental Variables
  5. 2024DiffusionCounterfactuals: Inferring High-dimensional Counterfactuals with Guidance of Causal Representations
  6. 2025Counterfactual Identifiability via Dynamic Optimal Transport
  7. 2024Ablation Based Counterfactuals
  8. 2026Causal Diffusion Models for Counterfactual Outcome Distributions in Longitudinal Data