Quantum approximate optimization algorithm applied to the binary perceptron

We apply digitized quantum annealing (QA) and approximate optimization algorithm (QAOA) to a paradigmatic task of supervised learning in artificial neural networks: the synaptic weights for binary perceptron. At variance with usual QAOA applications MaxCut, or spin-chains ground-state preparation, here classical cost function is characterized by highly nonlocal multispin interactions. Yet, we provide evidence existence optimal smooth solutions parameters, which are transferable among typical instances same problem, prove numerically an enhanced performance over traditional QA. also investigate on role cost-function landscape geometry this problem. By artificially breaking geometrical structure, show that detrimental effect gap-closing transition, encountered QA, negatively affecting our implementation.