Reinforcement learning assisted recursive QAOA.

Area of Science:

• Quantum Computing
• Combinatorial Optimization
• Machine Learning

Background:

• Variational quantum algorithms like QAOA show promise for NISQ devices in optimization.
• Low-depth QAOA performance is limited by locality constraints.
• Recursive QAOA (RQAOA) was proposed to overcome these limitations but is less understood.

Purpose of the Study:

• To identify failure cases of depth-1 RQAOA for combinatorial optimization problems.
• To propose a novel algorithm, RL-RQAOA, that enhances RQAOA performance.
• To demonstrate the synergy between reinforcement learning and quantum optimization.

Main Methods:

• Analysis of depth-1 RQAOA performance on specific problem instances.
• Development of a reinforcement learning enhanced RQAOA (RL-RQAOA).
• Comparative performance evaluation of RQAOA and RL-RQAOA.

Main Results:

• Identified specific instances where depth-1 RQAOA underperforms.
• RL-RQAOA demonstrates improved performance over RQAOA on these instances.
• RL-RQAOA matches RQAOA performance on instances where RQAOA is near-optimal.

Conclusions:

• RL-RQAOA offers a significant improvement over RQAOA for certain combinatorial optimization problems.
• Reinforcement learning can enhance quantum optimization algorithms.
• This work paves the way for developing more effective quantum heuristics.