About this document
Differentiable Birkhoff-von Neumann Decomposition by hans wurst is a document available to read on EtoBox.
This paper proposes a novel differentiable approach to approximate the Birkhoff-von-Neumann (BvN) decomposition of a doubly stochastic matrix. The BvN decomposition represents a doubly stochastic matrix as a convex combination of permutation matrices. The proposed method uses Riemannian gradient descent on the Birkhoff polytope to learn sub-permutation matrices that approximate the BvN decomposition. This differentiable approach allows precomputing an approximated BvN decomposition offline for applications
- Author
- hans wurst
- Language
- EN