Reproduce “A Reinforcement-Learning-Based Multiple-Column Selection Strategy for Column Generation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Haofeng Yuan, Lichang Fang, Shiji Song

A Reinforcement Learning (RL) strategy is introduced for selecting multiple columns in Column Generation (CG), demonstrating consistent reductions in total iterations and runtime for Cutting Stock and Graph Coloring Problems. The approach generalizes to larger problem instances and outperforms established heuristic and imitation learning baselines.

Key insights

  • Framing column selection as an RL problem enables optimization for the long-term goal of minimizing total CG iterations, leading to superior overall efficiency compared to short-sighted, single-step objective improvements.
  • Explicitly modeling the interdependencies and structural relationships between candidate columns using graph neural networks is crucial for effective multiple-column selection.
  • The learned RL policy can generalize to significantly larger and unseen problem instances, indicating that it captures fundamental, size-invariant principles of column selection.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.