Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
Identifiable Exchangeable Mechanisms for Causal Structure and Representation Learning
Authors: Patrik Reizinger, Siyuan Guo, Ferenc Huszar, Bernhard Schölkopf, Wieland Brendel
ICLR 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To demonstrate that both cause and mechanism variability enable causal structure identification, we ran synthetic experiments based on the publicly available repository of the Causal de Finetti paper5. [...] Results. Fig. 6 shows the proportion of correctly identified causal structures for different numbers of environments. The Causal-de-Finetti algorithm outperforms all the other methods with an accuracy close to 100%. |
| Researcher Affiliation | Academia | 1Max Planck Institute for Intelligent Systems, Tübingen, Germany 2University of Cambridge, Cambridge, United Kingdom 3ELLIS Institute Tübingen, Tübingen, Germany 4Tübingen AI Center, Tübingen, Germany |
| Pseudocode | No | The paper describes theoretical concepts and mathematical derivations but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code is available at https:// github.com/rpatrik96/IEM |
| Open Datasets | Yes | To demonstrate that both cause and mechanism variability enable causal structure identification, we ran synthetic experiments based on the publicly available repository of the Causal de Finetti paper5. [...] The Cd F parameters N = [ψ, θ] were randomly generated with distinct and independent elements in each environment. Samples within each environment have the noise variables S generated via Laplace distributions conditioned on the corresponding Cd F parameters i.e., the Cd F parameter is the location (mean) of the Laplace distribution. |
| Dataset Splits | Yes | We use, as in the original code, two samples per environment and ablate over {100, 200, 300, 400, 500} environments. Each experiment is repeated 100 times. |
| Hardware Specification | No | The paper mentions "compute resources at the Tübingen Machine Learning Cloud" in the acknowledgments but does not provide specific hardware models (e.g., CPU/GPU types, memory). |
| Software Dependencies | No | The paper mentions not using a "Mat Lab license" for a comparison method (CD-NOD) but does not specify any software dependencies with version numbers for their own implementation. |
| Experiment Setup | Yes | We ran synthetic experiments based on the publicly available repository of the Causal de Finetti paper5. [...] The Cd F parameters N = [ψ, θ] were randomly generated with distinct and independent elements in each environment. Samples within each environment have the noise variables S generated via Laplace distributions conditioned on the corresponding Cd F parameters i.e., the Cd F parameter is the location (mean) of the Laplace distribution. [...] We measure causal structure identification by three conditional independence tests with a significance level of α = 0.05. We choose the estimated causal structure to be the one corresponding to the test with the highest p-value. |