Machine Learning aiding the Discovery of better Strategies for Quantum Computing

As quantum technologies, including quantum computers, are scaling up, it becomes a challenge to harness the ever-growing complexity. Approaches based on machine learning are perfectly suited for this task. In this talk I will review how one can employ reinforcement learning, a powerful and general approach, to discover from scratch optimized feedback strategies. I will show how we are using model-free reinforcement learning to discover quantum error correction strategies like fault-tolerant logical state preparation. Moreover, we have applied the same technique in real experiments, for the first time training a real-time neural-network agent to feedback-control a superconducting qubit. Finally, I will mention our recently developed feedback-GRAPE technique for doing model-based reinforcement learning, and how this has already led to the discovery of an improved strategy for logical qubit stabilization in the GKP code setting.