DDRL: A DIFFUSION-DRIVEN REINFORCEMENT LEARNING APPROACH FOR ENHANCED TSP SOLUTIONS
Keywords: TSP, Diffusion model, Reinforcement Learning
Abstract: The Traveling Salesman Problem (TSP) is a fundamental challenge in combinatorial optimization, known for its NP-hard complexity. Reinforcement Learning (RL) has proven to be effective in managing larger and more complex TSP instances, yet it encounters challenges such as training instability and necessity for a substantial amount of training resources. Diffusion models, known for iteratively refining noisy inputs to generate high-quality solutions, offer scalability and exploration capabilities for TSP but may struggle with optimality in complex cases and require large, resource-intensive training datasets. To address these limitations, we propose DDRL (Diffusion-Driven Reinforcement Learning), which integrates diffusion models with RL. DDRL employs a latent vector to generate an adjacency matrix, merging image and graph learning within a unified RL framework. By utilizing a pre-trained diffusion model as a prior, DDRL exhibits strong scalability and enhanced convergence stability. We also provide theoretical analysis that training DDRL aligns with the diffusion policy gradient in the process of solving the TSP, demonstrating its effectiveness. Additionally, we introduce novel constraint datasets—obstacle, path, and cluster constraints—to evaluate DDRL's generalization capabilities. We demonstrate that DDRL offers a robust solution that outperforms existing methods in both basic and constrained TSP problems.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 4508
Loading