Scaling Combinatorial Optimization Neural Improvement Heuristics with Online Search and Adaptation

TL;DR

This paper presents Limited Rollout Beam Search (LRBS), a novel search strategy that enhances deep reinforcement learning heuristics for combinatorial optimization, improving performance, generalization, and adaptability across problem sizes and variants.

Contribution

Introduction of LRBS, a new beam search method that significantly improves DRL-based heuristics for TSP and related problems, with online and offline adaptation capabilities.

Findings

LRBS outperforms existing heuristics in optimality gaps.

LRBS generalizes well to larger problem instances.

Adaptive search with LRBS surpasses recent methods.

Abstract

We introduce Limited Rollout Beam Search (LRBS), a beam search strategy for deep reinforcement learning (DRL) based combinatorial optimization improvement heuristics. Utilizing pre-trained models on the Euclidean Traveling Salesperson Problem, LRBS significantly enhances both in-distribution performance and generalization to larger problem instances, achieving optimality gaps that outperform existing improvement heuristics and narrowing the gap with state-of-the-art constructive methods. We also extend our analysis to two pickup and delivery TSP variants to validate our results. Finally, we employ our search strategy for offline and online adaptation of the pre-trained improvement policy, leading to improved search performance and surpassing recent adaptive methods for constructive heuristics.