Squeezed states, characterized by the reduction of quantum fluctuations in specific quadratures, represent a vital resource for quantum metrology and information processing. A prominent platform for generating such states is the cavity optomechanical system, where mechanical oscillators can be prepared in squeezed states via tailored control. While periodic driving has served as a foundational technique for producing strong mechanical squeezing in this system, we propose an enhanced method based on deep reinforcement learning (DRL). Using both deep $Q$ network (DQN) and proximal policy optimization (PPO) algorithms, we optimize time-dependent control sequences. As a result, the machine-learning-driven control can surpass fundamental bounds imposed by traditional quantum control strategies due to the suppression of counter-rotating wave terms in the dynamics, with PPO achieving stronger squeezing than DQN. Moreover, our DRL-based scheme demonstrates robustness against both cavity decay and time delay in the control loop. This work presents the mapping of covariance matrix dynamics to reinforcement learning, significantly expanding artificial intelligence applications in continuous-variable quantum systems.

Close

PrevNext

View figure in article

• Machine learning
• Optomechanics
• Quantum optics