OptCRLQAS

OptCRLQAS reduces expensive evaluations from every step to roughly one evaluation per m edits. For 12-qubit H2O, OptCRLQAS reduced average wall-clock time per episode without degrading final quality. OptCRLQAS reduced quantum energy evaluation time and classical optimization time in runtime benchmarks. OptCRLQAS accumulates multiple architecture edits before running full variational optimization and energy evaluation. OptCRLQAS can improve learning signals by evaluating local gate combinations rather than isolated single-gate edits. OptCRLQAS is proposed as a way to make curriculum RL-based QAS more feasible at larger qubit counts.