Reinforcement Learning for Finding QAOA Parameters

Quantum Approximate Optimization Algorithm (QAOA) is an important variational hybrid quantum-classical algorithm for approximately solving combinatorial optimization problems on NISQ devices. The quality of the solution obtained by QAOA depends on the performance of the classical optimizer used to optimize the variational parameters, but finding these optimal parameters can be hard to achieve. To address this problem, we formulate the problem of finding optimal QAOA parameters as a learning task in which the knowledge gained from solving training instances can be leveraged to find high-quality solutions for unseen test instances using two machine learning based approaches. We used reinforcement learning (RL) framework to learn a policy network to optimize QAOA circuits and a kernel density estimation (KDE) technique to learn a generative model of optimal QAOA parameters. In both approaches, the training procedure is performed on the small-sized problem instances, which can be simulated on a classical computer; yet the learned RL policy and the generative model can be used to efficiently solve larger problems. Our proposed RL and KDE based approaches reduce the optimality gap by factors up to 30 when compared to other commonly used standard optimizers.