Modernizing Quantum Annealing II: Genetic Algorithms and Inference

I discuss how a quantum annealer can be used for an inference task to find the likely lowest energy state for classical optimization problems using a system of quantum bits, with uncertainty values for the state of individual qubits and/or clusters of qubits, depending on the structure of the driver Hamiltonian. I provide numerical evidence that annealer outputs provide information, not only about the solution of a problem but also about the relative uncertainty of different data bits. These calls to an annealer can be used as a computational primitive for hybrid computations. In particular, such a computational primitive can be used quite naturally as a directed mutation engine for genetic algorithms. I discuss how this can be done, as well as other ways in which such annealer calls can be applied to solving problems. I consider the practicality of implementing such a protocol on real devices, and demonstrate that these methods are compatible with many of the current efforts by others to improve the performance of annealers, including the use of non-stoquastic drivers, synchronizing freeze times for individual bits, and belief propagation techniques which allow problems to be solved which do not fit on the hardware graph.