Learning Linear Causal Representations from General Environments: Identifiability and Intrinsic Ambiguity

Authors: Jikai Jin, Vasilis Syrgkanis

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive experiments on synthetic data and demonstrate the effectiveness of Li NGCRe L in the finite-sample regime.
Researcher Affiliation Academia Jikai Jin Institute for Computational and Mathematical Engineering Stanford University Stanford, CA 94305 jkjin@stanford.edu Vasilis Syrgkanis Management Science and Engineering Stanford University Stanford, CA 94305 vsyrgk@stanford.edu
Pseudocode Yes Algorithm 1 Orthogonal-projections; Algorithm 2 Identify-Parents; Algorithm 3 Learn-Causal-Model
Open Source Code No Answer: [No] Justification: Code will be released after review.
Open Datasets No We generate the independent noise variables from generalized Gaussian distributions pβ(x) exp |x|β with parameters βk = 0.2k2, k = 1, 2, , d, multiplied by normalization constants to make their variances equal to 1. The ground-truth causal graph is generated by first fixing a total order of the vertices, say 1, 2, , d, then add directed edges i j(i < j) according to i.i.d. Bernoulli(p) distributions, where p (0, 1).
Dataset Splits No The paper discusses sample sizes for synthetic data but does not specify explicit training, validation, or test dataset splits.
Hardware Specification No Answer: [No] Justification: The experiments do not require huge computational resources and can be run on a local computer.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies used in the experiments.
Experiment Setup Yes In our implementation of Algorithm 3, in each iteration we instead choose i / S that has the largest ratio between the first and second singular values of [q1, q2, , q K]. And in line 6 of Algorithm 2, we introduce a hyper-parameter tl such that the matrix [q1, q2, , q K] is considered to have rank rm 1 if its rm -th singular value is smaller than tl.