Adaptive gateset design for superconducting qubits using reinforcement learning

1. The advantage of a quantum computer over its classical counterpart relies heavily on the ability to perform high-fidelity quantum logic operations. Theoretical and empirical studies of error sources have resulted in many promising designs for the standard single-qubit and two-qubit quantum gates. However, these candidates are limited to the considered errors and remain uninformed of the drifting nature of quantum hardwares and any other error processes. Given the vast design landscape at the pulse-level in a superconducting quantum computer, we employ a deep reinforcement learning agent to design quantum gates adaptively to different physical models. We simulate the superconducting qubits as transmons in an open quantum system. Instead of the commonly used average fidelity, we devise the figure of merit based on the worst-case fidelity, allowing for appropriate assessment of progress towards fault-tolerant quantum computation. We demonstrate the flexibility of the reinforcement learning scheme in learning single-qubit rotation and cross-resonance operations.