Sample Complexity of Interventional Causal Representation Learning
Authors: Emre Acartürk, Burak Varıcı, Karthikeyan Shanmugam, Ali Tajer
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we perform numerical assessments of our analyses to provide complementary insight into the sample complexity results of Section 5. Specifically, we evaluate the variations of the model constants with respect to problem dimensions n and d. |
| Researcher Affiliation | Collaboration | Emre Acartürk Rensselaer Polytechnic Institute Burak Varıcı Carnegie Mellon University Karthikeyan Shanmugam Google Deep Mind Ali Tajer Rensselaer Polytechnic Institute |
| Pseudocode | Yes | Algorithm 1 Causal order estimation Algorithm 2 Graph estimation Algorithm 3 Inverse transform estimation |
| Open Source Code | Yes | The codebase for the experiments can be found at https://github.com/acarturk-e/ finite-sample-linear-crl. |
| Open Datasets | No | The paper does not use a publicly available or open dataset. Instead, data is generated based on specified parameters: 'We consider problem dimensions n {3, 5, 10} and d {n, 15} and generate G using Erd os-Rényi model with density 0.5 on n nodes. We adopt linear Gaussian models as the latent causal model.' |
| Dataset Splits | No | The paper describes generating N samples (N {102.5, 103, 103.5, 104, 104.5, 105}) for experiments, but does not explicitly provide details about training, validation, or test dataset splits (e.g., percentages, sample counts, or predefined splits). |
| Hardware Specification | Yes | Experiments are run on a single commercial laptop CPU. |
| Software Dependencies | No | The paper does not provide specific version numbers for software dependencies or libraries used in the experiments. |
| Experiment Setup | Yes | Experimental details. We consider problem dimensions n {3, 5, 10} and d {n, 15} and generate G using Erd os-Rényi model with density 0.5 on n nodes. We adopt linear Gaussian models as the latent causal model. We consider N {102.5, 103, 103.5, 104, 104.5, 105} samples, and generate 100 latent models for each triplet (N, n, d). |