Causal Effect Estimation with Mixed Latent Confounders and Post-treatment Variables
Authors: Yaochen Zhu, Jing Ma, Liang Wu, Qi Guo, Liangjie Hong, Jundong Li
ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on both simulated and real-world datasets demonstrate significantly improved robustness of Ci VAE. |
| Researcher Affiliation | Collaboration | Yaochen Zhu1, Jing Ma2, Liang Wu3, Qi Guo3, Liangjie Hong3, Jundong Li1 1University of Virginia, 2Case Western Reserve University, 3Linked In Inc. EMAIL EMAIL EMAIL |
| Pseudocode | No | The paper describes the methodology in detail within Section 4 'METHODOLOGY' but does not include a distinct pseudocode block or algorithm listing. |
| Open Source Code | Yes | Code available at https://github.com/yaochenzhu/CiVAE. |
| Open Datasets | No | The paper uses 'Simulated Datasets' whose generative processes are described, and 'Real-world Datasets' from LinkedIn, which were built by the authors from collected job data. No public access links, repositories, or explicit statements of public availability are provided for these datasets. |
| Dataset Splits | No | The paper describes the generative processes for simulated datasets and the collection of real-world datasets, but it does not specify any training/test/validation splits in terms of percentages, absolute counts, or references to standard predefined splits. |
| Hardware Specification | No | The paper does not explicitly describe the specific hardware (e.g., GPU models, CPU types, memory specifications) used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details, such as library names with version numbers (e.g., Python, PyTorch, TensorFlow versions) used in the experiments. |
| Experiment Setup | Yes | For Latent Mediator, γ is set as [ 1, 1, 1], θ is set as [1, 1, 1], and τ is set as 2, which results in an ATE = 1. For the Latent Correlator dataset, we set the same γ and θ as the Latent Mediator dataset, where parameters ϕ and τ are set to 1, which results in ATE of 1. Z Gaussian(0, IKZ), X Multi(NNf(Z)), Y Gaussian(w Z, 1), where KZ = 8, and represents the element-wise product operator, respectively. We then treat the first KC = 5 dimensions of Z as the latent confounders C and the remaining KM = KZ KC dimensions as the latent mediators M. After learning NNf and w according to Eq. (11), we draw latent confounders C Gaussian(0, I), latent mediators M = T γ, and set the outcome Y = w [C||M] + τ T, where the true ATE can be calculated as sum(γ w KM:) + τ. |