Superconducting quantum computing optimization based on multi-objective deep reinforcement learning.

Area of Science:

• Quantum computing
• Artificial intelligence
• Optimization theory

Background:

• Deep reinforcement learning (DRL) is effective for quantum optimization and optimal control.
• Precise quantum measurements necessitate simulation control across multiple experimental stages.
• Current methods may yield local optima, limiting control over complex quantum systems.

Purpose of the Study:

• To enhance multi-process quantum optimal control using a novel DRL approach.
• To develop a method that achieves global optimal solutions considering multiple factors.
• To improve the precision of quantum measurements through advanced simulation control.

Main Methods:

• Improved a multi-objective deep reinforcement learning (DRL) method based on mathematical convex optimization.
• Utilized single-process quantum control optimization results as a truncation threshold and reward function transfer strategy.
• Applied the method to optimize microwave pulse parameters for superconducting qubits.

Main Results:

• Achieved global optimal solutions for multi-process quantum optimal control, surpassing local optima.
• Demonstrated excellent computational results on superconducting qubit systems.
• Provided a set of global parameter values and control strategies for optimum control.

Conclusions:

• The improved DRL method effectively addresses multi-process quantum optimal control challenges.
• This approach enables precise control over complex quantum systems, leading to better measurement outcomes.
• The findings offer a pathway to achieving global optimum control in quantum computing applications.