Learning Elimination Ordering for Tree Decomposition Problem

Taras Khakhulin, Roman Schutski, Ivan Oseledets

Keywords: Learning Heuristics, Reinforcement Learning, Graph Neural Networks

TL;DR: Reinforcement Learning-based approach to approximately solve the Tree Decomposition problem

Abstract:

We propose a Reinforcement Learning-based approach to approximately solve the Tree Decomposition problem.
Recently, it was shown that learned heuristics could successfully solve combinatorial problems.
We establish that our approach successfully generalizes from small graphs, where an optimal Tree Decomposition can be found by exact algorithms, to large instances of practical interest, while still having very low time-to-solution.
On the other hand, the agent-based approach surpasses all classical greedy heuristics by the quality of the solution.