Reinforcement Learning for Variational Quantum Circuit Design

Variational Quantum Algorithms (VQAs) are widely used for solving optimization problems in the Noisy Intermediate-Scale Quantum (NISQ) era. However, designing effective quantum circuits (ansatzes) that are compatible with the limitations of current quantum hardware remains a significant challenge. In this work, we introduce a Reinforcement Learning (RL) agent that autonomously generates ansatzes for VQAs. The RL agent is trained on several optimization problems, including Maximum Cut, Maximum Clique, and Minimum Vertex Cover, across different graph topologies. Our results show that the agent is able to generate effective quantum circuits, with approximation ratios that favorably compare to commonly used ansatzes. Additionally, we identify a novel family of ansatzes, termed “Ryz-connected”, particularly effective on Maximum Cut problems. These findings highlight the potential of RL techniques in designing efficient quantum circuits for a broad class of applications in quantum computing.