AI and the World's Hard Problems · Quantum physics & quantum computing · arXiv · 2026-10-22
Turning Quantum Compiler Configuration Selection Upside Down: A Top-Down Paradigm
Selecting an effective quantum compiler configuration requires joint decisions over hardware targets and compiler settings, but existing state-of-the-art methods choose a device first and then tune compiler options on it with RL-based search, which can cap the achievable fidelity. Taking advantage of the fact that quantum compiler configuration spaces remain small enough for exhaustive search, the authors build QuOPTBench, a benchmark of oracle configurations for 268 quantum circuits, and find that ten representative LLMs remain far from the oracle when asked to infer configurations directly. They then use an LLM to extract transferable configuration-selection rules, and QuTOP, a rule-guided ML framework combining these rules with trained predictors, selects configurations fast without iterative online search while giving the best balance of solution quality and selection latency among the baselines.