Causal Component Analysis

Authors: Liang Wendong, Armin Kekić, Julius von Kügelgen, Simon Buchholz, Michel Besserve, Luigi Gresele, Bernhard Schölkopf

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We introduce a likelihood-based approach using normalizing flows to estimate both the unmixing function and the causal mechanisms, and demonstrate its effectiveness through extensive synthetic experiments in the Cau CA and ICA setting.
Researcher Affiliation Academia 1 Max Planck Institute for Intelligent Systems, Tübingen, Germany 2 ENS Paris-Saclay, Gif-sur-Yvette, France 3 University of Cambridge, United Kingdom
Pseudocode No The paper does not contain any pseudocode or algorithm blocks.
Open Source Code Yes Code available at https://github.com/akekic/causal-component-analysis.
Open Datasets No The paper states: "Synthetic data-generating process. We first sample DAGs G..." and "We then sample observed mixtures from these latent CBNs, as described by eq. (2)." This indicates that the authors generated their own synthetic data and do not provide access to a publicly available dataset.
Dataset Splits Yes We train the model for 50-200 epochs with a batch size of 4096. The number of epochs was tuned manually for each type of experiment to ensure reliable convergence of the validation log probability. For each drawn latent CBN, we train three models with different initializations and select the model with the highest validation log probability at the end of training.
Hardware Specification Yes Each training run takes 2-8 hours on NVIDIA RTX-6000 gpus.
Software Dependencies No The paper mentions "ADAM optimizer" and "Neural Spline Flows" (which implies a deep learning framework), but it does not provide specific version numbers for these or other software libraries (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes We use the ADAM optimizer [27] with cosine annealing learning rate scheduling, starting with a learning rate of 5 imes 10^-3 and ending with 1 imes 10^-7. We train the model for 50-200 epochs with a batch size of 4096. The number of epochs was tuned manually for each type of experiment to ensure reliable convergence of the validation log probability.