Quantum approximate optimization algorithm parameter prediction using a convolutional neural network

Abstract The Quantum approximate optimization algorithm (QAOA) is a quantum-classical hybrid aiming to produce solutions for combinatorial problems. In the QAOA, quantum part prepares parameterized state that encodes solution, where parameters are optimized by classical optimizer. However, it difficult find optimal when circuit becomes deeper. Hence, there numerous active research on performance and cost of QAOA. this work, we build convolutional neural network predict depth p + 1 QAOA instance from counterpart. We propose two strategies based model. First, recurrently apply model generate set initial values certain It successfully initiates 10 instances, whereas each only trained with depths less than 6. Second, applied repetitively until maximum expected value reached. An average approximation ratio 0.9759 Max-Cut over 264 Erdős–Rényi graphs obtained, while optimizer adopted generating first input