Semi-supervised Diffusion Solver for Travelling Salesman Problem
Keywords: Travelling Salesman Problem, Semi-supervised Learning, Diffusion Model
Abstract: We propose a semi-supervised diffusion solver for solving the Travelling Sales- man Problem (TSP). Data-driven combinatorial optimization models recently at- tract an amount of attention, since they have shown promising results in solving various NP-hard problems without too much expert knowledge. However, most of them rely on reinforcement learning (RL) and supervised learning (SL) which face some intractable challenges: RL methods often encounter sparse reward problems and SL methods pose a strict assumption that the optimal solution (label) is always available. To address these challenges in arbitrarily large-scale TSP, this article proposes a novel semi-supervised learning-based diffusion framework towards a more general situation, i.e., we can freely produce instances as many as possible but the acquisition of optimal solution is costly. This semi-supervised paradigm is made viable by modeling the generative process upon a special transition matrix, which facilitates the effective learning of the generative diffusion, compared with learning the heatmap directly like other solvers do. Comprehensive experiments validate our method across various scales TSP, showing that our method remarkably outperforms state-of-the-art data-driven solvers on large benchmark datasets for Traveling Salesman Problems, and has an outstanding generalization ability.
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
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/2024/AuthorGuide.
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: 4355
Loading