Project: "Quantum Methods and Benchmarks for Resource Allocation (QuBRA)"; Teilvorhaben: "Maschinelles Lernen und Quantenalgorithmen für Optimierungsprobleme"

Abstract

This final report summarizes the objectives, methodology, and outcomes of the LUH subproject “Machine Learning and Quantum Algorithms for Optimization Problems” within the BMBF-funded consortium project QuBRA – Quantum Methods and Benchmarks for Resource Allocation. The overarching goal of QuBRA was to assess, in a rigorous and application-driven manner, whether quantum algorithms—particularly on near-term and future quantum hardware—can provide a practical advantage for industrially relevant combinatorial optimization problems. The LUH subproject combined complementary expertise from machine learning and quantum information theory to address five industrial application domains: vehicle configuration, job-shop scheduling, supply chain management, fleet management, and pick-up-and-delivery problems in the Internet of Things. Strong classical and machine-learning baselines were developed using reinforcement learning and graph neural networks, while novel quantum algorithmic approaches were designed, with a focus on tailored variants of the Quantum Approximate Optimization Algorithm (QAOA), hybrid classical–quantum workflows, and problem-specific encodings. All approaches were evaluated within a unified benchmarking methodology and integrated into the consortium-wide framework QuBRABench and the Benchmark Instances Project. The results demonstrate that advanced machine learning methods currently set the state of the art for the considered problems, while quantum algorithms show long-term potential but no near-term advantage on NISQ devices. Beyond scientific contributions, the project yielded peer-reviewed publications, patent applications, open-source benchmarking tools, and supported technology transfer, including the founding of a start-up. Overall, the LUH subproject provides a realistic and methodologically sound assessment of the opportunities and limitations of quantum optimization for industrial applications and lays a solid foundation for future research and exploitation.