Our recent preprint introduces a new scalable framework to solve the Gromov-Wasserstein problem.

I am really excited about this work because it brings together several tools/problems/ideas I like:

  • the Gromov-Wasserstein distance from optimal transport,
  • functional maps from shape analysis,
  • the inverse OT problem,
  • amortized optimization,
  • relative representations in representation learning, and
  • alignment problems in single-cell biology.