Generalized Independent Noise Condition for Estimating Latent Variable Causal Graphs

Authors: Feng Xie, Ruichu Cai, Biwei Huang, Clark Glymour, Zhifeng Hao, Kun Zhang

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on synthetic and real-world data demonstrate the effectiveness of our method.
Researcher Affiliation Academia Feng Xie 1,2, Ruichu Cai 1,3, Biwei Huang4, Clark Glymour4, Zhifeng Hao1,5, Kun Zhang 4 1 School of Computer Science, Guangdong University of Technology, Guangzhou, China 2 School of Mathematical Sciences, Peking University, Beijing, China 3 Pazhou Lab, Guangzhou, China 4 Department of Philosophy, Carnegie Mellon University, Pittsburgh, USA 5 School of Mathematics and Big Data, Foshan University, Foshan, China
Pseudocode Yes Algorithm 1 Identifying Causal Clusters and Algorithm 2 Learning the Causal Order of Latent Variables are explicitly provided in the paper.
Open Source Code Yes Our source code is available from https://github.com/xiefeng009/ GIN-Condition-for-Estimating-Latent-Variable-Causal-Graphs.
Open Datasets Yes Barbara Byrne conducted a study to investigate the impact of organizational (role ambiguity, role conflict, classroom climate, and superior support, etc.) and personality (selfesteem, external locus of control) on three facets of burnout in full-time elementary teachers [Byrne, 2010]. We applied our algorithm to this data set, with 28 observed variables in total.
Dataset Splits No The paper describes the generation of synthetic data with sample sizes (N = 500, 1000, 2000) and mentions a real-world dataset but does not specify explicit train/validation/test splits or percentages for either. It mentions 'Each experiment was repeated 10 times' for synthetic data, but this is a repetition count, not a data split.
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments, such as GPU models, CPU types, or memory specifications.
Software Dependencies No The paper mentions using 'Hilbert-Schmidt Independence Criterion (HSIC) test' and 'TETRAD package' for comparisons, but it does not provide specific version numbers for these or any other software dependencies.
Experiment Setup Yes In all four cases, the data are generated by Li NGLa M and the causal strength b is sampled from a uniform distribution between [ 2, 0.5] [0.5, 2], noise terms are generated from uniform[-1,1] variables to the fifth power, and the sample size N = 500, 1000, 2000. Each experiment was repeated 10 times with randomly generated data and the results were averaged. In the implementation, the kernel width in the HSIC test is set to 0.05.