Sampling-Enhanced Large Neighborhood Search for Solving Integer Linear Programs
Keywords: Integer Linear Program, Combinatorial Optimization, Large Neighborhood Search, Simulated Annealing, Locally-informed Proposals
TL;DR: We propose a sampling-enhanced neural LNS solver to tackle the "local optima" problem in existing neural LNS solvers for integer linear programs.
Abstract: Large Neighborhood Search (LNS) is a common heuristic in combinatorial optimization
that iteratively searches over a large neighborhood of the current solution for a better one. Recently, neural network-based LNS solvers have achieved great success in solving Integer Linear Program (ILP) problems
with a learnable
policy for neighborhood selection, followed by an off-the-shelf ILP solver for re-optimization.
Nonetheless, existing neural LNS solvers often get stuck in the same solution due to their greedy update strategy, i.e., only moving to the best solution found within the neighborhood. In this work, we try to theoretically identify the limitation of neural models in escaping the "local optima". Accordingly, we propose
a novel sampling-enhanced neural LNS solver, namely SPL-LNS, by reformulating LNS as a stochastic process,
which uses a locally-informed proposal to sample the next assignment and simulated annealing to alleviate the ``local optima'' issue. We also develop a novel hindsight relabeling method to efficiently train SPL-LNS on self-generated data. Experimental results reveal that our method substantially surpasses prior neural LNS solvers on multiple ILP problems.
Primary Area: optimization
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: 8542
Loading