Identifying General Mechanism Shifts in Linear Causal Representations

Authors: Tianyu Chen, Kevin Bello, Francesco Locatello, Bryon Aragam, Pradeep Ravikumar

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We corroborate our results on both synthetic experiments as well as an interesting psychometric dataset.
Researcher Affiliation Academia Department of Statistics and Data Sciences, University of Texas at Austin Booth School of Business, University of Chicago Machine Learning Department, Carnegie Mellon University Institute of Science and Technology Austria
Pseudocode Yes Algorithm 1 i LCS: Identifying Latent Causal Mechanisms Shifts
Open Source Code Yes The code can be found at https://github.com/Tianyu Codings/i LCS.
Open Datasets Yes The data can be downloaded via the link: https://www.kaggle.com/datasets/ lucasgreenwell/ocean-five-factor-personality-test-responses/data.
Dataset Splits No The paper does not explicitly specify training, validation, and test dataset splits (e.g., percentages or sample counts) for model training and evaluation.
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments, such as GPU or CPU models.
Software Dependencies No The paper mentions using ICA but does not provide specific version numbers for any software libraries, frameworks, or programming languages used in the experiments.
Experiment Setup Yes The parameter α is set to 0.2 for d 10 and 0.5 for higher dimensions, reflecting the increased complexity in estimating larger dimensional latent graphs and thus necessitating a higher tolerance for L1 norm differences in detecting shifted nodes. For each graph, the weights are independently sampled from Unif [0.25, 1] and the diagonal entries of Ωfrom Unif[2, 4]. In each environment k, 15% of the nodes are randomly selected for shifting. The new weights A(k) i for the shifted node i, and the new entries of Ω(k), specifically Ω(k) ii , are independently sampled from Unif[6, 8]. The mixing function G is independently generated from Unif[ 0.25, 0.25].