Reinforcement Learning assisted Pulse Shaping for Superconducting Qubit Readout

The readout performance of resource-efficient quantum processors comprising multiple superconducting qubits is often not on par with that of qubit-gate operations. Nonidealities such as crosstalk limit the readout performance. Some of these nonidealities can be mitigated or compensated by qubit-state discrimination or qubit-readout pulse shaping. Quantum error correction protocols depend on fast and efficient readout. Quick resonator ring-up and ring-down in a dispersive readout scheme ensure fast measurements and limited qubit dephasing in future operations. This talk focuses on readout pulse shaping for multiple superconducting qubits using deep reinforcement learning. Relative to conventional readout methods, our results reveal that deep reinforcement learning can significantly reduce measurement times.