Reinforcement Learning Based Quantum Circuit Optimization via ZX-Calculus - Jan Nogué Gómez

arXiv: https://arxiv.org/pdf/2312.11597.pdf

Abstract: We propose a novel Reinforcement Learning (RL) method for optimizing quantum circuits using the graph-like representation of a ZX-diagram. The agent, trained using the Proximal Policy Optimization (PPO) algorithm, employs Graph Neural Networks to approximate the policy and value functions. We test our approach for two differentiated circuit size regimes of increasing relevance, and benchmark it against the best-performing ZX-calculus based algorithm of the PyZX library, a state-of-the-art tool for circuit optimization in the field. We demonstrate that the agent can generalize the strategies learned from 5-qubit circuits to 20-qubit circuits of up to 450 Clifford gates, with enhanced compressions with respect to its counterpart while remaining competitive in terms of computational performance.

Joint-Work With: Jordi Riu, Gerard Vilaplana, Artur Garcia-Saez, and Marta P. Estarellas

Presented at the ZX-calculus seminar on the 17th of January 2024.