The Expressive Power of Binary Submodular Functions
Abstract
We investigate whether all Boolean submodular functions can be decomposed into a sum of binary submodular functions over a possibly larger set of variables. This question has been considered within several different contexts in computer science, including computer vision, artificial intelligence, and pseudo-Boolean optimisation. Using a connection between the expressive power of valued constraints and certain algebraic properties of functions, we answer this question negatively.
- Publication:
-
Lecture Notes in Computer Science
- Pub Date:
- 2009
- DOI:
- arXiv:
- arXiv:0811.1885
- Bibcode:
- 2009LNCS.5734..744Z
- Keywords:
-
- Computer Science - Discrete Mathematics;
- Computer Science - Artificial Intelligence;
- Computer Science - Computer Vision and Pattern Recognition;
- F.2.2;
- G.2.1;
- I.2.4;
- I.4.0;
- I.5.1
- E-Print:
- 16 pages