Automatic Discovery of Quantum Error Correction Codes with a Noise-Aware Reinforcement Learning Agent

Quantum error correction (QEC) is required to realize reliable quantum computing and communication tasks. Constructing codes is a complex task that has historically been powered by human creativity with the discovery of a large zoo of families of codes. However, while most codes are constructed to work best for idealized noise models, like the symmetric depolarizing noise channel, this is rarely the scenario that real quantum devices encounter. In this work, we realize a deep reinforcement learning agent that automatically discovers QEC codes and their encoding circuits for a given gate set and error model. In particular, this agent is noise-aware, meaning that it learns to switch its strategy depending on some parameter characterizing the noise channel. Our approach is very general and our implementation is extremely efficient, allowing us to (re)discover many codes from scratch within seconds, with physical qubit numbers of up to 20 and code distances varying from 3 to 5. All in all, our work is a powerful and versatile tool that we hope will accelerate the development of QEC across diverse quantum hardware platforms of interest.