Learning Large Neighborhood Search for Vehicle Routing in Airport Ground Handling

Abstract

published version features the final layout of the paper including the volume, issue and page numbers. Link to publication General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal.If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the "Taverne" license above, please follow below link for the End User

Learning to Generate Hard Instances: Towards Robust Solutions for Vehicle Routing Problems

“…Various deep models for VRPs are attacked, indicating the generality of the proposed framework. Specifically, we learn hard instances to attack POMO [28], Simulation-guided Beam Search (SGBS) [41], Adaptive Multi-distribution Knowledge Distillation (AMDKD) [10], and Omni-VRP [13], which serve as the environment models, respectively. These pretrained models are available at their repositories.…”

“…End-to-end models often deliver comparable solutions to learning-augmented models with reduced inference time and are more researched in the literature. Some of the above models are further enhanced in terms of the generalization across different distributions or sizes [31], [6], [9], [10], [32], [11], [12], [13], [14]. However, they simply rely on additional training on instances with manually specified distributions (e.g., Uniform, Gaussian, Diagonal distributions) or sizes (e.g., random numbers of nodes within [50,200]).…”

“…2) Zero-shot Generalizability: To assess zero-shot generalizability, we test the RET-trained POMO and baselines on the dataset which incorporates different distributions. Besides Uniform distribution, we employ commonly used distributions in the literature, including Gaussian mixture distribution [13], Cluster distribution [7], Diagonal distribution [33], and Explosion distribution [46]. We produce 1000 instances per distribution for the test.…”

“…both [13], [14]. In general, the above efforts simply train a deep model on instances with more than one random distribution or size, rendering it more generalizable to a broader range of VRP instances.…”

Learning Encodings for Constructive Neural Combinatorial Optimization Needs to Regret

“…on nine datasets of different scales and distributions. Results demonstrate that compared to POMO, and Sym-NCO, the proposed POMO-Regret significantly narrows the gaps in all nine datasets Zhou et al (2023). provides more datasets for generalization tests and in terms of the results in Appendix D of the SupplementaryThe Thirty-Eighth AAAI Conference on Artificial Intelligence (AAAI-24) The average solution costs (Cost), average optimal gaps (Gap), and total inference times (Time) on 10,000 uniform TSP instances and 10,000 CVRP instances.…”