Add Annealed Sinkhorn tutorial notebook, based on Chizat 2024

[elyanrgn]

Summary

This PR adds a new tutorial notebook on annealed Sinkhorn under docs/tutorials/linear/.

The notebook introduces entropy-regularized optimal transport and shows how annealing the regularization parameter can improve the speed-accuracy trade-off in Sinkhorn iterations. It includes:

• fixed-ε Sinkhorn baselines and Pareto-front comparisons,
• geometric schedules using OTT-JAX's native Epsilon scheduler,
• polynomial schedules motivated by convergence theory,
• an OTT-JAX adaptation of a debiased annealed variant.
• an experience on convergence conditions.

Notes

• The debiased section is presented as an OTT-JAX adaptation rather than a strict reproduction of the reference algorithm.
• The content focuses on practical intuition, implementation details, and empirical comparisons.

Submitted as part of the OTT-ENSAE 2025–2026 academic assignment.
Rougon Elyan

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB