Linear-Time Gromov Wasserstein Distances using Low Rank Couplings and Costs

Authors: Meyer Scetbon, Gabriel Peyré, Marco Cuturi

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our approach yields similar results, yet orders of magnitude faster computation than the So TA entropic GW approaches, on both simulated and real data.
Researcher Affiliation Collaboration Meyer Scetbon 1 CREST-ENSAE Gabriel Peyr e 2 CNRS and ENS, PSL Marco Cuturi 3 CREST-ENSAE, work partly done at Google, currently at Apple. Correspondence to: meyer scetbon <meyer.scetbon@ensae.fr>.
Pseudocode Yes Algorithm 1: Entropic-GW
Open Source Code Yes The code is available at https://github.com/meyerscetbon/Linear Gromov.
Open Datasets Yes Experiments were run on a Mac Book Pro 2019 laptop, and data from github.com/rsinghlab/SCOT. The code is available at https://github.com/meyerscetbon/Linear Gromov.
Dataset Splits No The paper mentions synthetic and real datasets, and specific sample sizes for experiments (e.g., 'n = m = 1000 samples'), but it does not provide explicit training/validation/test split percentages or sample counts for these splits.
Hardware Specification Yes Experiments were run on a Mac Book Pro 2019 laptop
Software Dependencies No The paper does not specify software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9).
Experiment Setup Yes In all other experiments, we always set γ = 100 and α = 10 10 for our methods, and only focus on rank r.