Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
102 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

GLOP: Learning Global Partition and Local Construction for Solving Large-scale Routing Problems in Real-time (2312.08224v2)

Published 13 Dec 2023 in cs.AI and cs.LG

Abstract: The recent end-to-end neural solvers have shown promise for small-scale routing problems but suffered from limited real-time scaling-up performance. This paper proposes GLOP (Global and Local Optimization Policies), a unified hierarchical framework that efficiently scales toward large-scale routing problems. GLOP partitions large routing problems into Travelling Salesman Problems (TSPs) and TSPs into Shortest Hamiltonian Path Problems. For the first time, we hybridize non-autoregressive neural heuristics for coarse-grained problem partitions and autoregressive neural heuristics for fine-grained route constructions, leveraging the scalability of the former and the meticulousness of the latter. Experimental results show that GLOP achieves competitive and state-of-the-art real-time performance on large-scale routing problems, including TSP, ATSP, CVRP, and PCTSP.

PDF HTML Abstract

Overview: GLOP - A Framework for Large-Scale Routing Problem Solutions

The paper presents GLOP, a hierarchical framework aimed at efficiently addressing large-scale routing problems in real-time. Traditional methods such as mathematical programming or iterative heuristics often fail to meet the demands of such problems in modern logistics, transportation, and robotic systems, which require rapid dispatch and handling of multiple nodes. Neural Combinatorial Optimization (NCO) approaches have shown promise for tackling routing problems, but they struggle with scaling to instances comprising thousands or tens of thousands of nodes.

GLOP Framework

GLOP is designed to operate under both global partition and local construction paradigms. Here's how the framework is structured:

  • Global Partition: GLOP initially divides larger routing problems into smaller subproblems. These are often particular instances of the Travelling Salesman Problem (TSP) or even Shortest Hamiltonian Path Problems (SHPP).
  • Local Construction: GLOP harnesses the meticulousness of autoregressive (AR) neural heuristics to refine routes at finer granularities by solving SHPP instances.

This hybridization of non-autoregressive (NAR) and AR approaches allows GLOP to leverage the strengths of both paradigms in terms of scalability and solution precision.

Results and Comparisons

The paper provides extensive experimental evaluations demonstrating GLOP's efficacy:

  1. Performance: GLOP achieves state-of-the-art real-time results for TSP and its variants, including Asymmetric TSP (ATSP), Capacitated Vehicle Routing Problem (CVRP), and Prize Collecting TSP (PCTSP). It surpasses the previous methods with significant speedups in computation time, while maintaining competitive solution quality.
  2. Scaling: GLOP is capable of solving larger instances than traditional solvers. Notably, it solves a TSP instance with 100k nodes, performing significantly faster than LKH-3 with a reasonable optimality gap of 5.1%.
  3. Cross-Distribution Performance: Unlike many existing methods, GLOP maintains consistent performance across varying scales and distributions, using the same set of learned policies without the need for scale-specific or distribution-specific tuning.

Implications and Future Directions

The implications of this research are twofold:

  • Practical Implications: GLOP can be readily applied to modern logistics systems needing real-time solutions for large routing problems. It provides a potential pathway for integrating neural solvers into industry practices, where scalability and efficiency are crucial.
  • Theoretical Implications: From a theoretical standpoint, GLOP presents a successful hybridization of AR and NAR approaches, paving the way for future research on hybrid methods in combinatorial optimization.

Future developments may focus on incorporating dynamic graph clustering techniques or more nuanced node classification tasks to enhance hierarchical routing problem solutions. Investigating the unified perspectives of AR and NAR probabilistic construction graphs could yield further efficiencies and broaden applicability.

