Learning to Solve Routing Problems via Distributionally Robust Optimization

TL;DR

This paper introduces a distributionally robust training method for deep routing models that enhances their ability to generalize across different node distributions, using group DRO and a CNN-based pattern learning module.

Contribution

It proposes a novel group DRO training framework combined with a CNN module to improve cross-distribution generalization in deep routing models.

Findings

Significant improvement in generalization on synthetic and benchmark datasets.

Effective enhancement of GCN and POMO models' robustness.

Demonstrated superiority over original models in diverse distributions.

Abstract

Recent deep models for solving routing problems always assume a single distribution of nodes for training, which severely impairs their cross-distribution generalization ability. In this paper, we exploit group distributionally robust optimization (group DRO) to tackle this issue, where we jointly optimize the weights for different groups of distributions and the parameters for the deep model in an interleaved manner during training. We also design a module based on convolutional neural network, which allows the deep model to learn more informative latent pattern among the nodes. We evaluate the proposed approach on two types of well-known deep models including GCN and POMO. The experimental results on the randomly synthesized instances and the ones from two benchmark dataset (i.e., TSPLib and CVRPLib) demonstrate that our approach could significantly improve the cross-distribution generalization performance over the original models.