M
Marco Cuturi
Researcher at Google
Publications - 155
Citations - 12954
Marco Cuturi is an academic researcher from Google. The author has contributed to research in topics: Computer science & Metric (mathematics). The author has an hindex of 42, co-authored 141 publications receiving 9403 citations. Previous affiliations of Marco Cuturi include École Normale Supérieure & Mines ParisTech.
Papers
More filters
Proceedings Article
Unsupervised Riemannian Metric Learning for Histograms Using Aitchison Transformations
Tam Le,Marco Cuturi +1 more
TL;DR: This paper considers the problem of learning a Riemannian metric on the simplex given unlabeled histogram data and proposes an algorithmic approach to maximize inverse volumes using sampling and contrastive divergences.
Posted Content
Wasserstein Training of Boltzmann Machines
TL;DR: A novel approach for Boltzmann training which assumes that a meaningful metric between observations is given, represented by the Wasserstein distance between distributions, is proposed, for which a gradient is derived with respect to the model parameters.
Proceedings Article
Projection Robust Wasserstein Distance and Riemannian Optimization
TL;DR: In this paper, it was shown that the Wasserstein projection pursuit (WPP) distance can be efficiently computed in practice using Riemannian optimization, yielding in relevant cases better behavior than its convex relaxation.
Posted Content
Fast and Robust Comparison of Probability Measures in Heterogeneous Spaces.
TL;DR: The Anchor Energy (AE) and Anchor Wasserstein (AW) distances are proposed, simpler alternatives to GW built upon the representation of each point in each distribution as the 1D distribution of its distances to all other points.
Posted Content
Precision-Recall Curves Using Information Divergence Frontiers
TL;DR: This paper presents a general evaluation framework for generative models that measures the trade-off between precision and recall using Renyi divergences, and provides a novel perspective on existing techniques and extends them to more general domains.