End-to-End Quantum Machine Learning with Quantum Control Systems

This work presents a hardware-friendly end-to-end quantum machine learning scheme that can be implemented with imperfect near-term intermediate-scale quantum processors. The proposal transforms the machine learning task to the optimization of a quantum control system, which parameterize the learning model by experimentally tunable control variables. Our design also enables automated feature selection by encoding the raw input data to quantum states through agent control variables. Comparing with the gate-based parameterized quantum circuits, the resulting end-to-end quantum learning models is easy to implement as there are only few ad-hoc parameters to be determined by the designer. Numerical simulations on the benchmarking MNIST dataset without down-sampling the images demonstrate that the proposed scheme can achieve comparable high performance with only 3-5 qubits than known quantum machine learning models. The scheme is promising for efficiently performing large-scale real-world learning tasks using intermediate-scale quantum processors.