pipette
ENEnglish

Embedding large subsets of Wasserstein spaces into Banach spaces

Cyril Letrouit, Quentin M\'erigot

Preprint

En palabras de los autores

Embedding subsets of Wasserstein spaces into linear spaces is a problem of both theoretical and computational interest. In this work, we prove that three classical and natural embeddings of Wasserstein spaces into Banach spaces of non-trivial type are bi-H\"older on subsets of Wasserstein spaces given by moment bounds. These embeddings, all considered in multiple applications, are the linearized optimal transport embedding, the sliced-Wasserstein embedding, and a particular kernel mean embedding. Our results contrast with known obstructions to embeddin the full Wasserstein space into Banach spaces of non-trivial type. We also show the sharpness of our results regarding the moments required for our conclusions to hold.

Resultado principalLimitación que admiten los autores

Apareció: viernes, 9 de octubre. arXiv. Preprint, todavía sin revisión por pares.