Reinforcement Learning for Quantum Gate Control

How can model-free reinforcement learning improve two-qubit gate control under modeled noise?

Approach

Combine model-free reinforcement learning with Walsh-function-based calibration and projection-based observations. Evaluate two-qubit gate control under modeled non-Markovian detuning noise.

My contribution

• Developed the pulse-level reinforcement learning workflow.
• Integrated calibration and projection-based observations.
• Used Slurm to run parameter sweeps and aggregate simulation results.

Results

Simulations achieve mean average gate fidelity above 0.98, with a tenfold reduction in gate error relative to an analytical baseline.

Ongoing research. A presentation at the APS Global Physics Summit 2027 is planned.