In conclusion, GLOP stands as a robust framework that enhances the capability of neural solvers in solving large-scale routing problems. Its scalability, speed, and cross-distribution consistency mark significant progress in the field of combinatorial optimization, with promising avenues for further research and practical application.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown Bookmark Chat Bubble Oval Streamline Icon: https://streamlinehq.com Chat (Pro)
Definition Search Book Streamline Icon: https://streamlinehq.com
References (72)
  1. Constrained Clustering for the Capacitated Vehicle Routing Problem (CC-CVRP). Applied artificial intelligence, 36(1): 1995658.
  2. Concorde TSP solver.
  3. Neural combinatorial optimization with reinforcement learning. ArXiv preprint, abs/1611.09940.
  4. Rl4co: an extensive reinforcement learning for combinatorial optimization benchmark. arXiv preprint arXiv:2306.17100.
  5. Learning Generalizable Models for Vehicle Routing Problems via Knowledge Distillation. ArXiv preprint, abs/2210.07686.
  6. Learning to Solve Vehicle Routing Problems: A Survey. ArXiv preprint, abs/2205.02453.
  7. The transformer network for the traveling salesman problem. ArXiv preprint, abs/2103.03012.
  8. Reinforcement Learning for Practical Express Systems with Mixed Deliveries and Pickups. ACM Transactions on Knowledge Discovery from Data (TKDD).
  9. Select and Optimize: Learning to solve large-scale TSP instances. In Ruiz, F.; Dy, J.; and van de Meent, J.-W., eds., Proceedings of The 26th International Conference on Artificial Intelligence and Statistics, volume 206 of Proceedings of Machine Learning Research, 1219–1231. PMLR.
  10. Simulation-guided beam search for neural combinatorial optimization. ArXiv preprint, abs/2207.06190.
  11. Learning 2-opt heuristics for the traveling salesman problem via deep reinforcement learning. In Asian Conference on Machine Learning, 465–480. PMLR.
  12. Deep Reinforcement Learning for UAV Routing in the Presence of Multiple Charging Stations. IEEE Transactions on Vehicular Technology.
  13. Generalize a Small Pre-trained Model to Arbitrarily Large TSP Instances. In Thirty-Fifth AAAI Conference on Artificial Intelligence, AAAI 2021, Thirty-Third Conference on Innovative Applications of Artificial Intelligence, IAAI 2021, The Eleventh Symposium on Educational Advances in Artificial Intelligence, EAAI 2021, Virtual Event, February 2-9, 2021, 7474–7482. AAAI Press.
  14. Combining reinforcement learning and optimal transport for the traveling salesman problem. ArXiv preprint, abs/2203.00903.
  15. Population-Based Reinforcement Learning for Combinatorial Optimization. ArXiv preprint, abs/2210.03475.
  16. Helsgaun, K. 2017. An extension of the Lin-Kernighan-Helsgaun TSP solver for constrained traveling salesman and vehicle routing problems. Roskilde: Roskilde University, 24–50.
  17. Efficient Active Search for Combinatorial Optimization Problems. In The Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022. OpenReview.net.
  18. Neural large neighborhood search for the capacitated vehicle routing problem. ArXiv preprint, abs/1911.09539.
  19. Generalize Learned Heuristics to Solve Large-scale Vehicle Routing Problems in Real-time. In The Eleventh International Conference on Learning Representations.
  20. Graph Neural Network Guided Local Search for the Traveling Salesperson Problem. In The Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022. OpenReview.net.
  21. Ensemble-based Deep Reinforcement Learning for Vehicle Routing Problems under Distribution Shift. In Thirty-seventh Conference on Neural Information Processing Systems.
  22. Learning to Solve Routing Problems via Distributionally Robust Optimization. In Thirty-Sixth AAAI Conference on Artificial Intelligence, AAAI 2022, Thirty-Fourth Conference on Innovative Applications of Artificial Intelligence, IAAI 2022, The Twelveth Symposium on Educational Advances in Artificial Intelligence, EAAI 2022 Virtual Event, February 22 - March 1, 2022, 9786–9794. AAAI Press.
  23. Pointerformer: Deep Reinforced Multi-Pointer Transformer for the Traveling Salesman Problem. ArXiv preprint, abs/2304.09407.
  24. Learning the travelling salesperson problem requires rethinking generalization. Constraints, 1–29.
  25. An efficient graph convolutional network technique for the travelling salesman problem. ArXiv preprint, abs/1906.01227.
