Black Box Quantum Error Correction with Reinforcement Learning

In this work, we introduce a general quantum error correction (QEC) framework based on Multi-Agent Reinforcement Learning (MARL) that operates independently of traditional QEC theory. Rather than relying on pre-existing error correction codes, our approach optimizes all the key components of a QEC circuit: encoder, syndrome measurement, and recovery circuits purely through circuit representations. RL agents iteratively refine these circuits by minimizing a cost function based on Knill-Laflamme conditions, ensuring codeword orthogonality and error subspace separability.