Mutual Information Minimization Based Disentangled Learning Framework For Causal Effect Estimation

Learning treatment effect from observational data is a fundamental problem in causal inference. Recently, disentangled representation learning methods, such as DR-CFR and DeR-CFR, have witnessed great success in treatment effect estimation, which aim to decompose covariates into three disjoint factors. However, we argue that these methods cannot identify underlying factors well, as they cannot obtain independent disentangled factors. Inspired by the success of mutual information minimization in disentangled representation learning, we propose a novel method called MimCE in this paper: Mutual Information Minimization based Disentangled Learning Framework for Causal Effect Estimation. MimCE mainly focuses on obtaining independent disentangled factors for treatment effect estimation and numerous experiments demonstrate that it performs better than the state-of-the-art methods both on the predictive performance and model stability.