A Summary of a Quantum Machine Learning (QML) State of the Art and Future Directions

A Summary Of A Quantum Machine Learning (Qml) State Of The Art And Future Directions

Quantum Machine Learning: State of the Art and Future Directions” is a 122-page report by the United States Federal Office for Information Security. It reviews “practically feasible” machine learning methods that not only include quantum algorithms for classical data and quantum algorithms for quantum data, but also classical algorithms for quantum data and “quantum inspired” algorithms for classical data.

Next to potential benefits due to quantum speedup, quantum machine learning may also entail risks regarding reliability, trustworthiness, safety, and security.

What’s in it for me?

The report includes a broad overview of quantum computing, including gate-based “universal” quantum computers and other technologies, as well as classical machine learning, quantum machine learning, and classical machine learning applied to quantum computing. It discusses classical pre-processing, quantum states, quantum circuits, errors, noise, and classical post-processing. It delves only superficially into hardware, oriented toward superconducting quantum computers without explicitly stating so, and it concludes with a glossary.

Who is the target audience?

The report begins with a non-technical introduction to artificial intelligence, machine learning, and quantum machine learning. The next section, however, is a mathematical introduction to quantum computing. A later section is a research summary, merely citing a number of papers that have been published on the topic. Therefore, there is no one target audience for this report. A layperson might not understand the math-heavy sections, a physicist or mathematician might be disappointed by the plain English sections, and the other sections direct the reader to other resources.

What does that mean?

The report includes non-standardized terms such as “quantum inspired,” “quantum enhanced,” “quantum boosting,” and others. Even “quantum machine learning” has more than one definition, but the report selects the one that would be better characterized as variational (aka, hybrid classical-quantum) machine learning. There’s possibly only one reference to controversial terminology in the entire report, and that concerns whether parameterized quantum circuits are “quantum neural networks” or not. However, “quantum computing” has more than one definition, “qubits” has more than one spelling, and the report, unfortunately, does not identify this problem.


There is probably something in this report for everyone, even though there probably isn’t everything in this report for anyone. It spends way too much virtual ink on quantum annealing, especially since it acknowledge the performance, cost, and energy efficiency advantages of digital annealing. It’s written as if it’s copy-and-pasted from its 290 cited works, with none of its editors being able to translate the content into a universal format, preferably a plain English explanation followed by the mathematics and then links to additional resources. It’s one redeeming quality is that it is a very broad overview of quantum machine learning, so that someone interested in pursuing related research at least has several points from which to commence such a journey.