Gromov-Wasserstein Averaging of Kernel and Distance Matrices

Authors: Gabriel Peyré, Marco Cuturi, Justin Solomon

ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate its application to the computation of shape barycenters and to the prediction of energy levels from molecular configurations in quantum chemistry.
Researcher Affiliation Academia Gabriel Peyr e GABRIEL.PEYRE@CEREMADE.DAUPHINE.FR CNRS and Univ. Paris-Dauphine, Pl. du M. De Lattre De Tassigny, 75775 Paris 16, FRANCE Marco Cuturi MCUTURI@I.KYOTO-U.AC.JP Kyoto University, 36-1 Yoshida-Honmachi, Sakyo-ku, Kyoto 606-8501, JAPAN Justin Solomon JSOLOMON@MIT.EDU Massachusetts Institute of Technology, Cambridge, MA 02139, USA
Pseudocode Yes Algorithm 1 Computation of GWε barycenters.
Open Source Code Yes The code to reproduce the results of this paper is available online.1
Open Datasets Yes In this experiment, we extract 500 point clouds of handwritten digits from the dataset (Le Cun et al., 1998), rotated arbitrarily in the plane... qm7 dataset of organic 7165 molecules
Dataset Splits Yes (Hansen et al., 2013, Table 3) provide out-of-sample mean absolute error (MAE) predictions for several techniques using 5 fold cross-validation.
Hardware Specification No The paper does not specify the hardware used for running experiments.
Software Dependencies No The paper mentions algorithms like Sinkhorn iterations and k-means but does not specify software dependencies with version numbers (e.g., Python, PyTorch, specific libraries).
Experiment Setup Yes We represent each digit as a symmetric Euclidean distance matrix and optimize for a 500 x 500 barycenter using Algorithm 1 (uniform weights, ε = 1 10 3)... We then cluster the point clouds by representing them as pairwise distance matrices and applying the k-means algorithm (k = 5), with k-means++ initialization (Arthur & Vassilvitskii, 2007)... Using a 3-nearest neighbor regression approach...