# Quantum Machine Learning: Challenges and Applications

An interview with Felix Gemeinhardt, Quantum and Deep Tech Enthusiast conducted by Mira Dechant

An interview with Felix Gemeinhardt, Quantum and Deep Tech Enthusiast conducted by Mira Dechant

**Felix, thank you very much for your time. Today we are diving into the topic of quantum machine learning together. You are an expert in the field and successfully defended your doctoral thesis in spring 2024. During your time at the JKU in Linz, you also worked with QMware and investigated a use case in the field of optimization in the manufacturing industry. We’ll come to that later. Let’s start with the basics: Quantum technology is just making the leap from the laboratory into the economy and the first industrial applications. Many questions remain unanswered. What motivated you to start your career in the mysterious world of quanta?**

Well, I didn’t start out directly in the world of quantum. I first studied physics and management, so I studied two subjects at the same time.

I was looking for a master’s thesis topic for my management studies that I could combine with my physics studies. So I thought to myself, well, I’ll write about the strategic importance of quantum computing because it was one of the technologies that you heard about relatively often back then. That was 2020.

In the course of this, I also conducted interviews with experts in the field of quantum computing. Among others, I had contact with Georg Gesek and also with my later dissertation supervisor. He then suggested that I start a dissertation with him in this field so that we could continue working on the topic together. The timing was just right. At that time, in 2020, it was really possible to start a dissertation in applied research on quantum computing. In other words, it was actually a chance hit and thus my entry into the quantum world.

**You said that you were interested in quantum computing from a strategic and economic perspective. What is the fascination here – especially beyond the hype?**

The fascination is that you simply use fundamentally different physical effects to process information. In other words, we are not making a processor that already exists better, or a different processor architecture with a transistor that already exists, but we are really using quantum mechanical effects.

We look at the theory of quantum mechanics, which regulates how the world behaves at the smallest level. Some people don’t understand this because they are really very counter-intuitive concepts, phenomena, but you can use these phenomena. This allows information to be processed in certain cases.

And this results in certain use cases that enable a real-world benefit. It may be that drugs are developed faster, chemical substances are developed faster, and so on and so forth. And that simply fascinated me at the time, this combination of, on the one hand, the fundamentally different physical phenomena that are used and, on the other, the really noticeable applications in business. The time horizon was still questionable at the time, but that’s what makes it exciting at the beginning.

**Quantum mechanics is sometimes counter-intuitive, a bit complex, difficult to explain, you said.**** ****When you talk to your grandparents and they say, Felix, what are you actually doing? How do you explain this area: quantum physics or quantum machine learning?**

Quantum mechanics are simply the rules of how atoms and molecules behave, which we simply cannot feel in our world. But there are simply rules that apply in the world. You simply have to accept that. It’s usually not that interesting for my grandparents to find out exactly what the rules are, because it usually gets too complicated. But they don’t ask any more questions. They simply accept that there are other rules that govern this world. And if you then say that you can use the effects, for example for information processing, i.e. that you can simply make a better computer, but with a completely different structure, then they usually understand that.

**The next step would be to explain not only quantum computing, but also quantum machine learning. And there are also differences. Perhaps you could go into this in more detail in advance: What is the difference to Quantum Machine Learning and when do we talk about Quantum AI?**

If you stay purely in the classical sense, there is a distinction between machine learning and AI in general, and in principle it is simply a matter of not telling a machine exactly what to do, but rather feeding the machine with data and the machine should then know on its own how to solve a certain problem.

For example, you can show the machine a bunch of data and the machine should then learn what a horse looks like. If the machine then sees a new image of a horse, it should know: Okay, this is a horse – based on what it has learned from the other data. And if I use quantum computers in such machine learning or AI applications, then I would speak of quantum machine learning and that brings certain advantages:

For example, one possible advantage would be the speed of the algorithm. Secondly, there is the speed of learning, in the sense of: I need less training data. If we stay with the horse example, this means that I may need fewer pictures of horses to learn what a horse looks like if I use a quantum computer.

In today’s world, data is a very important resource and its availability is sometimes limited. Quantum machine learning is therefore a very promising application of quantum computing, a very promising use case. Quantum machine learning can achieve good learning effects with limited data.

**Let’s move on to Quantum AI: what does the term stand for?**

The combination of quantum computing with AI. The same distinction remains as between classic machine learning and classic AI. If you add quantum, it’s quantum machine learning or quantum AI. So one is simply a subset of the other.

