MIP-GNN: A Data-Driven Framework for Guiding Combinatorial Solvers
نویسندگان
چکیده
Mixed-integer programming (MIP) technology offers a generic way of formulating and solving combinatorial optimization problems. While generally reliable, state-of-the-art MIP solvers base many crucial decisions on hand-crafted heuristics, largely ignoring common patterns within given instance distribution the problem interest. Here, we propose MIP-GNN, general framework for enhancing such with data-driven insights. By encoding variable-constraint interactions mixed-integer linear program (MILP) as bipartite graph, leverage graph neural network architectures to predict variable biases, i.e., component-wise averages (near) optimal solutions, indicating how likely will be set 0 or 1 in solutions binary MILPs. In turn, predicted biases stemming from single, once-trained model are used guide solver, replacing heuristic components. We integrate MIP-GNN into applying it tasks node selection warm-starting, showing significant improvements compared default setting solver two classes challenging Our code appendix publicly available at https://github.com/lyeskhalil/mipGNN.
منابع مشابه
A Human Computation Framework for Boosting Combinatorial Solvers
We propose a general framework for boosting combinatorial solvers through human computation. Our framework combines insights from human workers with the power of combinatorial optimization. The combinatorial solver is also used to guide requests for the workers, and thereby obtain the most useful human feedback quickly. Our approach also incorporates a problem decomposition approach with a gene...
متن کاملReoptimization Techniques in MIP Solvers
Recently, there have been many successful applications of optimization algorithms that solve a sequence of quite similar mixed-integer programs (MIPs) as subproblems. Traditionally, each problem in the sequence is solved from scratch. In this paper we consider reoptimization techniques that try to benefit from information obtained by solving previous problems of the sequence. We focus on the ca...
متن کاملa new approach to credibility premium for zero-inflated poisson models for panel data
هدف اصلی از این تحقیق به دست آوردن و مقایسه حق بیمه باورمندی در مدل های شمارشی گزارش نشده برای داده های طولی می باشد. در این تحقیق حق بیمه های پبش گویی بر اساس توابع ضرر مربع خطا و نمایی محاسبه شده و با هم مقایسه می شود. تمایل به گرفتن پاداش و جایزه یکی از دلایل مهم برای گزارش ندادن تصادفات می باشد و افراد برای استفاده از تخفیف اغلب از گزارش تصادفات با هزینه پائین خودداری می کنند، در این تحقیق ...
15 صفحه اولذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Proceedings of the ... AAAI Conference on Artificial Intelligence
سال: 2022
ISSN: ['2159-5399', '2374-3468']
DOI: https://doi.org/10.1609/aaai.v36i9.21262