Improving Generalization of Deep Reinforcement Learning-based TSP Solvers

TL;DR

This paper introduces MAGIC, a novel deep reinforcement learning approach with a specialized architecture and training techniques, significantly improving the generalization and efficiency of TSP solvers on larger instances.

Contribution

MAGIC combines a new neural architecture and training method, including curriculum learning and local search, to enhance DRL-based TSP solvers' performance and generalization.

Findings

MAGIC outperforms existing DRL-based TSP solvers on random instances.

MAGIC generalizes better to larger TSP instances.

MAGIC is competitive with traditional heuristics in performance and speed.

Abstract

Recent work applying deep reinforcement learning (DRL) to solve traveling salesman problems (TSP) has shown that DRL-based solvers can be fast and competitive with TSP heuristics for small instances, but do not generalize well to larger instances. In this work, we propose a novel approach named MAGIC that includes a deep learning architecture and a DRL training method. Our architecture, which integrates a multilayer perceptron, a graph neural network, and an attention model, defines a stochastic policy that sequentially generates a TSP solution. Our training method includes several innovations: (1) we interleave DRL policy gradient updates with local search (using a new local search technique), (2) we use a novel simple baseline, and (3) we apply curriculum learning. Finally, we empirically demonstrate that MAGIC is superior to other DRL-based methods on random TSP instances, both in terms of performance and generalizability. Moreover, our method compares favorably against TSP heuristics and other state-of-the-art approach in terms of performance and computational time.