Causal Discovery in Linear Latent Variable Models Subject to Measurement Error
Authors: Yuqin Yang, AmirEmad Ghassami, Mohamed Nafea, Negar Kiyavash, Kun Zhang, Ilya Shpitser
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We propose causal structure learning methods for both models, and evaluate their performance on synthetic data.We evaluated the performance of our recovery algorithm on randomly generated linear SEM-MEs and SEM-URs with different number of variables. |
| Researcher Affiliation | Academia | Yuqin Yang Georgia Institute of Technology Amir Emad Ghassami Johns Hopkins University Mohamed Nafea University of Detroit Mercy Negar Kiyavash École Polytechnique Fédérale de Lausanne (EPFL) Kun Zhang Carnegie Mellon University Ilya Shpitser Johns Hopkins University |
| Pseudocode | Yes | See Appendix A.2 for details and pseudo-code for AOG recovery. (Appendix A.2 contains 'Algorithm 1: AOG recovery algorithm') |
| Open Source Code | Yes | Our code is available at: https://github.com/Yuqin-Yang/SEM-ME-UR. |
| Open Datasets | No | Our simulations are conducted on synthetic data only. The paper mentions generating data rather than using a pre-existing, publicly available dataset that would require a direct link or citation for access. |
| Dataset Splits | No | The paper uses synthetic data generated for experiments and mentions 'different sample sizes'. While Appendix E.4 describes the experimental setup, it does not specify explicit training, validation, or test dataset splits (e.g., percentages or specific counts) for a fixed dataset, as the data is generated per run. |
| Hardware Specification | Yes | Simulations are conducted using the CPU of a Mac Book laptop. |
| Software Dependencies | No | The paper mentions software like 'Reconstruction ICA' and 'ICA-Li NGAM' (algorithms used), and that 'Our codes are written in Python'. However, it does not provide specific version numbers for Python or any libraries/packages used (e.g., Python 3.x, NumPy 1.x, PyTorch 1.x). |
| Experiment Setup | Yes | We considered two cases: (1) when a noisy version of the mixing matrix is given, where the noise is Gaussian with different choices of variance denoted by d2, and (2) when synthetic data comes from a linear generating model with non-Gaussian noises with different sample sizes, and the mixing matrix needs to be estimated. (Further details in Appendix E.4 state specific ranges for number of variables, noise variance, and sample sizes) |