Deep Reinforcement Learning for Real-time Feedback Calibration

Effective, efficient calibration is necessary for maximizing the potential of quantum computing (QC) hardware. As platforms advance toward the utility-scale, scalable calibration routines will become increasingly essential. But contemporary approaches to calibration are often very manual, requiring human intervention and extended experimental downtime. In this study, we explore deep reinforcement learning (DRL) as a promising approach for achieving scalable, high-precision calibration with real-time feedback. DRL, situated at the intersection of control theory and machine learning, uses agents that learn to make decisions through trial and error—an approach well-suited for quantum calibration tasks. We present results from two specific tasks: (i) on-line, single-parameter drift calibration and (ii) automated circuit selection, which can be integrated into a heuristic feedback scheme.