Quantum machine learning at Fraunhofer IIS: An interview with Dr. Daniel Scherer

Fraunhofer IIS works on technologies of the future, and quantum computing promises to enable a massive leap forward. Fraunhofer IIS research, particularly in the field of quantum machine learning (QML), is yielding results. Read this interview to discover what QML is really all about, how far QML research has advanced and to what extent the results can already be applied.

Dr. Scherer, what is quantum machine learning about?

Dr. Daniel Scherer: Essentially, what we mean by quantum machine learning (QML) is utilizing quantum resources for the purposes of machine learning. Machine learning (ML) is the term given to data-driven algorithmic methods used to generate solutions, for example to problems concerned with pattern recognition, classification or decision-making. Quantum resources can be quantum computers designed to process data or other information, but they can also be communication channels that exploit the properties of quantum states to transmit information.

Theoretical results from the field of quantum computing indicate that there are special classes of problem that a quantum computer – harnessing quantum mechanical effects such as superposition, entanglement and interference to process information – can solve much more efficiently than is possible within a conventional computer architecture. So in broad terms, this relatively new and dynamic research field of QML is trying to answer the question of if and how quantum computers can make ML methods more efficient and more accurate.

Why is Fraunhofer IIS working with quantum computers?

Dr. Daniel Scherer: Based on the latest research, quantum computing is a computational paradigm that utilizes quantum effects, and this paradigm is superior to the conventional computational paradigm that governs our electronic computers. Although there are still many hardware development hurdles to clear before the ideal – which is to say universal, error-tolerant and error-correcting – quantum computer becomes available, we have to start now to tap potential application areas and become familiar with algorithms and principles required to, say, program quantum computers. The quantum hardware currently available usually falls under the category of noisy intermediate-scale quantum, or NISQ for short. NISQ hardware has only limited data-processing capacity and is prone to errors that can disrupt the actual computational process or make it harder to read the results. But some people are speculating that there are areas of application for which NISQ computers could have the edge over conventional computers and algorithms. In our work on QML, we take both a medium- and long-term approach. Aligning our research accordingly, our goal is to pave the way for transferring technology from basic research into application so that we can develop and offer solutions based on quantum computing.  

What sets Fraunhofer IIS concepts and solutions apart from the competition?

Dr. Daniel Scherer: Our QML research activities focus on quantum reinforcement learning (QRL) – in other words, reinforcement learning (RL) that utilizes quantum computers. In RL, a learning agent – created using suitable algorithmic components – interacts with an environment that presents a particular problem to solve. By performing certain actions, the agent receives a corresponding reward signal from the environment. The goal is for the agent to find the optimum solution strategy for the problem more or less autonomously and with as little interaction as possible. At the moment, we’re focusing on researching the use of NISQ computers for this purpose, but we’re also exploring what types of quantum algorithms might be possible in the future with new hardware. Through our collaborations with industry partners as part of joint projects, we can take industrial application requirements into account at an early stage when developing our algorithms and software. Equally important here is to develop benchmarking pipelines. These make it possible for us to draw useful comparisons between different RL algorithms and compare RL and QRL algorithms to identify any potential quantum advantages.

How far has the research advanced?

Dr. Daniel Scherer: The dichotomy mentioned earlier in relation to hardware – NISQ computers versus universal, error-tolerant and error-correcting quantum computers – can also be found at the algorithm level. While algorithms designed for the virtually ideal hardware of the future have been thoroughly researched and promise verifiable quantum advantages, far less research has been done into NISQ algorithms, and from a theoretical  perspective, there is often zero or at best a very weak indication of potential quantum advantages. The past few years have yielded a better understanding – especially in the field of QML – of the properties of what are known as variational quantum algorithms, which essentially are tailored to NISQ computers and are also the focus of our research. But it’s not yet possible to make any definite statements regarding the quantum advantages they might hold. We keep a very close eye on how academic research is developing, and we’re in regular contact with our university partners. Key aspects of our own research include the coding of classical data – data indispensable for ML – within quantum algorithms, and the reading of information. This is where we see one of the best opportunities to boost the performance of QML algorithms. Going forward, we want to introduce automation and scalability to the design of coding and decoding strategies that are based on quantum information theory, as well as to other aspects of quantum algorithm design. There are also challenges to overcome when it comes to software engineering. The reproducibility of results, access models on quantum hardware, optimized job scheduling, optimized utilization of quantum resources, and many other factors play a decisive role in developing our methods.

To what extent is it possible to apply the results of this research?

Dr. Daniel Scherer: As it’s still not possible to make any definitive statements regarding the quantum advantages of QML based on variational quantum circuits, the question of the “QML killer app” sadly remains unanswered. We’re directing our research efforts toward a variety of application problems, ranging from managing industrial processes and data-driven learning to the optimized utilization of antenna arrays for mobile communications applications. Only once we have systematically scaled the problem’s parameters and analyzed the performance of our quantum algorithms will we uncover any potential quantum advantages.

The only strong indications that NISQ hardware and variational algorithms do indeed offer quantum advantages are for the class of native quantum applications – for example simulations of quantum systems such as molecules in pharmaceutical research. Here, the idea is to solve an intrinsically quantum mechanical problem with the help of a quantum computer. As we also research methods of using classical ML and RL processes to tackle complex computational problems encountered when working with quantum computers, we’re also addressing the question of the extent to which these problems can be expressed as native quantum applications and solved using QML methods. So we’re looking to use QML to help develop more powerful quantum computers.

Finally, let’s cast a brief glance to the future: How close are we to having the ideal quantum computer?

Dr. Daniel Scherer: I think it’s still difficult to estimate how close we are. The conventional computing paradigm that governs our computers and supercomputers is still adequate for problems up to a certain size. Only once we want to move beyond these problem parameters will using quantum computers really pay off. But when it comes to quantum simulations of things like molecules, I think we’re actually not all that far away from having error-correcting quantum computing. In that situation, around 100 logical (error-corrected) qubits ought to be sufficient to slowly make advances in the area of quantum advantages.

Dr. Scherer, thank you very much for speaking with us.