Unpaired Multi-Domain Causal Representation Learning
Authors: Nils Sturma, Chandler Squires, Mathias Drton, Caroline Uhler
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this paper, we give sufficient conditions for identifiability of the joint distribution and the shared causal graph in a linear setup. ... We conclude with a small simulation study as a proof of concept for the finite sample setting in Section 5. |
| Researcher Affiliation | Academia | Nils Sturma Technical University of Munich Munich Center for Machine Learning nils.sturma@tum.de Chandler Squires LIDS, Massachusetts Institute of Technology Broad Institute of MIT and Harvard csquires@mit.edu Mathias Drton Technical University of Munich Munich Center for Machine Learning mathias.drton@tum.de Caroline Uhler LIDS, Massachusetts Institute of Technology Broad Institute of MIT and Harvard cuhler@mit.edu |
| Pseudocode | Yes | Algorithm 1 Identify Joint Distribution... Algorithm 2 Identify Shared Graph... Algorithm 3 Identify Joint Distribution Empirical... Algorithm 4 Identify Shared Graph Empirical |
| Open Source Code | No | The paper does not provide any explicit statements or links indicating the availability of open-source code for the described methodology. |
| Open Datasets | No | The paper describes generating data for simulations ('In each experiment we generate 1000 random models') and specifies the error distributions in Appendix E, but does not refer to or provide access to any pre-existing public or open dataset. |
| Dataset Splits | No | The paper describes simulation studies with varying sample sizes ('ne = n for all e [m], and consider n {1000, 2500, 5000, 10000, 25000}'), but does not specify any training, validation, or test dataset splits. The simulations are used as a 'proof of concept' for the theoretical identifiability. |
| Hardware Specification | Yes | The computations were performed on a single thread of an Intel Xeon Gold 6242R processor (3.1 GHz), with a total computation time of 12 hours for all simulations presented in this paper (including Appendix). |
| Software Dependencies | No | The paper mentions types of algorithms for linear ICA (e.g., 'Fast ICA', 'Kernel ICA', 'JADE') and refers to their sources, but does not specify version numbers for any software dependencies used in their implementation or experiments. |
| Experiment Setup | Yes | The adapted algorithms have a hyperparameter γ, which is a threshold on singular values to determine the rank of a matrix. In our simulations we use γ = 0.2. |