Key points are not available for this paper at this time.
Abstract Quantum computing has recently emerged as a promising approach to solve complex combinatorial optimization problems beyond the reach of classical algorithms. This study explores the integration of machine learning techniques with the Quantum Approximate Optimization Algorithm (QAOA) to improve its effectiveness in graph-based optimization tasks, with a particular focus on the Max-Cut problem. We propose a hybrid framework that utilizes a neural network to dynamically adjust QAOA parameters, leveraging historical data to enhance performance. We conduct a comparative analysis between the machine learning-augmented QAOA and a traditional QAOA implementation. Through extensive simulations on random graphs with up to 2000 nodes, the machine learning-enhanced QAOA demonstrates superior performance, reducing training loss significantly and achieving a final loss of approximately 3.112 x 10-33 compared to 2.34 x 10-15 for the standard QAOA. Statistical analysis reveals a moderate negative correlation between training epochs and loss, highlighting the benefits of continued training while acknowledging other influencing factors. This research highlights the potential of combining quantum algorithms with machine learning to solve real-world optimization problems, offering promising applications in fields such as logistics, finance, and network design.
Naim Ajlouni (Tue,) studied this question.