Robust Heterogeneous Treatment Effect Estimation under Covariate Perturbation
Keywords: causal inference, treatment effect estimation, robust estimation
Abstract: Heterogeneous treatment effect estimation has important applications in fields such as healthcare, economics, and education, attracting increasing attention from both research and the industrial community. However, most existing causal machine learning methods may not perform well in practice due to the lack of robustness of the treatment effect estimation predicted by deep neural networks when an imperceptible perturbation has been added to the covariate. In this paper, we alleviate this problem using the idea of adversarial machine learning. We first show that our loss of interest, the adversarial loss, is partly bounded by the Lipschitz constant of the casual model. Next, we propose a representation learning framework called RHTE which estimates heterogeneous treatment effect under covariate perturbation by controlling the empirical loss, Lipschitz constant, and distance metric simulta neously. Theories are then derived to guarantee the performance and robustness of our estimation. To the best of our knowledge, this is the first work proposing robust representation learning methods under variable perturbation. Extensive experiments on both synthetic examples and standard benchmarks demonstrate the effectiveness and generality of our framework.
Supplementary Material: zip
Primary Area: causal reasoning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 13949
Loading