**And where do you really see a lot of potential in the field of quantum machine learning?**

Although I actually did my work in the field of optimization, together with a colleague from Johannes Keppler University, I actually see the greatest potential of quantum machine learning in learning about quantum systems in general. For example, I have a certain material or molecule that I want to use as a drug. And I want to learn something about this physical system, the molecule, the material. The molecule, the material is a quantum system.

This means, of course, that with a quantum computer, which is also a quantum system, I can learn something about this quantum system more efficiently than if I were to build a bridge via the classical world. And there are already initial approaches where it has been shown that you theoretically need exponentially less training data with a quantum computer than with any classical machine learning method.

**You mentioned drug development earlier.**** ****But I also keep hearing that we still have a long way to go in the area of simulation.**

Yes, I think that’s because the simulation use case is not so clearly defined. So, if I simply want to research certain properties of the quantum system for a specific case, without really having to simulate everything now, then that would be possible in the near future. But if I try to fully simulate a very complex system, it becomes difficult. The molecule, for example, is a quantum mechanical system, and it does not move in isolation, i.e. in a vacuum. There are lots of other molecules all around, it’s all a messy mixture. It is difficult to fully simulate this huge, complex structure. And that will certainly take even longer with a quantum computer. But as I said, if you want to make statements about general properties of a physical system, in a specific case, I would also see the potential of quantum computing in the near future.

**Let’s move on to your study project with QMware. Please explain how you came to this and what the specific problem was.**

The problem that the algorithm is supposed to solve is a problem that can be found in medium-term production planning. The question here is how much should I produce at what time so that I can satisfy a certain demand in the future.

So I have certain inputs in the company – what I buy, my inputs. And I know that I want to create certain end products. And the end products have certain demands in certain periods – let’s say day 1, day 2, day 3. But I don’t know them. I know with what probability customers will buy something on Saturday and I know with what probability customers will buy something on Monday. Nevertheless, future demand is uncertain.

To get from the input products to these different end products, I can create different components in the middle. And I can then put these components together in different ways to create my end products.

The question now is: How much do I need to generate in each period? How much of component X or end product Y and so on do I have to produce on day 1, how much of which intermediate product or end product do I have to produce in order to meet demand?

Since I don’t know the demand, what do I do? In classical optimization, it is now possible to carry out so-called stochastic optimizations. In other words, you don’t say: Demand is sure to show ten products on Monday – but I know certain probabilities. I can say that with a probability of 10% the customer will buy five products, with a probability of 20% he will buy ten products and so on. In other words, we are talking about different scenarios.

So you don’t optimize for the case that the customer buys exactly ten products, but you optimize for the mean value across all scenarios. This means that the probable quantity that he will buy is optimized. And because you have a lot of scenarios and perhaps also a lot of intermediate products and some end products, you can imagine that this will become relatively complicated to solve relatively quickly.

A computer simply takes far too long to find the optimum solution, it doesn’t work. So so-called heuristics are used. A heuristic is an algorithm that may not calculate the optimal solution, but a useful solution. And in less time. Some of these heuristics now break down this big problem into subproblems. Subproblem 1 is then solved and a result is obtained which serves as input for subproblem 2. Solving sub-problem 2 again produces a result, which in turn serves as input for sub-problem 1. In other words, you approach the problem iteratively by solving the two sub-problems.

And what I have now done is to solve the severity of these two sub-problems with a quantum computer and then compare them. This means that one part has always been solved classically, the easy part, and the hard part, once classically and once with quantum computing. And we then compared that.

**And what did the comparison reveal?**

To be able to solve this on the quantum computer at all, we had to change this heuristic on the one hand and also change the formulation of the problem. In other words, even the classic part that we used didn’t exist before.

We have a new classical part that is now quantum-inspired. And we compared pure quantum computing with each other. The realization: With today’s quantum computers, we are not yet better than the quantum-inspired classical part. But we have found a quantum-inspired classical part that, in certain situations, is better than anything that has gone before in the classical world. And this is a learning from quantum research in general: you shouldn’t just look at where quantum computing is immediately much better than everything classical. Instead, they ask what is the point of quantum research or working in the quantum field in general? And one of the benefits of this is that certain barriers in traditional algorithm development are being broken down. In other words, you learn an insane amount from research in the field of quantum computing. That’s super exciting.

