Better than classical? The subtle art of benchmarking quantum machine learning models

Benchmarking models via classical simulations is one of the main ways to\njudge ideas in quantum machine learning before noise-free hardware is\navailable. However, the huge impact of the experimental design on the results,\nthe small scales within reach today, as well as narratives influenced by the\ncommercialisation of quantum technologies make it difficult to gain robust\ninsights. To facilitate better decision-making we develop an open-source\npackage based on the PennyLane software framework and use it to conduct a\nlarge-scale study that systematically tests 12 popular quantum machine learning\nmodels on 6 binary classification tasks used to create 160 individual datasets.\nWe find that overall, out-of-the-box classical machine learning models\noutperform the quantum classifiers. Moreover, removing entanglement from a\nquantum model often results in as good or better performance, suggesting that\n"quantumness" may not be the crucial ingredient for the small learning tasks\nconsidered here. Our benchmarks also unlock investigations beyond simplistic\nleaderboard comparisons, and we identify five important questions for quantum\nmodel design that follow from our results.\n