The quantum-classical crossover in machine learning

Leveraging the valuable properties of quantum mechanics such as superposition, entanglement, and interference has the potential to provide dramatic advantages over current classical approaches to machine learning in terms of speed, efficiency, and data encoding. They may also display distinct advantages in the analysis of fundamentally quantum data generated by quantum systems. However, it remains to be seen whether quantum machine learning algorithms can provide such advantages in a consistent and scalable way. We examine this question by designing and analyzing a quantum circuit that can be continuously tuned between classical and quantum modes of operation. In the classical mode the circuit behaves as a classical multilayer perceptron with binarized activations. Through classical simulation of this quantum circuit we determine the capabilities of the circuit as a function of its "quantumness." We also explore the execution of the circuit on current commercially available quantum computation platforms to solve simple problems requiring a small number of neurons.