Learning Based Hardware-Centric Quantum Circuit Generation

In this paper we present an approach to find quantum circuits suitable to mimic probabilistic and search operations on a physical NISQ device. We present both a gradient based and a non-gradient based machine learning approach to optimize the created quantum circuits. In our optimization procedure we make use of a cost function that differentiates between the vector representing the probabilities of measurement of each basis state after applying our learned circuit and the desired probability vector. As such our quantum circuit generation (QCG) approach leads to thinner quantum circuits which behave better when executed on physical quantum computers. Our approach moreover ensures that the created quantum circuit obeys the restrictions of the chosen hardware. By catering to specific quantum hardware we can avoid unforeseen and potentially unnecessary circuit depth, and we return circuits that need no further transpilation. We present the results of running the created circuits on quantum computers by IBM, Rigetti and Quantum Inspire.