Skip to content

gpeyre/ot4ml

Repository files navigation

OT4ML - Optimal Transport for Machine Learners

Project webpage: https://www.gpeyre.com/ot4ml

Project homepage  ·  PDE4ML survey  ·  Interactive book  ·  Rendered figure gallery  ·  Resources  ·  GitHub source

This repository gathers the OT4ML book, the PDE4ML survey, the executable notebooks used to reproduce the figures, a shorter set of teaching notebooks, and an experimental MyST web prototype.

PDE4ML Survey

PDEs for Machine Learning is a long survey of PDE tools for machine learning, written with an optimal-transport bias. It reorganizes the OT4ML material most relevant to dynamic optimal transport, Wasserstein gradient flows, particle limits, diffusion models, flow matching, mean-field training, and transportation views of modern architectures.

PDE4ML survey visual

Sources and build notes live in PDE4ML/.

Book

The book Optimal Transport for Machine Learners is available on arXiv.

Figures of the book

The book figures are generated from executable notebooks and assembled by the LaTeX source. The current searchable gallery has been checked against the live LaTeX and MyST figure references: it exposes 113 figure views, covers all 112 referenced latex/figures/<figure-name>/ directories, and every active view has a notebook link, thumbnail, and generated PDF panels. The manuscript contains 115 LaTeX figure labels because some figure directories generate several labeled figures. Browse the rendered web gallery at www.gpeyre.com/ot4ml/notebooks-figures/index.html or the Markdown version in notebooks-figures/README.md, with thumbnails, notebook links, and Open in Colab badges.

Contact sheet of OT4ML book figures

Each live figure notebook writes PDF panels to latex/figures/<figure-name>/, where the LaTeX source assembles them into the book. Retired exploratory notebooks live in notebooks-figures/removed/ and are not part of the paper gallery.

Teaching Notebooks

The course notebooks below are compact, self-contained introductions to the main computational ideas. Each one can be opened locally or launched in Colab.

1. Optimal Transport with Linear Programming
1-linprog preview
Open In Colab
2. Entropic Regularization of Optimal Transport
2-sinkhorn preview
Open In Colab
3. Advanced Topics on Sinkhorn Algorithms
3-sinkhorn-advanced preview
Open In Colab
4. Semi-discrete Optimal Transport
4-semidiscrete preview
Open In Colab
5. Unbalanced Optimal Transport
5-unbalanced preview
Open In Colab
6. Diffusion Models and Optimal Transport
6-diffusion preview
Open In Colab
7. Wasserstein Gradient Flows of Interaction Functionals
7-wasserstein-flows preview
Open In Colab
8. Discrete Diffusion
8-discrete_diffusion preview
Open In Colab

Executable Web Prototype

An experimental MyST/Jupyter Book 2 prototype lives in myst/. It mirrors the LaTeX book front matter, 14 main chapters, conclusion, acknowledgements, and notation appendix while fusing the book text with executable figures and browser-native interactive demos. The rendered static version is available from the project homepage. The local workflow, the static-site build, and the offline behavior of the interactive demos are documented in myst/README.md.

Course Slides

Further Resources

A rendered resource portal is available at www.gpeyre.com/ot4ml/resources.html.

Bibliography

Core books and monographs cited in the book:

  • Mass Transportation Problems, Vol. I: Theory, Svetlozar T. Rachev & Ludger Rüschendorf, Springer, 1998.
  • Mass Transportation Problems, Vol. II: Applications, Svetlozar T. Rachev & Ludger Rüschendorf, Springer, 1998.
  • Topics in Optimal Transportation, Cédric Villani, AMS, 2003.
  • Optimal Transport: Old and New, Cédric Villani, Springer, 2009.
  • Optimal Transport for Applied Mathematicians, Filippo Santambrogio, Birkhäuser, 2015.
  • Gradient Flows in Metric Spaces and in the Space of Probability Measures, Luigi Ambrosio, Nicola Gigli & Giuseppe Savaré, Springer, 2006.
  • Optimal Transport Methods in Economics, Alfred Galichon, Princeton University Press, 2016.
  • Statistical Optimal Transport, Sinho Chewi, Jonathan Niles-Weed & Philippe Rigollet, 2024.

Computational references and long reviews cited in the book:

Software

Most figures in the book are generated with standard Python scientific tools, with several OT computations relying on POT, the Python Optimal Transport library cited and acknowledged in the manuscript.

About

Optimal Transport for Machine Learners

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors