We solicit submission of original research at the interface between optimal transport theory, statistics, optimization, machine learning and applications. Topics of interest to the workshop broadly include
Estimation of Optimal Transport Couplings and Maps
- Resolution of the Kantorovich problem and its variants (semi-discrete, continuous) at large scales
- Estimation of Monge maps between measures using parameterized or constrained families of potentials
- Estimation of Monge maps across time using JKO formulations.
Generalizations of OT
- Unbalanced formulations: OT between measures of different total mass
- Quadratic formulations of the OT problem, Gromov-Wasserstein metrics. OT with rigid transformations.
- Multi-marginal formulations for OT
- Quantum OT
- Martingale OT and time constraints in the estimation of OT
OT as a learning methodology
- Optimal transport costs as a loss: GAN type results, minimization of Wasserstein distances between parameterized measures and data measures.
- Optimal transport to define data transformations: domain adaptation, clustering, incorporation into MCMC methods.
- Learning of ground metrics and costs, OT carried out on projections (sliced and higher dimensional)
Statistics and Theory
- Interplay between OT and statistics: properties of finite sample OT estimators between continuous measures, either seen as Monge maps or couplings.
- OT as a criterion to study the convergence of sampling methods.
- OT as a tool to define convexity, sorting and quantiles
- Study of the complexity of OT algorithms such as the Sinkhorn iteration and variants.
OT in applications
- High-dimensional applications: Natural Language / Word Embeddings, Biology, Vision, etc...
- Low-dimensional applications: Graphics, shapes, imaging, univariate measures, etc...
Authors can submit work that overlaps with previously published or submitted work, as long as it adds a new perspective on that work.
Submission Format and Instructions
Authors can submit work that overlaps with previously published or submitted work, as long as it adds a new perspective on that work.
Authors should use the standard Neurips 2019 style files: https://neurips.cc/Conferences/2019/PaperInformation/StyleFiles
The recommended length for submissions is between 6 and 10 pages.
All submissions must be processed using the easychair website set up for this workshop. Please follow the link below. We cannot accept submissions by email.
https://easychair.org/conferences/?conf=otml19
We guarantee that all works submitted before September 15, 2019 (23:59 PDT) will be given full consideration and will be reviewed by at least two reviewers.
Presentation of Accepted Contributions
All accepted papers will have a poster presentation and selected papers will be given oral presentation slots.
The papers will not be published in proceedings but the authors will be able to provide a link to the arxiv preprint.