**May I ask again about the quantum-inspired algorithm: Does that mean that the computing is still absolutely classical, i.e. the hardware is still classical, but quantum elements or the mechanics of quantum computing are integrated into the software design?**

Exactly, so strictly speaking, quantum-inspired computing is something where you either imitate certain effects from quantum computing or where you say, okay, I’m just going to be inspired by quantum mechanics and say that certain solutions or a certain procedure uses effects from quantum mechanics. And that brings me to a better solution. Super exciting.

**By using the QMware platform, you have pursued a hybrid approach – i.e. using both classic and quantum resources. Can you explain what the advantage of hybrid quantum computing is?**

Because QMware offers such a complete package, it was quite practical to make this switch between the classic algorithm and the quantum algorithm. In fact, we have never used the actual quantum computer, only the quantum simulator. You have to differentiate again. This has nothing to do with quantum-inspired computing – the quantum simulator really is a one-to-one replica of a quantum computer. Also because quantum computing is currently not so big that it can no longer be simulated classically, it makes sense for research purposes to simply simulate it one-to-one with a classical computer. But this classical simulator really does behave exactly like the quantum computer, with all the errors it has.

**Finally, let’s look ahead once again: what’s next for you now that you’ve successfully joined the program? Are you staying true to quantum machine learning?**

The next step is now to move from academia to industry. And I want to stay connected with quantum computing, quantum machine learning, simply because I see the potential and because it’s an extremely exciting topic. Since the time horizon is very uncertain, and there may well be something like a quantum winter, I am now taking the step where I am still solving quantitative problems, but the central topic, the focus is no longer quantum machine learning. It plays a role, but it no longer plays the central role.

**You said it before, you’re going into management consulting. To what extent is quantum machine learning relevant for companies today?**

It is a question for companies, especially for company managers who want to know: Do I really have to jump on the quantum hype now? What am I missing? And that is the first question that is important to clarify. What can the company expect in concrete terms? Of course, a management consultancy with expertise in this area can help a lot. Once concrete benefits have been worked out, you can of course take the next step and build a prototype.

It will be too late to get started when quantum computing is already scalable. Similar to machine learning, you have to know in advance what you need it for. Build up the structures first, get the employees on board, create the competence and awareness that you can then simply implement the whole thing more quickly in the company. And that’s what management consultancies are good for.

**In which industries do you expect an increased demand for consulting on quantum machine learning applications?**

So, in my opinion, mainly in industries that have really important, quantitative problems and that have already exhausted the state of the art in classical computing in order to remain competitive. If a company is just starting to work with classic algorithms, then there is still so much room for improvement in terms of classic computing that quantum machine learning is not needed at first. Quantum machine learning is relevant for industries that really need to be state of the art in order to remain competitive, such as the financial industry. Or, as mentioned, the pharmaceutical industry or material production, which can save a lot of resources with quantum machine learning if they can make certain simulations better or improve certain properties of their materials. Everything I can find out on the computer, I no longer have to do in the lab. Everything I do in the lab takes a lot of time and a lot of resources. In this respect, there are examples that could benefit greatly from quantum computing.

**If you could give young school leavers with an interest in the world of quanta a recommendation for their education today: What advice would you give them?**

First of all, to get a rough idea of how diverse this world of quanta actually is and where you actually see yourself. Do you see yourself more in basic research, where, like Anton Zeilinger, you create basic experiments or basic theories about the whole thing? Or do you see yourself more in the application and want to work on the quantum computer of tomorrow, on the quantum communication technologies of tomorrow? You should look at the whole spectrum: Quantum computing, quantum communication, quantum sensing. And don’t get caught up in the hype and think, okay, I have to learn everything now so that I’m always on top of the bandwagon, because in two years it’ll be history again. It won’t happen that quickly.

At the same time, you should keep in mind that you are still aiming for a broad education. Quantum technologies are still a very young field, which means that it is very dynamic and you simply need generalists who are broadly positioned in order to be able to familiarize themselves very quickly with the specific topics. Depending on which field you want to work in later, it also makes sense to take additional subjects. From electrical engineering to computer science and so on. These are the three things I would like to give young graduates: get a rough picture, take a step-by-step approach and still aim for a broad education.

**Thank you very much for the interview.**