Auxiliary Task-Based Deep Reinforcement Learning for Quantum Control.

Area of Science:

• Quantum Physics
• Artificial Intelligence
• Control Theory

Background:

• Quantum control problems are complex and often require environment-specific knowledge.
• Reinforcement learning (RL) presents a promising, model-free approach for quantum control.
• Deep deterministic policy gradient is a key RL algorithm for continuous control tasks.

Purpose of the Study:

• To investigate the effectiveness of continuous control policies for quantum systems using deep RL.
• To propose and evaluate an auxiliary task-based deep RL (AT-DRL) framework for high-fidelity quantum control.
• To enhance the exploration of quantum dynamics and improve state preparation.

Main Methods:

• Implementation of continuous control policies using deep deterministic policy gradient.
• Development of an auxiliary task to predict quantum system fidelity, sharing parameters with the main RL task.
• Design of a guided reward function based on quantum state fidelity for gradual improvement.

Main Results:

• The proposed AT-DRL framework successfully controls quantum systems with high fidelity.
• AT-DRL demonstrates faster learning rates compared to standard RL approaches.
• Numerical simulations validate the method's effectiveness in exploring quantum dynamics.

Conclusions:

• AT-DRL provides an effective solution for high-fidelity quantum control and state preparation.
• The auxiliary task aids in extracting intrinsic environmental features, improving agent performance.
• This approach shows significant potential for designing advanced quantum control pulses.