Efficient Machine Learning Generation of Target Quantum Processes

This work addresses the problem of developing a machine-learning approach for the generation of a desired quantum process, i.e., a predefined trajectory in the set of quantum channels. To this end, a physical model of N-level quantum systems driven by coherent and incoherent controls is constructed, in which both Hamiltonian and dissipative dynamics are controlled. The main challenge in generating a target quantum process is the high complexity of finding optimal controls with conventional local-search optimization methods. To efficiently solve this problem, we develop a machine learning method that leverages a special representation of the dynamics. The proposed machine learning method achieves high efficiency and fast execution times: it delivers significantly higher accuracy (by an order of magnitude) and accelerates the generation of target quantum processes compared to standard local search methods for qubit, qutrit, and qudit (five-level) systems.