A flexible initializer for parametrized quantum circuits

Compared to fault-tolerant quantum computing, variational quantum algorithms (VQAs) hold the hope for a quantum advantage of practical relevance in a not too distant future. However, optimization of circuit parameters remains arduous and is impeded by many obstacles such as the presence of barren plateaus and many local minima in the optimization landscape. Hence, developing more efficient strategies for training parametrized quantum circuits (PQCs) is needed to unlock the full potential offered by VQAs. Extending ideas from the field of meta-learning, we address this task from an initialization perspective, and propose a FLexible Initializer for Parametrized quantum circuits (FLIP) scheme which can be applied to any family of PQCs. FLIP is tailored to learn the structure of successful parameters from a small number of related problems used as the training set. Once trained it can be used on similar problems and show a dramatic advantage over random initialization and also over more involved meta-learning initialization strategies. Furthermore FLIP accommodates quantum circuits of arbitrary sizes and we show that it can be employed on circuits larger than the ones seen during training: a feature lacking in other meta-learning parameter initializing strategies proposed to date.