Optimizing quantum heuristics with meta-learning

Abstract Variational quantum algorithms, a class of heuristics, are promising candidates for the demonstration useful computation. Finding best way to amplify performance these methods on hardware is an important task. Here, we evaluate optimization heuristics with existing techniques called “meta-learners.” We compare meta-learner evolutionary strategies, L-BFGS-B and Nelder-Mead approaches, two (quantum alternating operator ansatz variational eigensolver), three problems, in simulation environments. show that comes near global optima more frequently than all other optimizers tested noisy parameter setting environment. also find generally resistant noise, example, seeing smaller reduction Noisy Sampling environments, performs better average by “gain” metric its closest comparable competitor L-BFGS-B. Finally, present evidence indicates trained small problems will generalize larger problems. These results indication meta-learning associated machine learning be integral application near-term computers.