Reinforcement learning assisted quantum adiabatic algorithm design

We develop a framework to optimize the adiabatic quantum algorithm for prime factorization using a quadratic Ising Hamiltonian encoding. In comparing the quantum adiabatic algorithm with the classical simulated annealing methods, we find rare problem instances which are difficult to solve for both classical and quantum annealing. We then adopt deep reinforcement learning directly targeting the rare difficult problem instances. By machine learning against rare problem instances, the overall performance of the quantum adiabatic algorithm is substantially improved. This provides a novel approach for quantum algorithm design with reinforcement learning.