Description.
LUBP is a software that implements a heuristic for the construction of unitary branching programs (UBPs) consistent with a given dataset. The idea is that such a UBP may be regarded as a point in a complex Stiefel manifold of appropriate dimension, while the search for an UBP consistent with a given dataset may be regarded as a minimization problem over this manifold. We approach this minimization problem by combining Riemannian gradient descent theory with local optimization.
Download.
Please visit github.com/autoproving/lubp for instructions on how to download, install and use our program.
Citation.
The algorithm implemented in our program is described in detail in the following publication, where you will also find experimental results and a theoretical study of unitary branching programs.
- Unitary Branching Programs: Learnability and Lower Bounds.
Fidel Ernesto Diaz Andino, Maria Kokkou, Mateus de Oliveira Olieira, Farhad Vadiee.
Proceedings of the 38th International Conference on Machine Learning.