On a Neural Implementation of Brenier’s Polar Factorization

Authors: Nina Vesseron, Marco Cuturi

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5. Experiments, 5.1. Accuracy Metrics for NPF, Table 1. Polar factorization and Inverse multivalued map metrics for learning the gradient of the elevation in Chamonix area.
Researcher Affiliation Collaboration 1CREST-ENSAE, IP Paris 2Apple.
Pseudocode Yes Algorithm 1 Training of Xψ
Open Source Code No The paper does not provide an explicit statement or a link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We use the Python package elevation to get the elevation of three regions of the world: Chamonix, London, and Cyprus. classify MNIST digits.
Dataset Splits Yes To assess the quality of our method NPF, we used a 85% training / 15% test split.
Hardware Specification No This work was performed using HPC resources from GENCI IDRIS (Grant 2023-103245).
Software Dependencies No The paper mentions software components like 'ADAM solver' and 'diffrax', but does not provide specific version numbers for these or other key software dependencies required for reproducibility.
Experiment Setup Yes G. Hyperparameters section, specifically Figure 17 (and subsequent figures) which details 'model hyperparameter value' for various components including activation functions, architectures, learning rates, schedulers, and solver parameters.