PhD Student Gloria Turati

Reinforcement Learning for Quantum Circuit Design

Most research has focused on using Quantum Computing to accelerate tasks such as optimization, simulation, and more recently, machine learning and artificial intelligence (AI). However, there has been less focus on employing machine learning or AI to support quantum computers by tackling tasks that are challenging to perform and to mitigate its current limitations. We focus in particular on Variational Quantum Algorithms to tackle optimization problems. These algorithms rely on a parametric quantum circuit, known as an ansatz, optimized by a classical algorithm. However, how to design efficient ansatzes for specific problems that are resilient to noise and account for the hardware limitations of current generation devices remains a significant challenge. This talk will present our ongoing work on how to use Reinforcement Learning (RL) to design new ansatzes for variational quantum algorithms. The RL agent was trained on various optimization problems, including Maximum Cut, Maximum Clique, and Minimum Vertex Cover, using different graph topologies. Our study shows that the RL agent was able to discover useful quantum circuits, with approximation ratios that favorably compare to commonly used ansatzes . These results highlight the potential of RL techniques in designing efficient quantum circuits and their broad applicability in quantum computing, opening new directions in the generation of efficient ansatzes.