Independent mechanism analysis, a new concept?

Authors: Luigi Gresele, Julius von Kügelgen, Vincent Stimper, Bernhard Schölkopf, Michel Besserve

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide theoretical and empirical evidence that our approach circumvents a number of nonidentifiability issues arising in nonlinear blind source separation. (...) We experimentally validate our theoretical claims and propose a regularised maximum-likelihood learning approach based on the IMA constrast which outperforms the unregularised baseline ( 5).
Researcher Affiliation Academia 1 Max Planck Institute for Intelligent Systems, Tübingen, Germany 2 University of Cambridge
Pseudocode No The paper describes algorithms and models using mathematical equations and textual explanations, but it does not include any formal pseudocode or algorithm blocks.
Open Source Code Yes Code available at: https://github.com/lgresele/independent-mechanism-analysis
Open Datasets No The paper uses synthetically generated data: "We sample the ground truth sources from a uniform distribution in [0, 1]n". It does not use a named, publicly available dataset.
Dataset Splits No The paper does not specify percentages or counts for training, validation, and test splits. It mentions generating "1000 random mixing functions" and "50 random mixings" for evaluation but not specific data splits.
Hardware Specification No The main text of the paper does not specify hardware details such as GPU/CPU models, memory, or cloud resources used for experiments. It mentions in the checklist that these are in Appendix E, which is not provided in the context.
Software Dependencies No In all of our experiments, we use JAX [12] and Distrax [13]. No specific version numbers for these libraries are provided in the main text.
Experiment Setup Yes Experimental setup. To use CIMA as a learning signal, we consider a regularised maximum-likelihood approach, with the following objective: L(g) = Ex[log pg(x)] λ CIMA(g 1, py), where g denotes the learnt unmixing, y = g(x) the reconstructed sources, and λ 0 a Lagrange multiplier. (...) We train a residual flow g (with full Jacobian) to maximise L. (...) Quantitative results for 50 learnt models for each λ {0.0, 0.5, 1.0} and n {5, 7} are summarised in Fig. 5.