Quantum architecture search via truly proximal policy optimization.

Area of Science:

• Quantum Computing
• Artificial Intelligence
• Algorithm Optimization

Background:

• Quantum Architecture Search (QAS) automates quantum circuit design using intelligent algorithms.
• Existing deep reinforcement learning methods like QAS-PPO lack strict policy ratio control and trust domain constraints, hindering performance.
• The need for more robust and efficient QAS methods is critical for advancing quantum computing.

Purpose of the Study:

• To introduce a novel deep reinforcement learning-based QAS method, QAS-TR-PPO-RB, for automated quantum gate sequence generation.
• To enhance policy stability and performance by incorporating trust region constraints and rollback mechanisms.
• To demonstrate superior performance and efficiency compared to previous QAS approaches.

Main Methods:

• Developed Trust Region-based Proximal Policy Optimization with Rollback (QAS-TR-PPO-RB) for QAS.
• Implemented an improved clipping function for rollback behavior, limiting policy ratio changes.
• Utilized trust domain triggering conditions to ensure policy optimization within defined boundaries.

Main Results:

• QAS-TR-PPO-RB demonstrated improved policy performance in experiments on multi-qubit circuits.
• The new method achieved a lower algorithm running time compared to the original QAS-PPO.
• Guaranteed monotone improvement was observed due to policy optimization within the trust domain.

Conclusions:

• QAS-TR-PPO-RB offers a more stable and efficient approach to Quantum Architecture Search.
• The method effectively addresses limitations of previous deep reinforcement learning-based QAS techniques.
• This advancement holds promise for accelerating the design and optimization of quantum circuits.