RQC Seminar

285th RQC Seminar

• Speaker

  Mr. Manuel Guatto
  ( Forschungszentrum Jülich )

• Date

  14:00-14:50, (2:00 p.m.- 2:50 p.m.) Tuesday, June 16, 2026

• Venue

  Hybrid( Zoom,
  345-347 Seminar Room, 3F, Main Research Building, Wako Campus / 和光地区 研究本館3階 セミナー室 (345-347) (C01))

• Title

  Real-time noise adaptive quantum error correction by model-free, multi-agent learning

• Inquiries

  norilab_rqc_assist[at]ml.riken.jp

Abstract
Quantum error correction is fundamental for scalable quantum computing, yet existing approaches rely on static assumptions about noise that break down in realistic hardware where error channels drift over time. We introduce a unified framework in which QEC is both learned from first principles, without imposing a predefined code family or stabilizer ansatz, and continuously adapted in real time for a chosen family of noise processes. Our central idea is to separate the problem into offline code discovery and online lightweight recalibration. We realize this by combining multi-agent reinforcement learning (MARL), which autonomously discovers a complete QEC cycle—encoding, syndrome extraction, and recovery—as an explicit circuit realization of the code, with BRAVE, a real-time agent that augments the discovered code with a low-dimensional adaptive variational layer that can be re-optimized online without retraining the full MARL stack. This yields a “discover once, adapt continuously” strategy: MARL provides the base architecture, while BRAVE tracks drifts in the effective error basis under the non-stationary noise present in nowadays quantum processors. As a result, in the slowly varying-noise regime and at sufficiently high sampling rates, our method reduces logical infidelity by a factor approximately of 18x for qubit codes and 3x for qutrit codes, while substantially extending robustness to noise fluctuations. These results establish a paradigm in which QEC is no longer static but dynamically optimized for realistic quantum hardware.