Online Sinkhorn: Optimal Transport distances from sample streams

Authors: Arthur Mensch, Gabriel Peyré

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate our method on synthetic 1-d to 10-d data and on real 3-d shape data.
Researcher Affiliation Academia Arthur Mensch PSL University CNRS, ENS, DMA Paris, France arthur.mensch@m4x.org Gabriel Peyré PSL University CNRS, ENS, DMA Paris, France gabriel.peyre@ens.fr
Pseudocode Yes Algorithm 1 Online Sinkhorn
Open Source Code No The paper does not provide a concrete access link or statement about open-sourcing the code for the methodology.
Open Datasets Yes Stanford 3D scans Turk and Levoy, 1994
Dataset Splits No The paper does not specify exact split percentages or methods for validation datasets beyond general use of samples.
Hardware Specification No The paper does not specify the hardware used for experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes We report quantitative results for ε = 10 2 and non fully-corrective online Sinkhorn in the main text, and all other curves in Supp. Fig. 4. In Supp. Fig. 7, we also report results for OT between Gaussians, which is a simpler and less realistic setup, but for which closed-form expressions of the potentials are known Janati et al., 2020. ... We use n(t) = N 100(1 + 0.1t)1/2 results vary little with the exponent.