Learning to learn with quantum neural networks via classical neural networks

Quantum Neural Networks (QNNs) are a promising variational learning paradigm applications to near-term quantum processors, however they still face some challenges. One such challenge is finding good parameter heuristics that ensure rapid and consistent convergence to local of the parameterized quantum circuit landscape. In this work, we train neural networks to assist in the quantum learning process, also know meta-learning, to rapidly find approximate optima in the parameter landscape several classes of quantum variational algorithms. Specifically, we train recurrent neural networks to find approximately optimal parameters a small number of queries of the cost function for the Quantum Optimization Algorithm (QAOA) for MaxCut, QAOA for-Kirkpatrick Ising model, and for a Variational Quantum Eigensolver the Hubbard model. By initializing other optimizers at parameter values by the classical neural network, we demonstrate a significant in the total number of optimization iterations required to reach a accuracy. We further demonstrate that the optimization strategies learned the neural network generalize well across a range of problem instance sizes. opens up the possibility of training on small, classically simulatable instances, in order to initialize larger, classically intractably problem instances on quantum devices, thereby significantly the number of required quantum-classical optimization iterations.