BayQS — Bavarian Competence Center for Quantum Security and Data Science

Since the development of Shor’s algorithm in 1994, asymmetric cryptographic methods have become much less future-proof: In particular, cryptocomponents that are permanently installed in industrial environments have life cycles that can span decades and are therefore susceptible to QC-based cryptanalysis. The manufacturers and users of these components have a need right now for new, QC-resistant methods or update mechanisms to enable the methods to be switched in the future. Furthermore, there is a real need for evaluation services with regard to the post-quantum resilience of a company’s security infrastructure and for consultation when it comes to creating and implementing suitable concepts.

At the same time, another priority of BayQS is to investigate symmetric methods using QC-based cryptanalysis in order to assess their future-proofness and security credentials.

Quantum machine learning is a young, interdisciplinary field of research which aims to use quantum computers and algorithms to accelerate and improve classic machine learning and deep learning approaches and to make them more economical in terms of data. The spectrum ranges from learning variational circuits to hybrid approaches with special quantum layers within neural networks, right through to pure quantum algorithms.

In addition to studying the practicability and performance of these approaches at BayQS, we are also investigating their reliability and security (for example, their susceptibility to adversarial examples) and how these aspects can be improved.

Furthermore, quantum computers can be used to certify classic machine learning models.

One of the key priorities for BayQS is to make sure that quantum computing platforms in cloud environments are secure in order to protect know-how and intellectual property and to ensure users’ data sovereignty, while also ensuring legal certainty and operational security for cloud service providers. Procuring and operating a quantum computer requires investments that run into the millions, which means that the majority of companies — particularly SMEs — will be reliant on cloud quantum computing services. With these services, there is a risk that know-how and data may be lost due to non-secure QC platforms.

Following the introduction of the European General Data Protection Regulation, service providers are under increased pressure to implement technical security measures in the context of data processing. The principle of shared responsibility applies here, which means that, while cloud operators are responsible for offering technical security features — e.g., encrypted data storage — the users of these services are likewise obligated to use these features and configure them correctly.

BayQS is working on identifying and developing these technical features for both operators and users.

The use of modern sensors, especially for 2D and 3D imaging, is extremely diverse. This applies in particular to industrial measuring technology in all areas of high-tech production — ranging from series manufacturing to a batch size of one.

The task-oriented use of complex sensors involves two challenges which, in some cases, give rise to major combinatorial optimization problems as well as huge volumes of data which need to be processed.

When it comes to the two challenges — processing image signals and solving combinatorial optimization problems — there are established approaches for quantum computers; for example, the quantum Fourier transform and the possibilities of quantum annealing.

As part of the BayQS project, we are working to adapt and further develop these approaches so they can be used in the context of modern sensors for the first time. Our aim here is to facilitate QC-driven image and signal processing. To do this, we are using the example of X-ray computed tomography — a 3D imaging technology. For this technology to be implemented successfully, we need to solve the challenges mentioned above time and time again.

Many issues and challenges in the field of logistics can be formulated as mathematical problems and then solved using algorithmic methods; for example, planning and route problems.

Since the discovery of Shor’s factoring algorithm in 1994, followed shortly after by Grover’s search algorithm, a whole host of other quantum algorithms have been developed which are able to solve all kinds of problems of this nature more quickly than the established, conventional methods can. The two key elements that have led to this acceleration are the quantum Fourier transform and the process of amplitude amplification. A fundamental problem that all of these quantum algorithms have in common is the demand placed on the hardware: For practical problem sizes, this demand is so high that it cannot be fulfilled by current quantum computers, nor by those planned for the future.

One possible way of getting around this problem involves the use of the hybrid variational quantum eigensolver, which combines classic algorithmic methods of solving optimization problems with quantum algorithms in order to make the best possible use of even the smallest of quantum resources. In this context, the classic optimization algorithm uses a parameterizable quantum algorithm such as quantum annealing or the quantum approximate optimization algorithm in order to generate possible solutions and then attempts to improve their quality by selecting appropriate parameters.

One of the aims of the BayQS project is to identify possible benefits arising from the use of quantum computers in the field of logistics — both for quantum computers that already exist or are being planned, and at a conceptual level for the future.

Mobile communications technology has evolved into one of the most influential technologies of our time. For more than two decades, Fraunhofer institutes have played a key role in shaping this development. As it has expanded to include higher and higher frequency ranges and more complex systems, new challenges have emerged which can only be solved with fundamentally new technology.

When it comes to identifying locations in mobile communications networks, 5G provides a new framework with localization methods which measure the transmission time and angle of radio signals. Particularly in scenarios with shadowing and multipath scattering, the boundary conditions for localization are difficult. A promising way of obtaining good measurement data in this context is to use radio channel estimation with the aid of multiple-antenna systems known as multiple-input multiple-output (MIMO) antennas. At higher frequency ranges and in distributed systems, the measuring, configuration and signaling tasks all become very complex. Due to the size of the system and the many dynamic parameters, the processes are very computationally intensive and highly time-critical.

The underlying non-convex optimization problems can be solved more efficiently using artificial intelligence methods rather than conventional algorithms such as exhaustive search. As tasks such as beam coordination and localization in radio networks can only be carried out interactively in real time, methods based on reinforcement learning are particularly suitable in these cases. With these methods, the “learner” interacts with its environment. It can perceive the state of this environment either fully or only partially on the basis of an observation made by executing an action in accordance with an underlying strategy.

We are working on incorporating quantum computers into these reinforcement learning methods in order to make them more efficient. Although there are already some theoretical proposals for how to implement quantum reinforcement learning — based in part on Grover-style amplitude amplification — the expected time savings in terms of transmission time are moderate. In general, quantum methods can provide benefits with regard to various dimensions of algorithmic complexity. Hybrid reinforcement learning methods are being developed and examined for quantum benefits by means of appropriate numerical experiments or by incorporating real quantum hardware.