AI RESEARCH

Debiased Counterfactual Generation via Flow Matching from Observations

arXiv CS.LG

ArXi:2605.07665v1 Announce Type: cross 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.