Abstract
We study how to extend the use of the diffusion model to answer the causalquestion from the observational data under the existence of unmeasuredconfounders. In Pearl's framework of using a Directed Acyclic Graph (DAG) tocapture the causal intervention, a Diffusion-based Causal Model (DCM) wasproposed incorporating the diffusion model to answer the causal questions moreaccurately, assuming that all of the confounders are observed. However,unmeasured confounders in practice exist, which hinders DCM from beingapplicable. To alleviate this limitation of DCM, we propose an extended modelcalled Backdoor Criterion based DCM (BDCM), whose idea is rooted in theBackdoor criterion to find the variables in DAG to be included in the decodingprocess of the diffusion model so that we can extend DCM to the case withunmeasured confounders. Synthetic data experiment demonstrates that ourproposed model captures the counterfactual distribution more precisely than DCMunder the unmeasured confounders.