Quantum Compilation

In 2-qubit ZZ(pi) approximation, HER did not reach the target fidelity threshold within 100,000 episodes. Compilation agents were trained for 50,000 episodes and evaluated on 100,000 independently sampled Haar-random target unitaries. Quantum compilation was modeled as off-policy deep Q-learning in the experiments. Compilation represented the residual unitary using real and imaginary entries. The compilation experiments included 1-qubit small-rotation tasks, an HRC-basis task, and a 2-qubit setting with single- and two-qubit rotations.