POT: Python Optimal Transport

By Rémi Flamary, Nicolas Courty, Alexandre Gramfort, Mokhtar Z. Alaya, Aurélie Boisbunon, Stanislas Chambon, Adrien Corenflos, Nemo Fournier, N. Gayraud, H. Janati and 8 more. Journal of machine learning research, 2021

1,118 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Continuous Control with Action Quantization from Demonstrations (2021, methodology): “We compute it using the POT library [Flamary et al., 2021] and use the Sinkhorn distance, a regularized version of the Wasserstein distance, as it is faster to compute.”
Imitating Human Behaviour with Diffusion Models (2023, methodology): “We then use the POT library’s emd2 function (Flamary et al., 2021) to compute the Wasserstein distance, using the L 2 cost function between the 30 dimensional entries.”
Independent SE(3)-Equivariant Models for End-to-End Rigid Protein Docking (2021, methodology): “The optimal transport plan is computed using an Earth Mover’s Distance and the POT library (Flamary et al., 2021), while being kept ﬁxed during back-propagation and optimization when only the cost matrix is trained.”
Improving and generalizing flow-based generative models with minibatch optimal transport (2023, unknown): “We use σ = 1 and batchsize=512.”
Participatory provenance as representational auditing for AI-mediated public consultation (2026, unknown): “The Wasserstein-2 distance [Villani et al., 2009, Flamary et al., 2021] between the full participant embedding distribution P and the summary sentence distribution Q (both in PCA-50 space) expresses the same comparison in distributional terms: W 2 = 0 .”