Sinkhorn Barycenters with Free Support via Frank-Wolfe Algorithm

Authors: Giulia Luise, Saverio Salzo, Massimiliano Pontil, Carlo Ciliberto

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments validate the effectiveness of our method in practice.We empirically evaluate the performance of the proposed algorithm.
Researcher Affiliation Academia Giulia Luise1, Saverio Salzo2, Massimiliano Pontil1,2, Carlo Ciliberto3 g.luise.16@ucl.ac.uk, saverio.salzo@iit.it, m.pontil@cs.ucl.ac.uk,c.ciliberto@ic.ac.uk 1 Department of Computer Science, University College London, UK 2 CSML, Istituto Italiano di Tecnologia, Genova, Italy 3 Department of Electrical and Electronic Engineering, Imperial College London, UK
Pseudocode Yes Algorithm 1 FRANK-WOLFE IN DUAL BANACH SPACES
Open Source Code Yes Code has been made publicly available1. 1 https://github.com/GiulsLu/Sinkhorn-Barycenters
Open Datasets Yes k-means on MNIST digits. We consider a subset of 500 random images from the MNIST dataset. Continuous measures: barycenter of Gaussians. We compute the barycenter of 5 Gaussian distributions N(mi, Ci) i = 1, . . . , 5 in R2, with mean mi 2 R2 and covariance Ci randomly generated.
Dataset Splits No The information is insufficient. The paper mentions datasets like MNIST and Gaussian distributions but does not provide specific train/validation/test dataset splits (percentages, sample counts, or explicit standard split references) for reproducibility.
Hardware Specification No The information is insufficient. The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The information is insufficient. The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes We compute the barycenter of 30 randomly generated nested ellipses on a 50 50 grid similarly to [14]. We apply Alg. 2 to empirical measures obtained by sampling n = 500 points from each N(mi, Ci), i = 1, . . . , 5. We initialize 20 centroids according to the k-means++ strategy [4].