Function Classes for Identifiable Nonlinear Independent Component Analysis
Authors: Simon Buchholz, Michel Besserve, Bernhard Schölkopf
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | An experimental illustration of this setting and Theorem 4 below can be found in Appendix H. The notion of locality in Definition 5 should be understood as non-global and notably does not imply restrictions to a small neighbourhood, as local properties often do. (Page 5) And from Appendix H: In this Appendix, we illustrate our theoretical results from Section 4 and 5 on local identifiability of OCTs vs. general nonlinear functions numerically. We train a normalizing flow, following the setup in [13, 39], to learn the inverse of an orthogonal map f : Rd ! Rd (for d = 2) and investigate its behaviour in regions where the mixing changes continuously. |
| Researcher Affiliation | Academia | Simon Buchholz, Michel Besserve & Bernhard Schölkopf Max Planck Institute for Intelligent Systems Tübingen, Germany {sbuchholz,mbesserve,bs}@tue.mpg.de |
| Pseudocode | No | The paper contains theoretical discussions, definitions, and proofs, but no section or figure explicitly labeled 'Pseudocode' or 'Algorithm'. |
| Open Source Code | Yes | 3. If you ran experiments... (a) Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] See Appendix H (Page 9) and in Appendix H: Our code is available at https://anonymous.4open.science/r/function-classes-for-nonlinear-ica-19EF/ |
| Open Datasets | No | For the experiments, we used synthetic data generated according to Eq. (1). (Appendix H). This is not a pre-existing publicly available dataset. No specific public dataset with access information is provided. |
| Dataset Splits | No | The paper mentions drawing 10000 samples in each epoch for training, but it does not specify any explicit train/validation/test dataset splits, proportions, or sample counts for these subsets. |
| Hardware Specification | Yes | All experiments were run on a single NVIDIA A100 GPU. |
| Software Dependencies | No | The paper mentions using 'nflows' (implied from reference [13] 'nflows: normalizing flows in Py Torch') and 'Adam' [31] as an optimizer. However, it does not provide specific version numbers for these software components or frameworks (e.g., 'PyTorch 1.9' or 'nflows 0.1.0'), which are required for full reproducibility. |
| Experiment Setup | Yes | We use Adam [31] with a learning rate of 10^-3 and train for 100 epochs with a batch size of 256. |