  26. Learning Collaborative Policies to Solve NP-hard Routing Problems. In Ranzato, M.; Beygelzimer, A.; Dauphin, Y. N.; Liang, P.; and Vaughan, J. W., eds., Advances in Neural Information Processing Systems 34: Annual Conference on Neural Information Processing Systems 2021, NeurIPS 2021, December 6-14, 2021, virtual, 10418–10430.
  27. Sym-NCO: Leveraging Symmetricity for Neural Combinatorial Optimization. ArXiv preprint, abs/2205.13209.
  28. Adam: A Method for Stochastic Optimization. In Bengio, Y.; and LeCun, Y., eds., 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings.
  29. Deep policy dynamic programming for vehicle routing problems. In International Conference on Integration of Constraint Programming, Artificial Intelligence, and Operations Research, 190–213. Springer.
  30. Attention, Learn to Solve Routing Problems! In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019. OpenReview.net.
  31. POMO: Policy Optimization with Multiple Optima for Reinforcement Learning. In Larochelle, H.; Ranzato, M.; Hadsell, R.; Balcan, M.; and Lin, H., eds., Advances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems 2020, NeurIPS 2020, December 6-12, 2020, virtual.
  32. Matrix encoding networks for neural combinatorial optimization. In Ranzato, M.; Beygelzimer, A.; Dauphin, Y. N.; Liang, P.; and Vaughan, J. W., eds., Advances in Neural Information Processing Systems 34: Annual Conference on Neural Information Processing Systems 2021, NeurIPS 2021, December 6-14, 2021, virtual, 5138–5149.
  33. Deep reinforcement learning for solving the heterogeneous capacitated vehicle routing problem. IEEE Transactions on Cybernetics, 52(12): 13572–13585.
  34. Deep reinforcement learning for combinatorial optimization: Covering salesman problems. IEEE transactions on cybernetics, 52(12): 13142–13155.
  35. Learning to delegate for large-scale vehicle routing. In Ranzato, M.; Beygelzimer, A.; Dauphin, Y. N.; Liang, P.; and Vaughan, J. W., eds., Advances in Neural Information Processing Systems 34: Annual Conference on Neural Information Processing Systems 2021, NeurIPS 2021, December 6-14, 2021, virtual, 26198–26211.
  36. Combinatorial Optimization with Graph Convolutional Networks and Guided Tree Search. In Bengio, S.; Wallach, H. M.; Larochelle, H.; Grauman, K.; Cesa-Bianchi, N.; and Garnett, R., eds., Advances in Neural Information Processing Systems 31: Annual Conference on Neural Information Processing Systems 2018, NeurIPS 2018, December 3-8, 2018, Montréal, Canada, 537–546.
  37. Reinforcement Graph Clustering with Unknown Cluster Number. In Proceedings of the 31st ACM International Conference on Multimedia, 3528–3537.
  38. A Survey of Deep Graph Clustering: Taxonomy, Challenge, and Application. arXiv preprint arXiv:2211.12875.
  39. SGDR: Stochastic Gradient Descent with Warm Restarts. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings. OpenReview.net.
  40. Decoupled Weight Decay Regularization. In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019. OpenReview.net.
  41. A Learning-based Iterative Method for Solving Vehicle Routing Problems. In 8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, April 26-30, 2020. OpenReview.net.
  42. Combinatorial optimization by graph pointer networks and hierarchical reinforcement learning. ArXiv preprint, abs/1911.04936.
  43. Learning to Search Feasible and Infeasible Regions of Routing Problems with Flexible Neural k-Opt. In Thirty-seventh Conference on Neural Information Processing Systems.
  44. Learning to Iteratively Solve Routing Problems with Dual-Aspect Collaborative Transformer. In Ranzato, M.; Beygelzimer, A.; Dauphin, Y. N.; Liang, P.; and Vaughan, J. W., eds., Advances in Neural Information Processing Systems 34: Annual Conference on Neural Information Processing Systems 2021, NeurIPS 2021, December 6-14, 2021, virtual, 11096–11107.
  45. Reinforcement learning for combinatorial optimization: A survey. Computers & Operations Research, 134: 105400.
  46. Unsupervised Learning for Solving the Travelling Salesman Problem. ArXiv preprint, abs/2303.10538.
  47. A cluster-first route-second approach for the swap body vehicle routing problem. Annals of Operations Research, 253: 935–956.
  48. Revised note on learning quadratic assignment with graph neural networks. In 2018 IEEE Data Science Workshop (DSW), 1–5. IEEE.
  49. H-TSP: Hierarchically Solving the Large-Scale Travelling Salesman Problem. ArXiv preprint, abs/2304.09395.
  50. DIMES: A Differentiable Meta Solver for Combinatorial Optimization Problems. ArXiv preprint, abs/2210.04123.
