Enhancing Optimization Techniques with Quantum-Inspired Generative Models (Part 1)

Large-scale integer combinatorial problems represent some of the most commonly occurring optimization problems in industrial settings. Quantum-inspired optimizers based on tensor networks can find unique optimization routes that may solve these problems faster than traditional approaches. In this work, we utilize such a quantum-inspired optimizer to enhance traditional optimization methods and analyze performance on a BMW plant optimization problem. Specifically, we investigate optimizer performance under basic data encodings and parameterizations. We also explore a subspace of the hyperparameters for the quantum-inspired optimizer and show that a maximum performance can be achieved as compared to other hyperparameter configurations. Finally, we compile these datasets to show the limits of quantum-inspired improvement of traditional optimization methods in cases of little problem-knowledge.