Quantum Control through Reinforcement Learning

Description

Quantum technology is believed to be much more powerful than present-day computational technologies in many problems. Lying at its heart is quantum control, that is, the precise manipulation of quantum systems robust to noises. Various numerical techniques have been employed to find control protocols that achieve maximum accuracy while costing minimum resources. However, traditional algorithms such as gradient based ones and the Krotov method have limitations under various situations.On the other hand, machine-learning techniques have demonstrated great power in many problems. Reinforcement learning, a main branch of machine learning, is particularly suitable for quantum control as it casts the control problem into a "game", with rules and goals derived from the desired problem. In this game, an agent seeks for the maximum reward following the prescribed rules, which, at the same time, gives optimal solutions. The reinforcement learning have been found to be more efficient thantraditional methods under some situations.In this work, we will further reveal the power of reinforcement learning in quantum control problems. We will apply reinforcement learning to quantum gate optimization, quantum metrology, and other problems such as shortcuts to adiabaticity. As the outcome of this project, we expect that the control protocols suggested by reinforcement learning would be much easier to be experimentally implemented than those exist in the literature, which should greatly improve the accuracy of quantum control, providing asolid foundation stone for quantum technology.