  51. Stochastic Economic Lot Scheduling via Self-Attention Based Deep Reinforcement Learning. IEEE Transactions on Automation Science and Engineering.
  52. Learning 3-opt heuristics for traveling salesman problem via deep reinforcement learning. In Asian Conference on Machine Learning, 1301–1316. PMLR.
  53. DIFUSCO: Graph-based Diffusion Solvers for Combinatorial Optimization. ArXiv preprint, abs/2302.08224.
  54. POPMUSIC for the travelling salesman problem. European Journal of Operational Research, 272(2): 420–429.
  55. Pointer Networks. In Cortes, C.; Lawrence, N. D.; Lee, D. D.; Sugiyama, M.; and Garnett, R., eds., Advances in Neural Information Processing Systems 28: Annual Conference on Neural Information Processing Systems 2015, December 7-12, 2015, Montreal, Quebec, Canada, 2692–2700.
  56. A Game-Theoretic Approach for Improving Generalization Ability of TSP Solvers. ArXiv preprint, abs/2110.15105.
  57. ASP: Learn a Universal Neural Solver! ArXiv preprint, abs/2303.00466.
  58. Williams, R. J. 1992. Simple statistical gradient-following algorithms for connectionist reinforcement learning. Reinforcement learning, 5–32.
  59. Learning Large Neighborhood Search Policy for Integer Programming. In Ranzato, M.; Beygelzimer, A.; Dauphin, Y. N.; Liang, P.; and Vaughan, J. W., eds., Advances in Neural Information Processing Systems 34: Annual Conference on Neural Information Processing Systems 2021, NeurIPS 2021, December 6-14, 2021, virtual, 30075–30087.
  60. Learning improvement heuristics for solving routing problems.. IEEE transactions on neural networks and learning systems.
  61. Neural Airport Ground Handling. ArXiv preprint, abs/2303.02442.
  62. An evolutionary multiobjective route grouping-based heuristic algorithm for large-scale capacitated vehicle routing problems. IEEE transactions on cybernetics, 51(8): 4173–4186.
  63. Multi-Decoder Attention Model with Embedding Glimpse for Solving Vehicle Routing Problems. In Thirty-Fifth AAAI Conference on Artificial Intelligence, AAAI 2021, Thirty-Third Conference on Innovative Applications of Artificial Intelligence, IAAI 2021, The Eleventh Symposium on Educational Advances in Artificial Intelligence, EAAI 2021, Virtual Event, February 2-9, 2021, 12042–12049. AAAI Press.
  64. NeuroLKH: Combining Deep Learning Model with Lin-Kernighan-Helsgaun Heuristic for Solving the Traveling Salesman Problem. In Ranzato, M.; Beygelzimer, A.; Dauphin, Y. N.; Liang, P.; and Vaughan, J. W., eds., Advances in Neural Information Processing Systems 34: Annual Conference on Neural Information Processing Systems 2021, NeurIPS 2021, December 6-14, 2021, virtual, 7472–7483.
  65. Memory-efficient Transformer-based network model for Traveling Salesman Problem. Neural Networks, 161: 589–597.
  66. DeepACO: Neural-enhanced Ant Systems for Combinatorial Optimization. ArXiv preprint, abs/2309.14032.
  67. Learning to solve multiple-TSP with time window and rejections via deep reinforcement learning. IEEE Transactions on Intelligent Transportation Systems.
  68. A route clustering and search heuristic for large-scale multidepot-capacitated arc routing problem. IEEE Transactions on Cybernetics, 52(8): 8286–8299.
  69. Learning to Solve Travelling Salesman Problem with Hardness-Adaptive Curriculum. In Thirty-Sixth AAAI Conference on Artificial Intelligence, AAAI 2022, Thirty-Fourth Conference on Innovative Applications of Artificial Intelligence, IAAI 2022, The Twelveth Symposium on Educational Advances in Artificial Intelligence, EAAI 2022 Virtual Event, February 22 - March 1, 2022, 9136–9144. AAAI Press.
  70. Towards Omni-generalizable Neural Methods for Vehicle Routing Problems. In the 40th International Conference on Machine Learning (ICML 2023).
  71. RBG: Hierarchically Solving Large-Scale Routing Problems in Logistic Systems via Reinforcement Learning. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 4648–4658.
  72. MAPDP: Cooperative Multi-Agent Reinforcement Learning to Solve Pickup and Delivery Problems. In Thirty-Sixth AAAI Conference on Artificial Intelligence, AAAI 2022, Thirty-Fourth Conference on Innovative Applications of Artificial Intelligence, IAAI 2022, The Twelveth Symposium on Educational Advances in Artificial Intelligence, EAAI 2022 Virtual Event, February 22 - March 1, 2022, 9980–9988. AAAI Press.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (6)
  1. Haoran Ye (12 papers)
  2. Jiarui Wang (33 papers)
  3. Helan Liang (2 papers)
  4. Zhiguang Cao (48 papers)
  5. Yong Li (628 papers)
  6. Fanzhang Li (7 papers)
Citations (20)
View on Semantic Scholar