文献类型：期刊文献

英文题名：Denoising for a Balanced Representation in Treatment Effect Estimation

作者：Yang, Hai[1]; Yao, Jing[1]; Wang, Zhe[1]; Yang, Yijing[2]

机构：[1] Department of Computer Science and Engineering, East China University of Science and Technology, Shanghai, 200237, China; [2] Department of Computer Science, University of Illinois Urbana-Champaign, Champaign, 61820, United States

年份：2024

外文期刊名：SSRN

收录：EI(收录号：20240042782)

语种：英文

外文关键词：Machine learning

摘要：Estimating individual treatment effects with observational data is of great importance in various domains. However, it faces two crucial challenges, including the absence of counterfactual outcomes and selection bias due to confounders. In this paper, we focus on the fundamental reasons for the divergence between the treated and control groups and try to quantify it from a generative perspective. Then we propose a method called Denoising for a Balanced Representation in Treatment Effect Estimation (DBRT). Motivated by the reverse denoising process to generate similar samples in the diffusion probabilistic model, we consider the bias between the treated and control groups as noise. Then we construct a specific network to denoise it progressively to achieve balanced representations in the latent space. In order to enhance the representation, we incorporate memory vector into the attention mechanism to capture prior information about the data and correlations between covariates. Furthermore, we employ the Hilbert-Schmidt Independence Criterion (HSIC) to constrain the learned representation, ensuring its relevance to the original data. Our proposed method has shown superior results on different datasets compared to previous classic works, which demonstrates our success in representation balance and predictive ability. ? 2024, The Authors. All rights reserved.