BTBS-LNS: A Binarized-Tightening, Branch and Search Approach of Learning Large Neighborhood Search Policies for MIP

Hao Yuan, Wenli Ouyang, Changwen Zhang, Yong Sun, Liming Gong, Ziao Guo, Zhichen Dong, Junchi Yan

Keywords: large neighborhood search, bound tightening, hybrid branch and search, reinforcement learning

Abstract: Learning to solve large-scale Mixed Integer Program (MIP) problems is an emerging research topic, and policy learning-based Large Neighborhood Search (LNS) has recently shown its effectiveness. However, prevailing approaches predominantly concentrated on binary variables and local search strategies, often susceptible to becoming ensnared in local optima. In response to these challenges, we introduce a novel technique, termed Binarized-Tightening Branch-and-Search LNS (BTBS-LNS). Specifically, we propose the ``Binarized Tightening" technique for integer variables to deal with their wide range by encoding and bound tightening, and design an attention-based tripartite graph to capture global correlations within MIP instances. Furthermore, we devised an extra branching network at each step, to identify and optimize some wrongly-fixed backdoor variables by pure LNS. We empirically show that our approach can effectively escape local optimum. Extensive experiments on different problems, including instances from Mixed Integer Programming Library (MIPLIB), show that it significantly outperforms the open-source solver SCIP and LNS baselines. It performs competitively with, and sometimes even better than the commercial solver Gurobi (v9.5.0), especially at an early stage. Source code will be made publicly available.

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/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: 5004