Design of QML benchmarking metric for NISQ devices

The research fields of machine learning and quantum computing have seen rapid progress within the last two decades and are expected to have a disruptive impact on science, technology, and our society overall. Machine learning, having benefited from methodological progress and increased availability of inexpensive computing power, is a set of methods sought to enable computers to perform tasks where the human mind holds (or previously held) a natural superiority, such as image recognition or generative tasks. Machine learning methods are in the process of become ubiquitous in our everyday life and are a cornerstone of future economic and scientific development. The emerging field of quantum machine learning (QML) explores how machine learning and quantum technology can be combined. While many algorithms and methods from machine learning have already been extended to the quantum domain, it still is rather unpredictable how much added value and impact QML will actually have, as it strongly depends on the progress of quantum computer developments. 

To inform and steer the development of quantum computer platforms better, key performance comparisons become a necessity. Development of such benchmarks is notoriously difficult. It prerequisites a detailed understanding of all concepts involved ranging from quantum science to machine learning and potential use cases, since the performance metrics are highly interrelated and context-dependent. 

The central goal of our presented work is to provide a simplified, condensed, and unified method for assessing the performance of quantum computers in the context of QML applications. It should serve decisionmakers from industry, politics, and other scientific domains and to allow them to evaluate and understand the progress of the field, without delving into advanced theoretical concepts or understanding details of a particular hardware platform.